rvsttd: increase fade-out to 25ms

This reverts commit 84474731e7.

.gitignore

@@ -27,3 +27,6 @@ Robovoice.Tts.LibPiper/piper.h
 ## Temp files
 *.wav
 *.raw
+
+## Rust build output
+target/

BACKLOG.md

@@ -0,0 +1,21 @@
+# Robovoice Backlog
+
+## Second-order PTT
+
+When a game is running, it has its own PTT button (e.g. a voip push-to-talk
+key). Robovoice should simulate pressing the game's PTT key before TTS audio
+output begins, and release it after playback finishes.
+
+This lets the TTS audio be transmitted through the game's voip channel to
+other players.
+
+### Considerations
+
+- Needs a configurable "game PTT key" (separate from Robovoice's own PTT key)
+- Use `SendInput` or `keybd_event` to synthesize the keypress
+- Press the game PTT key right before buffered audio starts playing
+- Release it after `AudioOutput` finishes playback (need a playback-complete
+  signal — currently `Flush()` doesn't provide one)
+- Edge cases: what if the user presses Robovoice PTT while game PTT is still
+  held from a previous utterance? Flush should release game PTT too.
+- Should this be a per-output-device setting? (CABLE Output vs speakers)

DhcpTunnelTest/Program.cs

@@ -1,106 +0,0 @@
-using System.Net;
-using System.Net.NetworkInformation;
-using System.Net.Sockets;
-using System.Text;
-
-// Args: [interface_ip] [interval_ms]
-// Defaults: auto-detect LAN IP, 50ms
-
-string localIp = args.Length > 0 ? args[0] : GetLanInterfaceIp() ?? "0.0.0.0";
-int intervalMs = args.Length > 1 && int.TryParse(args[1], out int iv) ? iv : 50;
-
-using var sock = new Socket(AddressFamily.InterNetwork, SocketType.Dgram, ProtocolType.Udp);
-sock.SetSocketOption(SocketOptionLevel.Socket, SocketOptionName.ReuseAddress, true);
-sock.EnableBroadcast = true;
-
-var localEp = new IPEndPoint(IPAddress.Parse(localIp), 68);
-sock.Bind(localEp);
-
-Console.WriteLine($"Bound to {localEp} (SO_REUSEADDR)");
-Console.WriteLine("Sending NOPs + listening. Press Ctrl+C to stop.");
-Console.WriteLine();
-
-var cts = new CancellationTokenSource();
-var recvTask = Task.Run(() => ReceiveLoop(sock, cts.Token));
-
-var destEp = new IPEndPoint(IPAddress.Broadcast, 67);
-uint session = 1;
-uint nonce = 0;
-
-while (!cts.Token.IsCancellationRequested)
-{
-    nonce++;
-    string msg = $"HKMSTR {session} {nonce}\n";
-    byte[] payload = Encoding.UTF8.GetBytes(msg);
-
-    try
-    {
-        int sent = sock.SendTo(payload, destEp);
-        Console.WriteLine($"[{DateTime.Now:HH:mm:ss.fff}] SENT s={session} n={nonce} {sent} bytes");
-    }
-    catch (Exception ex)
-    {
-        Console.WriteLine($"[{DateTime.Now:HH:mm:ss.fff}] SEND FAILED: {ex.Message}");
-    }
-
-    try { Thread.Sleep(intervalMs); }
-    catch (OperationCanceledException) { break; }
-}
-
-cts.Cancel();
-await recvTask;
-
-static string? GetLanInterfaceIp()
-{
-    foreach (var nic in NetworkInterface.GetAllNetworkInterfaces())
-    {
-        if (nic.OperationalStatus != OperationalStatus.Up)
-            continue;
-        if (nic.NetworkInterfaceType == NetworkInterfaceType.Loopback)
-            continue;
-        if (nic.Description.Contains("WireGuard", StringComparison.OrdinalIgnoreCase))
-            continue;
-
-        foreach (var addr in nic.GetIPProperties().UnicastAddresses)
-        {
-            if (addr.Address.AddressFamily == AddressFamily.InterNetwork)
-            {
-                var ip = addr.Address.ToString();
-                if (ip.StartsWith("192.168.") || ip.StartsWith("10.") || ip.StartsWith("172."))
-                    return ip;
-            }
-        }
-    }
-    return null;
-}
-
-static void ReceiveLoop(Socket sock, CancellationToken ct)
-{
-    byte[] buffer = new byte[4096];
-    EndPoint fromEp = new IPEndPoint(IPAddress.Any, 0);
-
-    while (!ct.IsCancellationRequested)
-    {
-        int received;
-        try
-        {
-            if (!sock.Poll(500_000, SelectMode.SelectRead))
-                continue;
-            received = sock.ReceiveFrom(buffer, ref fromEp);
-        }
-        catch (Exception ex)
-        {
-            Console.WriteLine($"[{DateTime.Now:HH:mm:ss.fff}] RECV ERROR: {ex.Message}");
-            continue;
-        }
-
-        string text = Encoding.UTF8.GetString(buffer, 0, received).TrimEnd('\n', '\r');
-        if (!text.StartsWith("HKMSTR"))
-        {
-            Console.WriteLine($"[{DateTime.Now:HH:mm:ss.fff}] RECV {received} bytes from {fromEp} (no magic)");
-            continue;
-        }
-
-        Console.WriteLine($"[{DateTime.Now:HH:mm:ss.fff}] RECV {received} bytes from {fromEp}: {text}");
-    }
-}

DhcpTunnelTest/Robovoice.DhcpTunnelTest.csproj

@@ -1,10 +0,0 @@
-<Project Sdk="Microsoft.NET.Sdk">
-
-  <PropertyGroup>
-    <OutputType>Exe</OutputType>
-    <TargetFramework>net10.0</TargetFramework>
-    <ImplicitUsings>enable</ImplicitUsings>
-    <Nullable>enable</Nullable>
-  </PropertyGroup>
-
-</Project>

PROTOCOL.md

@@ -0,0 +1,106 @@
+# Robovoice STT Protocol
+
+## Overview
+
+Robovoice connects to the STT server over TCP (typically through a gatuna
+L2 tunnel). The server captures audio, runs Moonshine STT, and sends
+transcript segments back. The client pre-synthesizes TTS on segments and
+plays audio on final.
+
+```
+[Robovoice client] --TCP--> [STT server 127.0.0.1:6996]
+     │                            │
+     ├── ON <session>\n  ────────►│  (abort old, start new session)
+     ├── OFF <session>\n ────────►│  (stop, final STT pass)
+     │◄── P <session> <text>\n ──┤  (completed segment)
+     │◄── F <session> <text>\n ──┤  (all done; text may be empty)
+```
+
+## Transport
+
+- **Protocol:** TCP (reliable, ordered, connection-oriented)
+- **Server:** `127.0.0.1:6996` (hardcoded loopback)
+- **Framing:** newline-delimited text (`\n`), UTF-8
+- **Auto-reconnect:** client retries every 3s if connection drops
+
+## Wire format
+
+### Client → Server
+
+**ON (PTT pressed):**
+```
+ON <session>\n
+```
+Starts a new STT session. The server aborts any active session and starts
+recording. `<session>` is an incrementing unsigned integer chosen by the
+client. Replies from the server echo this session ID.
+
+**OFF (PTT released):**
+```
+OFF <session>\n
+```
+Stops the session. The server does a final STT pass on remaining audio and
+sends any new segments followed by `F`.
+
+### Server → Client
+
+**Segment (completed VAD segment):**
+```
+P <session> <text>\n
+```
+A completed, VAD-separated utterance segment. The client starts TTS
+synthesis immediately and buffers the audio (does not play yet).
+
+**Final (all done):**
+```
+F <session> <text>\n
+```
+Signals all segments have been sent. `<text>` may be empty (`F <session>\n`).
+Triggers playback of all buffered audio on the client. If text is non-empty,
+the client synthesizes it before playing.
+
+## Session IDs
+
+- Client increments session ID on each PTT press
+- Server echoes the session ID in all replies for that session
+- Client drops any reply with a stale session ID (handles the race where
+  stale segments from an aborted session are still in the TCP buffer)
+- Server aborts old session on receiving `ON` with a new session ID
+
+## Client playback model
+
+1. `P` arrives → start TTS synthesis immediately, buffer audio (don't play)
+2. More `P` arrive → keep synthesizing and buffering
+3. `F` arrives → play all buffered audio immediately
+4. PTT pressed → flush: stop playback, cancel synthesis, clear buffers
+
+The purpose of pre-synthesis is to minimize latency between PTT release
+and audio playback. By the time `F` arrives, audio is already buffered.
+
+## Server state machine
+
+```
+              ┌──────────┐  ON <session>   ┌──────────────┐
+              │  IDLE    │ ──────────────► │  RECORDING   │
+              └──────────┘                 └──────────────┘
+                                                │      │
+                                        OFF     │      │
+                                        recv'd  │      │
+                                                ▼      │
+                                            ┌─────────────┐
+                                            │  PROCESSING │
+                                            └─────────────┘
+                                                 │
+                                          send   │
+                                          P/F    │
+                                                 ▼
+                                           back to IDLE
+```
+
+- **IDLE → RECORDING:** `ON <session>` received, start mic capture
+- **RECORDING:** Moonshine streaming produces completed segments → send `P`
+- **RECORDING → PROCESSING:** `OFF <session>` received
+- **PROCESSING → IDLE:** final STT pass, send remaining `P` + `F`
+
+If `ON` arrives while recording, the current session is aborted (no final
+flush) and a new session starts immediately.

Robovoice.App/AppConfig.cs

@@ -12,7 +12,7 @@ public sealed class AppConfig
     public int LengthScale { get; set; } = 100;
     public int NoiseWScale { get; set; } = 800;
     public bool MinimizeToTray { get; set; } = true;
-    public string InterfaceIp { get; set; } = string.Empty;
+    public string SttEndpoint { get; set; } = "127.0.0.1:6996";
 
     public static string AppDataDir => Path.Combine(
         Environment.GetFolderPath(Environment.SpecialFolder.LocalApplicationData),

Robovoice.App/MainForm.Designer.cs

@@ -9,21 +9,15 @@ partial class MainForm
     private TextBox txtPttKey = null!;
     private Label lblVoice = null!;
     private ComboBox cmbVoice = null!;
-    private Button btnTestVoice = null!;
     private Button btnManageVoices = null!;
     private Label lblOutput = null!;
     private ComboBox cmbOutput = null!;
-    private Label lblFile = null!;
-    private Button btnBrowseFile = null!;
-    private Label lblFileName = null!;
     private Label lblServer = null!;
-    private ComboBox cmbInterface = null!;
-    private Label lblLineStatus = null!;
+    private TextBox txtSttEndpoint = null!;
+    private Button btnLock = null!;
     private RichTextBox txtLog = null!;
     private CheckBox chkMinimizeToTray = null!;
     private Button btnClearLog = null!;
-    private Label lblTextInput = null!;
-    private TextBox txtTextInput = null!;
     private Label lblNoise = null!;
     private TrackBar trkNoise = null!;
     private Label lblNoiseVal = null!;
@@ -49,21 +43,15 @@ partial class MainForm
         txtPttKey = new TextBox();
         lblVoice = new Label();
         cmbVoice = new ComboBox();
-        btnTestVoice = new Button();
         btnManageVoices = new Button();
         lblOutput = new Label();
         cmbOutput = new ComboBox();
-        lblFile = new Label();
-        btnBrowseFile = new Button();
-        lblFileName = new Label();
         lblServer = new Label();
-        cmbInterface = new ComboBox();
-        lblLineStatus = new Label();
+        txtSttEndpoint = new TextBox();
+        btnLock = new Button();
         txtLog = new RichTextBox();
         chkMinimizeToTray = new CheckBox();
         btnClearLog = new Button();
-        lblTextInput = new Label();
-        txtTextInput = new TextBox();
         lblNoise = new Label();
         trkNoise = new TrackBar();
         lblNoiseVal = new Label();
@@ -100,65 +88,38 @@ partial class MainForm
         cmbVoice.Size = new Size(200, 23);
         cmbVoice.DropDownStyle = ComboBoxStyle.DropDownList;
 
-        // btnTestVoice
-        btnTestVoice.Text = "Test";
-        btnTestVoice.Location = new Point(423, 11);
-        btnTestVoice.Size = new Size(45, 25);
-        btnTestVoice.UseVisualStyleBackColor = true;
-
         // btnManageVoices
         btnManageVoices.Text = "Add/Remove...";
-        btnManageVoices.Location = new Point(474, 11);
+        btnManageVoices.Location = new Point(423, 11);
         btnManageVoices.Size = new Size(95, 25);
         btnManageVoices.UseVisualStyleBackColor = true;
 
         // lblOutput
         lblOutput.Text = "Output:";
-        lblOutput.Location = new Point(580, 15);
+        lblOutput.Location = new Point(530, 15);
         lblOutput.Size = new Size(50, 23);
         lblOutput.TextAlign = ContentAlignment.MiddleLeft;
 
         // cmbOutput
-        cmbOutput.Location = new Point(633, 12);
-        cmbOutput.Size = new Size(180, 23);
+        cmbOutput.Location = new Point(583, 12);
+        cmbOutput.Size = new Size(150, 23);
         cmbOutput.DropDownStyle = ComboBoxStyle.DropDownList;
 
-        // lblFile
-        lblFile.Text = "Text File:";
-        lblFile.Location = new Point(12, 48);
-        lblFile.Size = new Size(60, 23);
-        lblFile.TextAlign = ContentAlignment.MiddleLeft;
-
-        // btnBrowseFile
-        btnBrowseFile.Text = "Browse...";
-        btnBrowseFile.Location = new Point(75, 45);
-        btnBrowseFile.Size = new Size(75, 25);
-        btnBrowseFile.UseVisualStyleBackColor = true;
-
-        // lblFileName
-        lblFileName.Text = "(none)";
-        lblFileName.Location = new Point(155, 48);
-        lblFileName.Size = new Size(280, 23);
-        lblFileName.TextAlign = ContentAlignment.MiddleLeft;
-        lblFileName.ForeColor = Color.Gray;
-
         // lblServer
-        lblServer.Text = "Interface:";
-        lblServer.Location = new Point(440, 48);
-        lblServer.Size = new Size(55, 23);
+        lblServer.Text = "STT:";
+        lblServer.Location = new Point(12, 48);
+        lblServer.Size = new Size(35, 23);
         lblServer.TextAlign = ContentAlignment.MiddleLeft;
 
-        // cmbInterface
-        cmbInterface.Location = new Point(498, 45);
-        cmbInterface.Size = new Size(165, 23);
-        cmbInterface.DropDownStyle = ComboBoxStyle.DropDownList;
+        // txtSttEndpoint
+        txtSttEndpoint.Location = new Point(50, 45);
+        txtSttEndpoint.Size = new Size(200, 23);
 
-        // lblLineStatus
-        lblLineStatus.Text = "";
-        lblLineStatus.Location = new Point(645, 48);
-        lblLineStatus.Size = new Size(160, 23);
-        lblLineStatus.TextAlign = ContentAlignment.MiddleRight;
-        lblLineStatus.ForeColor = Color.DarkBlue;
+        // btnLock
+        btnLock.Text = "Lock Model";
+        btnLock.Location = new Point(260, 44);
+        btnLock.Size = new Size(90, 25);
+        btnLock.UseVisualStyleBackColor = true;
 
         // lblNoise
         lblNoise.Text = "Noise:";
@@ -166,7 +127,7 @@ partial class MainForm
         lblNoise.Size = new Size(40, 23);
         lblNoise.TextAlign = ContentAlignment.MiddleLeft;
 
-        // trkNoise (0-1000 → 0.0-1.0, default 667)
+        // trkNoise
         trkNoise.Location = new Point(52, 78);
         trkNoise.Size = new Size(120, 45);
         trkNoise.Minimum = 0;
@@ -187,7 +148,7 @@ partial class MainForm
         lblSpeed.Size = new Size(40, 23);
         lblSpeed.TextAlign = ContentAlignment.MiddleLeft;
 
-        // trkSpeed (50-300 → 0.5-3.0, default 100)
+        // trkSpeed
         trkSpeed.Location = new Point(260, 78);
         trkSpeed.Size = new Size(120, 45);
         trkSpeed.Minimum = 50;
@@ -208,7 +169,7 @@ partial class MainForm
         lblNoiseW.Size = new Size(45, 23);
         lblNoiseW.TextAlign = ContentAlignment.MiddleLeft;
 
-        // trkNoiseW (0-1000 → 0.0-1.0, default 800)
+        // trkNoiseW
         trkNoiseW.Location = new Point(475, 78);
         trkNoiseW.Size = new Size(120, 45);
         trkNoiseW.Minimum = 0;
@@ -224,19 +185,8 @@ partial class MainForm
         lblNoiseWVal.TextAlign = ContentAlignment.MiddleLeft;
 
         // txtLog
-        txtLog.Location = new Point(12, 149);
-        txtLog.Size = new Size(800, 310);
-
-        // lblTextInput
-        lblTextInput.Text = "Text:";
-        lblTextInput.Location = new Point(12, 121);
-        lblTextInput.Size = new Size(35, 23);
-        lblTextInput.TextAlign = ContentAlignment.MiddleLeft;
-
-        // txtTextInput
-        txtTextInput.Location = new Point(50, 118);
-        txtTextInput.Size = new Size(762, 23);
-
+        txtLog.Location = new Point(12, 121);
+        txtLog.Size = new Size(800, 338);
         txtLog.ReadOnly = true;
         txtLog.Font = new Font("Consolas", 9F);
         txtLog.BackColor = Color.FromArgb(30, 30, 30);
@@ -262,16 +212,12 @@ partial class MainForm
         Controls.Add(txtPttKey);
         Controls.Add(lblVoice);
         Controls.Add(cmbVoice);
-        Controls.Add(btnTestVoice);
         Controls.Add(btnManageVoices);
         Controls.Add(lblOutput);
         Controls.Add(cmbOutput);
-        Controls.Add(lblFile);
-        Controls.Add(btnBrowseFile);
-        Controls.Add(lblFileName);
         Controls.Add(lblServer);
-        Controls.Add(cmbInterface);
-        Controls.Add(lblLineStatus);
+        Controls.Add(txtSttEndpoint);
+        Controls.Add(btnLock);
         Controls.Add(lblNoise);
         Controls.Add(trkNoise);
         Controls.Add(lblNoiseVal);
@@ -281,8 +227,6 @@ partial class MainForm
         Controls.Add(lblNoiseW);
         Controls.Add(trkNoiseW);
         Controls.Add(lblNoiseWVal);
-        Controls.Add(lblTextInput);
-        Controls.Add(txtTextInput);
         Controls.Add(txtLog);
         Controls.Add(chkMinimizeToTray);
         Controls.Add(btnClearLog);

Robovoice.App/MainForm.cs

@@ -2,7 +2,7 @@ using NAudio.Wave;
 using Robovoice.App;
 using Robovoice.Core;
 using Robovoice.Core.Voices;
-using Robovoice.Stt.Dhcp;
+using Robovoice.Stt.Tcp;
 using Robovoice.Tts.LibPiper;
 using System.Diagnostics;
 
@@ -16,11 +16,12 @@ internal sealed partial class MainForm : Form
 
     private LibPiperTtsEngine? _tts;
     private AudioOutput? _audioOutput;
-    private DhcpSttSource? _sttSource;
-    private Orchestrator? _orchestrator;
+    private TcpSttSource? _sttSource;
+    private VoicePipeline? _pipeline;
     private PttHotkey? _pttHotkey;
     private NotifyIcon? _trayIcon;
     private bool _trayInit;
+    private bool _locked;
 
     public MainForm()
     {
@@ -35,6 +36,8 @@ internal sealed partial class MainForm : Form
     }
 
     private async void OnLoad(object? sender, EventArgs e)
+    {
+        try
         {
             ActiveControl = txtLog;
             PopulateVoices();
@@ -61,17 +64,14 @@ internal sealed partial class MainForm : Form
             OnSliderScroll(null, EventArgs.Empty);
 
             chkMinimizeToTray.Checked = _config.MinimizeToTray;
-        PopulateInterfaces();
+            txtSttEndpoint.Text = _config.SttEndpoint;
 
-        btnBrowseFile.Click += OnBrowseFile;
-        btnTestVoice.Click += OnTestVoice;
             btnManageVoices.Click += OnManageVoices;
             btnClearLog.Click += (_, _) => txtLog.Clear();
+            btnLock.Click += OnLockToggle;
             txtPttKey.Enter += OnPttKeyFocus;
             txtPttKey.KeyDown += OnPttKeyDown;
-        cmbOutput.SelectedIndexChanged += OnOutputChanged;
-        cmbVoice.SelectedIndexChanged += OnVoiceChanged;
-        cmbInterface.SelectedIndexChanged += OnInterfaceChanged;
+            txtSttEndpoint.Leave += OnSttEndpointChanged;
 
             trkNoise.Scroll += OnSliderScroll;
             trkSpeed.Scroll += OnSliderScroll;
@@ -85,8 +85,11 @@ internal sealed partial class MainForm : Form
             Resize += OnResize;
 
             SetupTray();
-
-        await InitializeEngineAsync();
+        }
+        catch (Exception ex)
+        {
+            Debug.WriteLine($"OnLoad failed: {ex}");
+        }
     }
 
     private Keys _pttKey = Keys.F8;
@@ -209,7 +212,21 @@ internal sealed partial class MainForm : Form
             cmbOutput.SelectedIndex = 0;
     }
 
-    private async Task InitializeEngineAsync()
+    // ─── Lock / Release ────────────────────────────────────────────────────
+
+    private async void OnLockToggle(object? sender, EventArgs e)
+    {
+        if (_locked)
+        {
+            ReleaseModel();
+        }
+        else
+        {
+            await LockModelAsync();
+        }
+    }
+
+    private async Task LockModelAsync()
     {
         if (cmbVoice.SelectedItem is not string voiceName)
         {
@@ -230,42 +247,97 @@ internal sealed partial class MainForm : Form
             return;
         }
 
+        btnLock.Enabled = false;
         Log($"Loading voice: {voiceName}...");
-        _audioOutput?.Dispose();
-        _tts?.DisposeAsync().AsTask().Wait();
 
+        try
+        {
             _tts = new LibPiperTtsEngine(
                 modelPath,
                 _espeakDataPath,
                 noiseScale: trkNoise.Value / 1000.0f,
                 lengthScale: trkSpeed.Value / 100.0f,
                 noiseWScale: trkNoiseW.Value / 1000.0f);
+
+            await Task.Run(() => _tts.InitializeAsync());
+            Log($"TTS ready (sample rate: {_tts.SampleRate} Hz)");
+
             _audioOutput = new AudioOutput();
 
-        if (_sttSource is null)
-        {
-            string ifaceIp = cmbInterface.SelectedItem as string ?? "";
-            _sttSource = new DhcpSttSource { InterfaceIp = ifaceIp, Log = Log };
+            _sttSource = new TcpSttSource { Endpoint = txtSttEndpoint.Text, Log = Log };
+            _sttSource.TranscriptReceived += OnTranscript;
             await _sttSource.StartAsync();
-        }
 
-        _orchestrator?.DisposeAsync().AsTask().Wait();
-        _orchestrator = new Orchestrator(_tts, _audioOutput, _sttSource, Log)
+            _pipeline = new VoicePipeline(_tts, _audioOutput, Log)
             {
                 OutputDeviceName = cmbOutput.SelectedItem as string ?? string.Empty,
             };
+            _pipeline.Start();
+
+            _locked = true;
+            SetControlsLocked(true);
+            btnLock.Text = "Release Model";
+            btnLock.Enabled = true;
 
-        try
-        {
-            await _orchestrator.InitializeTtsAsync();
-            Log("Engine ready. Press PTT to send to STT server.");
             SetupHotkey();
+            Log("Model locked. Press PTT to talk.");
         }
         catch (Exception ex)
         {
-            Log($"Init failed: {ex.Message}");
+            Log($"Lock failed: {ex.Message}");
+            btnLock.Enabled = true;
+            _tts?.DisposeAsync().AsTask().Wait();
+            _tts = null;
+            _audioOutput?.Dispose();
+            _audioOutput = null;
+            if (_sttSource is not null)
+            {
+                _sttSource.TranscriptReceived -= OnTranscript;
+                _sttSource.DisposeAsync().AsTask().Wait();
+                _sttSource = null;
             }
         }
+    }
+
+    private void ReleaseModel()
+    {
+        if (!_locked) return;
+
+        _pttHotkey?.Dispose();
+        _pttHotkey = null;
+
+        if (_sttSource is not null)
+        {
+            _sttSource.TranscriptReceived -= OnTranscript;
+            _sttSource.DisposeAsync().AsTask().Wait();
+            _sttSource = null;
+        }
+
+        _pipeline?.Dispose();
+        _pipeline = null;
+
+        // VoicePipeline.Dispose calls _tts.DisposeAsync and _audioOutput.Dispose
+        _tts = null;
+        _audioOutput = null;
+
+        _locked = false;
+        SetControlsLocked(false);
+        btnLock.Text = "Lock Model";
+        Log("Model released.");
+    }
+
+    private void SetControlsLocked(bool locked)
+    {
+        cmbVoice.Enabled = !locked;
+        cmbOutput.Enabled = !locked;
+        trkNoise.Enabled = !locked;
+        trkSpeed.Enabled = !locked;
+        trkNoiseW.Enabled = !locked;
+        txtSttEndpoint.Enabled = !locked;
+        btnManageVoices.Enabled = !locked;
+    }
+
+    // ─── PTT ────────────────────────────────────────────────────────────────
 
     private void SetupHotkey()
     {
@@ -279,86 +351,56 @@ internal sealed partial class MainForm : Form
 
     private void OnPttPressed(object? sender, EventArgs e)
     {
-        Log("PTT pressed");
-        _orchestrator?.NotifyPttPressed();
-        try
-        {
+        _pipeline?.OnPttPressed();
         _sttSource?.SendOn();
     }
-        catch (Exception ex)
-        {
-            Log($"SendOn failed: {ex.Message}");
-        }
-    }
 
     private void OnPttReleased(object? sender, EventArgs e)
-    {
-        Log("PTT released");
-        try
     {
         _sttSource?.SendOff();
+        _pipeline?.OnPttReleased();
     }
-        catch (Exception ex)
+
+    // ─── STT transcript → pipeline ──────────────────────────────────────────
+
+    private void OnTranscript(object? sender, TranscriptEventArgs e)
     {
-            Log($"SendOff failed: {ex.Message}");
+        if (_pipeline is null || !_locked) return;
+
+        var msg = e.Message;
+        if (msg.Type == TranscriptType.Partial)
+        {
+            if (!string.IsNullOrWhiteSpace(msg.Text))
+            {
+                Log($"SEGMENT: \"{msg.Text}\"");
+                _pipeline.EnqueueSegment(msg.Text);
+            }
+        }
+        else
+        {
+            if (!string.IsNullOrWhiteSpace(msg.Text))
+            {
+                Log($"FINAL: \"{msg.Text}\"");
+                _pipeline.EnqueueFinal(msg.Text);
+            }
+            else
+            {
+                Log("FINAL: (empty)");
+                _pipeline.EnqueueFinal("");
+            }
         }
     }
 
-    private void OnBrowseFile(object? sender, EventArgs e)
-    {
-        using var dlg = new OpenFileDialog
-        {
-            Filter = "Text files (*.txt)|*.txt|All files (*.*)|*.*",
-            Title = "Select a text file to speak",
-        };
-
-        if (dlg.ShowDialog() == DialogResult.OK)
-        {
-            string text = File.ReadAllText(dlg.FileName);
-            lblFileName.Text = Path.GetFileName(dlg.FileName);
-            lblFileName.ForeColor = Color.Black;
-            Log($"Loaded: {dlg.FileName} ({text.Length} chars)");
-            _ = _orchestrator?.SynthesizeAsync(text);
-        }
-    }
-
-    private void OnTestVoice(object? sender, EventArgs e)
-    {
-        if (_orchestrator is null)
-        {
-            Log("Engine not initialized.");
-            return;
-        }
-
-        if (cmbVoice.SelectedItem is not string voiceName)
-        {
-            Log("No voice selected.");
-            return;
-        }
-
-        if (!VoiceCatalogue.IsVoiceInstalled(_voicesDir, voiceName))
-        {
-            Log($"Voice '{voiceName}' is not installed. Use Add/Remove to download it.");
-            return;
-        }
-
-        string text = txtTextInput.Text.Trim();
-        if (string.IsNullOrEmpty(text))
-            text = "Hello, this is a voice test.";
-
-        btnTestVoice.Enabled = false;
-        try
-        {
-            _ = _orchestrator.SynthesizeAsync(text);
-        }
-        finally
-        {
-            btnTestVoice.Enabled = true;
-        }
-    }
+    // ─── Voice manager ─────────────────────────────────────────────────────
 
     private void OnManageVoices(object? sender, EventArgs e)
     {
+        if (_locked)
+        {
+            Log("Release the model before managing voices.");
+            return;
+        }
+
         string currentVoice = cmbVoice.SelectedItem as string ?? string.Empty;
         using var dlg = new VoiceManagerForm(_voicesDir, currentVoice);
         dlg.ShowDialog(this);
@@ -376,31 +418,9 @@ internal sealed partial class MainForm : Form
         }
 
         SaveConfig();
-        _ = InitializeEngineAsync();
     }
 
-    private void OnVoiceChanged(object? sender, EventArgs e)
-    {
-        btnTestVoice.Enabled = cmbVoice.SelectedItem is string voiceName
-            && VoiceCatalogue.IsVoiceInstalled(_voicesDir, voiceName);
-
-        if (cmbVoice.SelectedItem is string name)
-        {
-            SaveConfig();
-            _ = ReinitializeEngineAsync(name);
-        }
-    }
-
-    private async Task ReinitializeEngineAsync(string voiceName)
-    {
-        if (!VoiceCatalogue.IsVoiceInstalled(_voicesDir, voiceName))
-        {
-            btnTestVoice.Enabled = false;
-            return;
-        }
-
-        await InitializeEngineAsync();
-    }
+    // ─── Slider / endpoint ─────────────────────────────────────────────────
 
     private void OnSliderScroll(object? sender, EventArgs e)
     {
@@ -412,72 +432,15 @@ internal sealed partial class MainForm : Form
     private void OnSliderReleased(object? sender, MouseEventArgs e)
     {
         SaveConfig();
-        if (cmbVoice.SelectedItem is string name && VoiceCatalogue.IsVoiceInstalled(_voicesDir, name))
-        {
-            _ = InitializeEngineAsync();
-        }
     }
 
-    private void OnOutputChanged(object? sender, EventArgs e)
+    private void OnSttEndpointChanged(object? sender, EventArgs e)
     {
-        if (_orchestrator is not null)
-            _orchestrator.OutputDeviceName = cmbOutput.SelectedItem as string ?? string.Empty;
-        Log($"Output device: {_orchestrator?.OutputDeviceName}");
+        _config.SttEndpoint = txtSttEndpoint.Text;
         SaveConfig();
     }
 
-    private void PopulateInterfaces()
-    {
-        cmbInterface.Items.Clear();
-        foreach (var (ip, name) in DhcpSttSource.GetAvailableInterfaces())
-        {
-            cmbInterface.Items.Add(name);
-            cmbInterface.Items[^1] = name;
-            cmbInterface.Items[cmbInterface.Items.Count - 1] = name;
-        }
-
-        if (!string.IsNullOrEmpty(_config.InterfaceIp))
-        {
-            for (int i = 0; i < cmbInterface.Items.Count; i++)
-            {
-                if (cmbInterface.Items[i] is string s && s.Contains(_config.InterfaceIp))
-                {
-                    cmbInterface.SelectedIndex = i;
-                    return;
-                }
-            }
-        }
-
-        if (cmbInterface.Items.Count > 0)
-            cmbInterface.SelectedIndex = 0;
-    }
-
-    private void OnInterfaceChanged(object? sender, EventArgs e)
-    {
-        if (cmbInterface.SelectedItem is not string selected)
-            return;
-
-        string? ip = ExtractIpFromDisplay(selected);
-        if (ip is null) return;
-
-        _config.InterfaceIp = ip;
-        SaveConfig();
-
-        if (_sttSource is not null)
-        {
-            _sttSource.DisposeAsync().AsTask().Wait(2000);
-            _sttSource = null;
-            _ = InitializeEngineAsync();
-        }
-    }
-
-    private static string? ExtractIpFromDisplay(string display)
-    {
-        int start = display.IndexOf('(');
-        int end = display.IndexOf(')');
-        if (start < 0 || end <= start) return null;
-        return display[(start + 1)..end];
-    }
+    // ─── Tray ──────────────────────────────────────────────────────────────
 
     private void SetupTray()
     {
@@ -535,6 +498,8 @@ internal sealed partial class MainForm : Form
         Activate();
     }
 
+    // ─── Config ────────────────────────────────────────────────────────────
+
     private void SaveConfig()
     {
         _config.Voice = cmbVoice.SelectedItem as string ?? string.Empty;
@@ -544,10 +509,12 @@ internal sealed partial class MainForm : Form
         _config.LengthScale = trkSpeed.Value;
         _config.NoiseWScale = trkNoiseW.Value;
         _config.MinimizeToTray = chkMinimizeToTray.Checked;
-        _config.InterfaceIp = ExtractIpFromDisplay(cmbInterface.SelectedItem as string ?? "") ?? "";
+        _config.SttEndpoint = txtSttEndpoint.Text;
         _config.Save();
     }
 
+    // ─── Logging ───────────────────────────────────────────────────────────
+
     private void Log(string message)
     {
         if (IsDisposed) return;
@@ -562,12 +529,13 @@ internal sealed partial class MainForm : Form
         txtLog.ScrollToCaret();
     }
 
+    // ─── Shutdown ──────────────────────────────────────────────────────────
+
     private void OnFormClosing(object? sender, FormClosingEventArgs e)
     {
         _pttHotkey?.Dispose();
         _trayIcon!.Visible = false;
-        _orchestrator?.DisposeAsync().AsTask().Wait(2000);
-        _sttSource?.DisposeAsync().AsTask().Wait(2000);
+        ReleaseModel();
         SaveConfig();
     }
 }

Robovoice.App/Orchestrator.cs

@@ -1,283 +0,0 @@
-using Robovoice.Core;
-using Robovoice.Tts.LibPiper;
-
-namespace Robovoice.App;
-
-internal sealed class Orchestrator : IAsyncDisposable
-{
-    private readonly LibPiperTtsEngine _tts;
-    private readonly AudioOutput _audioOutput;
-    private readonly ISttSource _sttSource;
-    private readonly Action<string> _log;
-    private CancellationTokenSource? _currentCts;
-    private bool _disposed;
-
-    private readonly object _stateLock = new();
-    private readonly Queue<string> _pendingTexts = new();
-    private readonly List<float[]> _audioBuffer = new();
-    private int _bufferSampleRate;
-    private bool _playing;
-    private Task? _synthTask;
-    private CancellationTokenSource? _synthCts;
-    private System.Threading.Timer? _segmentTimer;
-    private static readonly TimeSpan SegmentTimeout = TimeSpan.FromMilliseconds(250);
-
-    public string OutputDeviceName { get; set; } = string.Empty;
-
-    public Orchestrator(
-        LibPiperTtsEngine tts,
-        AudioOutput audioOutput,
-        ISttSource sttSource,
-        Action<string> log)
-    {
-        _tts = tts;
-        _audioOutput = audioOutput;
-        _audioOutput.Log = log;
-        _sttSource = sttSource;
-        _log = log;
-        _sttSource.TranscriptReceived += OnTranscript;
-    }
-
-    private void OnTranscript(object? sender, TranscriptEventArgs e)
-    {
-        var msg = e.Message;
-
-        lock (_stateLock)
-        {
-            if (msg.Type == TranscriptType.Partial)
-            {
-                if (!string.IsNullOrWhiteSpace(msg.Text))
-                {
-                    _log($"SEGMENT: \"{msg.Text}\" ({msg.Text.Length} chars)");
-                    _pendingTexts.Enqueue(msg.Text);
-                    EnsureSynthTask();
-                    ResetSegmentTimer();
-                }
-            }
-            else
-            {
-                if (!string.IsNullOrWhiteSpace(msg.Text))
-                {
-                    _log($"FINAL: \"{msg.Text}\" ({msg.Text.Length} chars)");
-                    _pendingTexts.Enqueue(msg.Text);
-                    EnsureSynthTask();
-                }
-                else
-                {
-                    _log("FINAL: (empty)");
-                }
-
-                CancelSegmentTimer();
-                TransitionToPlaying();
-            }
-        }
-    }
-
-    private void ResetSegmentTimer()
-    {
-        _segmentTimer?.Dispose();
-        _segmentTimer = new System.Threading.Timer(_ => OnSegmentTimeout(), null, SegmentTimeout, Timeout.InfiniteTimeSpan);
-    }
-
-    private void CancelSegmentTimer()
-    {
-        _segmentTimer?.Dispose();
-        _segmentTimer = null;
-    }
-
-    private void OnSegmentTimeout()
-    {
-        _log("STT: segment timeout, starting playback early");
-        lock (_stateLock)
-        {
-            TransitionToPlaying();
-        }
-    }
-
-    private void TransitionToPlaying()
-    {
-        if (_playing)
-            return;
-
-        if (_audioBuffer.Count == 0)
-        {
-            if (_synthTask is null || _synthTask.IsCompleted)
-            {
-                _log("TTS: nothing to play");
-                CancelSegmentTimer();
-            }
-            return;
-        }
-
-        _playing = true;
-        CancelSegmentTimer();
-
-        int sampleRate = _bufferSampleRate;
-        var chunks = _audioBuffer.ToList();
-        _audioBuffer.Clear();
-
-        _log($"TTS: playing {chunks.Count} buffered chunks ({sampleRate} Hz)");
-
-        _audioOutput.Start(sampleRate, OutputDeviceName);
-        foreach (var samples in chunks)
-        {
-            _audioOutput.WriteSamples(samples);
-        }
-    }
-
-    private void EnsureSynthTask()
-    {
-        if (_synthTask is not null && !_synthTask.IsCompleted)
-            return;
-
-        _synthCts?.Cancel();
-        _synthCts = new CancellationTokenSource();
-        _synthTask = SynthLoopAsync(_synthCts.Token);
-    }
-
-    private async Task SynthLoopAsync(CancellationToken ct)
-    {
-        while (true)
-        {
-            string text;
-            lock (_stateLock)
-            {
-                if (_pendingTexts.Count == 0)
-                    break;
-                text = _pendingTexts.Dequeue();
-            }
-
-            try
-            {
-                await foreach (var chunk in _tts.SynthesizeAsync(text, ct))
-                {
-                    lock (_stateLock)
-                    {
-                        if (_playing)
-                        {
-                            _audioOutput.WriteSamples(chunk.Samples);
-                        }
-                        else
-                        {
-                            _bufferSampleRate = chunk.SampleRate;
-                            _audioBuffer.Add(chunk.Samples);
-                        }
-                    }
-                }
-            }
-            catch (OperationCanceledException)
-            {
-                break;
-            }
-            catch (Exception ex)
-            {
-                _log($"TTS error: {ex.Message}");
-            }
-        }
-
-        lock (_stateLock)
-        {
-            if (_playing)
-            {
-                _audioOutput.Flush();
-                _log("TTS: synthesis complete, flushed");
-            }
-        }
-    }
-
-    public void NotifyPttPressed()
-    {
-        lock (_stateLock)
-        {
-            CancelSegmentTimer();
-            _synthCts?.Cancel();
-            _synthCts?.Dispose();
-            _synthTask = null;
-
-            if (_playing)
-            {
-                _audioOutput.Stop();
-                _playing = false;
-            }
-
-            _pendingTexts.Clear();
-            _audioBuffer.Clear();
-            _bufferSampleRate = 0;
-        }
-    }
-
-    public async Task InitializeTtsAsync()
-    {
-        _log("Initializing TTS engine...");
-        await _tts.InitializeAsync();
-        _log($"TTS ready (sample rate: {_tts.SampleRate} Hz)");
-    }
-
-    public async Task SynthesizeAsync(string text)
-    {
-        _currentCts?.Cancel();
-        _currentCts = new CancellationTokenSource();
-        var ct = _currentCts.Token;
-
-        var sw = System.Diagnostics.Stopwatch.StartNew();
-        _log($"Speak: \"{text}\" ({text.Length} chars)");
-
-        try
-        {
-            bool started = false;
-            int chunkCount = 0;
-            int totalSamples = 0;
-
-            await foreach (var chunk in _tts.SynthesizeAsync(text, ct))
-            {
-                if (!started)
-                {
-                    started = true;
-                    _audioOutput.Start(chunk.SampleRate, OutputDeviceName);
-                    _log($"TTS: first chunk ({sw.ElapsedMilliseconds}ms)");
-                }
-
-                _audioOutput.WriteSamples(chunk.Samples);
-                chunkCount++;
-                totalSamples += chunk.Samples.Length;
-            }
-
-            if (!started)
-            {
-                _log("TTS: no audio produced");
-            }
-            else
-            {
-                _audioOutput.Flush();
-                double durationSec = (double)totalSamples / _tts.SampleRate;
-                _log($"TTS: done ({chunkCount} chunks, {durationSec:F2}s audio, {sw.ElapsedMilliseconds}ms)");
-            }
-        }
-        catch (OperationCanceledException)
-        {
-            _log("TTS: cancelled");
-            _audioOutput.Stop();
-        }
-        catch (Exception ex)
-        {
-            _log($"TTS error: {ex.Message}");
-        }
-
-        sw.Stop();
-    }
-
-    public async ValueTask DisposeAsync()
-    {
-        if (_disposed) return;
-        _currentCts?.Cancel();
-        _currentCts?.Dispose();
-        _synthCts?.Cancel();
-        _synthCts?.Dispose();
-        CancelSegmentTimer();
-        _sttSource.TranscriptReceived -= OnTranscript;
-        await _sttSource.DisposeAsync();
-        await _tts.DisposeAsync();
-        _audioOutput.Dispose();
-        _disposed = true;
-    }
-}

Robovoice.App/Robovoice.App.csproj

@@ -3,7 +3,7 @@
   <ItemGroup>
     <ProjectReference Include="..\Robovoice.Core\Robovoice.Core.csproj" />
     <ProjectReference Include="..\Robovoice.Tts.LibPiper\Robovoice.Tts.LibPiper.csproj" />
-    <ProjectReference Include="..\Robovoice.Stt.Dhcp\Robovoice.Stt.Dhcp.csproj" />
+    <ProjectReference Include="..\Robovoice.Stt.Tcp\Robovoice.Stt.Tcp.csproj" />
   </ItemGroup>
 
   <ItemGroup>

Robovoice.App/VoicePipeline.cs

@@ -0,0 +1,187 @@
+using System.Collections.Concurrent;
+using Robovoice.Core;
+using Robovoice.Tts.LibPiper;
+
+namespace Robovoice.App;
+
+internal sealed record TextItem(uint Session, string Text);
+
+internal sealed record AudioItem(uint Session, float[]? Samples);
+
+internal sealed class VoicePipeline : IDisposable
+{
+    private readonly LibPiperTtsEngine _tts;
+    private readonly AudioOutput _audioOutput;
+    private readonly Action<string> _log;
+
+    private BlockingCollection<TextItem> _textQueue = new();
+    private BlockingCollection<AudioItem> _audioQueue = new();
+    private ManualResetEventSlim _gate = new(false);
+    private volatile uint _currentSession;
+    private volatile bool _running;
+
+    private Thread? _synthThread;
+    private Thread? _playerThread;
+    private bool _disposed;
+
+    public string OutputDeviceName { get; set; } = string.Empty;
+
+    public VoicePipeline(
+        LibPiperTtsEngine tts,
+        AudioOutput audioOutput,
+        Action<string> log)
+    {
+        _tts = tts;
+        _audioOutput = audioOutput;
+        _audioOutput.Log = log;
+        _log = log;
+    }
+
+    public void Start()
+    {
+        _running = true;
+        _synthThread = new Thread(SynthLoop) { IsBackground = true, Name = "VoicePipeline-Synth" };
+        _playerThread = new Thread(PlayerLoop) { IsBackground = true, Name = "VoicePipeline-Player" };
+        _synthThread.Start();
+        _playerThread.Start();
+    }
+
+    public void EnqueueSegment(string text)
+    {
+        if (!_running) return;
+        uint session = _currentSession;
+        _textQueue.Add(new TextItem(session, text));
+    }
+
+    public void EnqueueFinal(string text)
+    {
+        if (!_running) return;
+        uint session = _currentSession;
+        if (!string.IsNullOrEmpty(text))
+            _textQueue.Add(new TextItem(session, text));
+        _textQueue.Add(new TextItem(session, ""));
+        _gate.Reset();
+    }
+
+    public void OnPttPressed()
+    {
+        _currentSession++;
+        _gate.Reset();
+        _audioOutput.Stop();
+    }
+
+    public void OnPttReleased()
+    {
+        _gate.Set();
+    }
+
+    private void SynthLoop()
+    {
+        foreach (var item in _textQueue.GetConsumingEnumerable())
+        {
+            if (!_running) break;
+
+            if (item.Session != _currentSession)
+                continue;
+
+            if (item.Text.Length == 0)
+            {
+                // :F sentinel — signal end of utterance to player
+                _audioQueue.Add(new AudioItem(item.Session, null));
+                continue;
+            }
+
+            try
+            {
+                foreach (var chunk in _tts.SynthesizeSync(item.Text))
+                {
+                    if (item.Session != _currentSession)
+                    {
+                        // Session changed mid-synthesis — drain piper cleanly
+                        _tts.SynthesizeDrain();
+                        break;
+                    }
+
+                    _audioQueue.Add(new AudioItem(item.Session, chunk.Samples));
+                }
+            }
+            catch (Exception ex)
+            {
+                _log($"TTS error: {ex.Message}");
+            }
+        }
+    }
+
+    private void PlayerLoop()
+    {
+        while (_running)
+        {
+            _gate.Wait();
+            if (!_running) break;
+
+            // Drain stale items, then wait for real audio for current session
+            AudioItem firstItem = default!;
+            while (_running)
+            {
+                try { firstItem = _audioQueue.Take(); }
+                catch (InvalidOperationException) { return; }
+
+                if (firstItem.Session != _currentSession)
+                    continue;
+
+                break;
+            }
+
+            if (!_running) break;
+
+            if (firstItem.Samples == null)
+            {
+                // :F with no audio — nothing to play
+                _gate.Reset();
+                continue;
+            }
+
+            _audioOutput.Start(_tts.SampleRate, OutputDeviceName);
+            _audioOutput.WriteSamples(firstItem.Samples);
+            _log($"TTS: playback started (session {firstItem.Session})");
+
+            foreach (var item in _audioQueue.GetConsumingEnumerable())
+            {
+                if (!_running) break;
+
+                if (item.Session != _currentSession)
+                    break;
+
+                if (item.Samples == null)
+                    break;
+
+                _audioOutput.WriteSamples(item.Samples);
+            }
+
+            _audioOutput.Flush();
+            _log("TTS: playback finished");
+            _gate.Reset();
+        }
+    }
+
+    public void Dispose()
+    {
+        if (_disposed) return;
+        _disposed = true;
+        _running = false;
+
+        _gate.Set();
+        _textQueue.CompleteAdding();
+        _audioQueue.CompleteAdding();
+
+        _synthThread?.Join(5000);
+        _playerThread?.Join(5000);
+
+        _textQueue.Dispose();
+        _audioQueue.Dispose();
+        _gate.Dispose();
+
+        _tts.DisposeAsync().AsTask().Wait();
+        _audioOutput.Dispose();
+    }
+}

Robovoice.Stt.Dhcp/DhcpSttSource.cs

@@ -1,277 +0,0 @@
-using System.Net;
-using System.Net.NetworkInformation;
-using System.Net.Sockets;
-using System.Text;
-using Robovoice.Core;
-
-namespace Robovoice.Stt.Dhcp;
-
-public sealed class DhcpSttSource : ISttSource
-{
-    private const string Magic = "HKMSTR";
-    private const int DhcpClientPort = 68;
-    private const int DhcpServerPort = 67;
-    private static readonly TimeSpan NopInterval = TimeSpan.FromMilliseconds(50);
-
-    private Socket? _sock;
-    private CancellationTokenSource? _cts;
-    private Task? _receiveTask;
-    private Task? _nopTask;
-    private EndPoint _broadcastEp = new IPEndPoint(IPAddress.Broadcast, DhcpServerPort);
-    private uint _nonce;
-    private uint _session;
-    private bool _disposed;
-
-    public string InterfaceIp { get; set; } = string.Empty;
-
-    public Action<string>? Log { get; set; }
-
-    public event TranscriptEventHandler? TranscriptReceived;
-
-    public Task StartAsync(CancellationToken ct = default)
-    {
-        ObjectDisposedException.ThrowIf(_disposed, this);
-        if (_cts is not null)
-            return Task.CompletedTask;
-
-        string ip = string.IsNullOrEmpty(InterfaceIp) ? AutoDetectInterfaceIp() ?? "0.0.0.0" : InterfaceIp;
-
-        _sock = new Socket(AddressFamily.InterNetwork, SocketType.Dgram, ProtocolType.Udp);
-        _sock.SetSocketOption(SocketOptionLevel.Socket, SocketOptionName.ReuseAddress, true);
-        _sock.EnableBroadcast = true;
-        _sock.Bind(new IPEndPoint(IPAddress.Parse(ip), DhcpClientPort));
-
-        Log?.Invoke($"STT: bound to {ip}:{DhcpClientPort}, broadcasting to :{DhcpServerPort}");
-
-        _cts = CancellationTokenSource.CreateLinkedTokenSource(ct);
-        _receiveTask = ReceiveLoopAsync(_cts.Token);
-        return Task.CompletedTask;
-    }
-
-    public async Task StopAsync(CancellationToken ct = default)
-    {
-        StopNop();
-
-        if (_cts is not null)
-            _cts.Cancel();
-
-        _sock?.Dispose();
-        _sock = null;
-
-        if (_receiveTask is not null)
-        {
-            try { await _receiveTask.WaitAsync(ct); }
-            catch { }
-            _receiveTask = null;
-        }
-
-        _cts?.Dispose();
-        _cts = null;
-    }
-
-    public void SendOn()
-    {
-        if (_cts is null)
-            return;
-
-        _session++;
-        _nonce = 0;
-        Log?.Invoke($"STT: session {_session} started");
-        SendNop();
-        _nopTask = NopLoopAsync(_cts.Token);
-    }
-
-    public void SendOff()
-    {
-        StopNop();
-        SendControl($"HKMSTR:OFF {_session} {_nonce}");
-    }
-
-    private void StopNop()
-    {
-        if (_nopTask is not null)
-        {
-            try { _nopTask.Wait(2000); } catch { }
-            _nopTask = null;
-        }
-    }
-
-    private async Task NopLoopAsync(CancellationToken ct)
-    {
-        while (!ct.IsCancellationRequested)
-        {
-            try { await Task.Delay(NopInterval, ct); }
-            catch (OperationCanceledException) { break; }
-
-            SendNop();
-        }
-    }
-
-    private void SendNop()
-    {
-        _nonce++;
-        SendControl($"HKMSTR {_session} {_nonce}");
-    }
-
-    private void SendControl(string message)
-    {
-        if (_sock is null)
-            return;
-
-        try
-        {
-            byte[] payload = Encoding.UTF8.GetBytes(message + "\n");
-            _sock.SendTo(payload, _broadcastEp);
-        }
-        catch (Exception ex)
-        {
-            Log?.Invoke($"STT: send failed: {ex.Message}");
-        }
-    }
-
-    private async Task ReceiveLoopAsync(CancellationToken ct)
-    {
-        byte[] buffer = new byte[4096];
-        EndPoint fromEp = new IPEndPoint(IPAddress.Any, 0);
-
-        while (!ct.IsCancellationRequested)
-        {
-            int received;
-            try
-            {
-                if (!_sock!.Poll(500_000, SelectMode.SelectRead))
-                    continue;
-                received = _sock.ReceiveFrom(buffer, ref fromEp);
-            }
-            catch (OperationCanceledException)
-            {
-                break;
-            }
-            catch (ObjectDisposedException)
-            {
-                break;
-            }
-            catch (Exception ex)
-            {
-                Log?.Invoke($"STT: receive error: {ex.Message}");
-                continue;
-            }
-
-            string text = Encoding.UTF8.GetString(buffer, 0, received).TrimEnd('\n', '\r');
-            if (!text.StartsWith(Magic))
-                continue;
-
-            TranscriptMessage? message = ParseReply(text);
-            if (message is null)
-                continue;
-
-            TranscriptReceived?.Invoke(this, new TranscriptEventArgs
-            {
-                Message = message,
-            });
-        }
-    }
-
-    private TranscriptMessage? ParseReply(string text)
-    {
-        // Format: HKMSTR:P <session> <text>  or  HKMSTR:F <session> <text>
-        // <text> may be empty.
-        string prefix;
-        TranscriptType type;
-
-        if (text.StartsWith("HKMSTR:P "))
-        {
-            prefix = "HKMSTR:P ";
-            type = TranscriptType.Partial;
-        }
-        else if (text.StartsWith("HKMSTR:F "))
-        {
-            prefix = "HKMSTR:F ";
-            type = TranscriptType.Final;
-        }
-        else
-        {
-            return null;
-        }
-
-        string rest = text[prefix.Length..];
-
-        int spaceIndex = rest.IndexOf(' ');
-        if (spaceIndex < 0)
-        {
-            if (uint.TryParse(rest, out uint sessionOnly))
-            {
-                if (sessionOnly != _session)
-                    return null;
-                return new TranscriptMessage(type, string.Empty);
-            }
-            return null;
-        }
-
-        string sessionStr = rest[..spaceIndex];
-        if (!uint.TryParse(sessionStr, out uint session))
-            return null;
-
-        if (session != _session)
-        {
-            Log?.Invoke($"STT: dropping stale reply (session {session} != current {_session})");
-            return null;
-        }
-
-        string transcript = rest[(spaceIndex + 1)..];
-        return new TranscriptMessage(type, transcript);
-    }
-
-    public static List<(string Ip, string Name)> GetAvailableInterfaces()
-    {
-        var result = new List<(string, string)>();
-
-        foreach (var nic in NetworkInterface.GetAllNetworkInterfaces())
-        {
-            if (nic.OperationalStatus != OperationalStatus.Up)
-                continue;
-            if (nic.NetworkInterfaceType == NetworkInterfaceType.Loopback)
-                continue;
-
-            string desc = nic.Description;
-            if (desc.Contains("WireGuard", StringComparison.OrdinalIgnoreCase))
-                continue;
-
-            foreach (var addr in nic.GetIPProperties().UnicastAddresses)
-            {
-                if (addr.Address.AddressFamily != AddressFamily.InterNetwork)
-                    continue;
-
-                string ip = addr.Address.ToString();
-                if (ip.StartsWith("192.168.") || ip.StartsWith("10.") ||
-                    ip.StartsWith("172.16.") || ip.StartsWith("172.17.") ||
-                    ip.StartsWith("172.18.") || ip.StartsWith("172.19.") ||
-                    ip.StartsWith("172.20.") || ip.StartsWith("172.21.") ||
-                    ip.StartsWith("172.22.") || ip.StartsWith("172.23.") ||
-                    ip.StartsWith("172.24.") || ip.StartsWith("172.25.") ||
-                    ip.StartsWith("172.26.") || ip.StartsWith("172.27.") ||
-                    ip.StartsWith("172.28.") || ip.StartsWith("172.29.") ||
-                    ip.StartsWith("172.30.") || ip.StartsWith("172.31."))
-                {
-                    result.Add((ip, $"{nic.Name} ({ip})"));
-                }
-            }
-        }
-
-        return result;
-    }
-
-    private static string? AutoDetectInterfaceIp()
-    {
-        foreach (var (ip, _) in GetAvailableInterfaces())
-            return ip;
-        return null;
-    }
-
-    public async ValueTask DisposeAsync()
-    {
-        if (_disposed) return;
-        await StopAsync();
-        _disposed = true;
-    }
-}

Robovoice.Stt.Tcp/TcpSttSource.cs

@@ -0,0 +1,291 @@
+using System.Net;
+using System.Net.Sockets;
+using System.Text;
+using Robovoice.Core;
+
+namespace Robovoice.Stt.Tcp;
+
+public sealed class TcpSttSource : ISttSource
+{
+    private TcpClient? _tcp;
+    private NetworkStream? _stream;
+    private StreamWriter? _writer;
+    private Thread? _connectThread;
+    private volatile bool _running;
+    private readonly object _sendLock = new();
+    private uint _session;
+    private bool _disposed;
+
+    public string Endpoint { get; set; } = "127.0.0.1:6996";
+
+    public Action<string>? Log { get; set; }
+
+    public event TranscriptEventHandler? TranscriptReceived;
+
+    public Task StartAsync(CancellationToken ct = default)
+    {
+        ObjectDisposedException.ThrowIf(_disposed, this);
+        if (_running)
+            return Task.CompletedTask;
+
+        _running = true;
+        _connectThread = new Thread(ConnectLoop) { IsBackground = true, Name = "TcpSttSource-Connect" };
+        _connectThread.Start();
+
+        return Task.CompletedTask;
+    }
+
+    public Task StopAsync(CancellationToken ct = default)
+    {
+        _running = false;
+
+        lock (_sendLock)
+        {
+            _writer?.Dispose();
+            _stream?.Dispose();
+            _tcp?.Close();
+            _writer = null;
+            _stream = null;
+            _tcp = null;
+        }
+
+        // Threads are background — they'll die when the process exits.
+        // Closing the socket unblocks any pending Read.
+        _connectThread?.Join(1000);
+
+        return Task.CompletedTask;
+    }
+
+    private void ConnectLoop()
+    {
+        while (_running)
+        {
+            IPEndPoint? endpoint = ParseEndpoint(Endpoint);
+            if (endpoint is null)
+            {
+                Log?.Invoke($"STT: invalid endpoint '{Endpoint}'");
+                SleepInterruptible(3000);
+                continue;
+            }
+
+            try
+            {
+                var tcp = new TcpClient();
+                tcp.Connect(endpoint.Address, endpoint.Port);
+                tcp.NoDelay = true;
+
+                lock (_sendLock)
+                {
+                    _tcp = tcp;
+                    _stream = tcp.GetStream();
+                    _writer = new StreamWriter(_stream, new UTF8Encoding(false)) { AutoFlush = true };
+                }
+
+                Log?.Invoke($"STT: connected to {Endpoint}");
+
+                // Blocking receive loop — runs until disconnected or stopped.
+                ReceiveLoop();
+
+                Log?.Invoke("STT: disconnected");
+            }
+            catch (Exception ex)
+            {
+                if (_running)
+                    Log?.Invoke($"STT: connection failed ({ex.Message}), retrying...");
+            }
+            finally
+            {
+                lock (_sendLock)
+                {
+                    _writer?.Dispose();
+                    _stream?.Dispose();
+                    _tcp?.Close();
+                    _writer = null;
+                    _stream = null;
+                    _tcp = null;
+                }
+            }
+
+            if (_running)
+                SleepInterruptible(3000);
+        }
+    }
+
+    private void ReceiveLoop()
+    {
+        byte[] buffer = new byte[4096];
+        StringBuilder lineBuf = new();
+
+        while (_running)
+        {
+            NetworkStream? stream;
+            lock (_sendLock)
+            {
+                stream = _stream;
+            }
+
+            if (stream is null)
+                break;
+
+            int bytesRead;
+            try
+            {
+                bytesRead = stream.Read(buffer, 0, buffer.Length);
+            }
+            catch
+            {
+                break;
+            }
+
+            if (bytesRead == 0)
+                break;
+
+            for (int i = 0; i < bytesRead; i++)
+            {
+                byte b = buffer[i];
+                if (b == '\n')
+                {
+                    string line = lineBuf.ToString().TrimEnd('\r');
+                    lineBuf.Clear();
+
+                    TranscriptMessage? message = ParseReply(line);
+                    if (message is not null)
+                    {
+                        TranscriptReceived?.Invoke(this, new TranscriptEventArgs
+                        {
+                            Message = message,
+                        });
+                    }
+                }
+                else
+                {
+                    lineBuf.Append((char)b);
+                }
+            }
+        }
+    }
+
+    private void SleepInterruptible(int ms)
+    {
+        int slice = 100;
+        int waited = 0;
+        while (_running && waited < ms)
+        {
+            int chunk = Math.Min(slice, ms - waited);
+            Thread.Sleep(chunk);
+            waited += chunk;
+        }
+    }
+
+    public void SendOn()
+    {
+        _session++;
+        Send($"ON {_session}");
+    }
+
+    public void SendOff()
+    {
+        Send($"OFF {_session}");
+    }
+
+    private void Send(string message)
+    {
+        lock (_sendLock)
+        {
+            if (_writer is null)
+                return;
+
+            try
+            {
+                _writer.WriteLine(message);
+            }
+            catch
+            {
+                Log?.Invoke($"STT: failed to send '{message}' (not connected?)");
+            }
+        }
+    }
+
+    private static IPEndPoint? ParseEndpoint(string endpoint)
+    {
+        int colon = endpoint.LastIndexOf(':');
+        if (colon <= 0)
+            return null;
+
+        string host = endpoint[..colon];
+        if (!int.TryParse(endpoint[(colon + 1)..], out int port))
+            return null;
+
+        if (IPAddress.TryParse(host, out var addr))
+            return new IPEndPoint(addr, port);
+
+        try
+        {
+            var addresses = Dns.GetHostAddresses(host);
+            addr = addresses.FirstOrDefault(a => a.AddressFamily == AddressFamily.InterNetwork);
+            if (addr is null)
+                return null;
+            return new IPEndPoint(addr, port);
+        }
+        catch
+        {
+            return null;
+        }
+    }
+
+    private TranscriptMessage? ParseReply(string line)
+    {
+        if (line.StartsWith("P ", StringComparison.Ordinal))
+        {
+            string rest = line["P ".Length..];
+            int space = rest.IndexOf(' ');
+            if (space < 0)
+                return null;
+
+            if (!uint.TryParse(rest[..space], out uint session))
+                return null;
+
+            if (session != _session)
+            {
+                Log?.Invoke($"STT: dropping stale reply (session {session} != current {_session})");
+                return null;
+            }
+
+            return new TranscriptMessage(TranscriptType.Partial, rest[(space + 1)..]);
+        }
+
+        if (line.StartsWith("F ", StringComparison.Ordinal))
+        {
+            string rest = line["F ".Length..];
+            int space = rest.IndexOf(' ');
+
+            if (space < 0)
+            {
+                if (uint.TryParse(rest, out uint session) && session == _session)
+                    return new TranscriptMessage(TranscriptType.Final, string.Empty);
+                return null;
+            }
+
+            if (!uint.TryParse(rest[..space], out uint ses))
+                return null;
+
+            if (ses != _session)
+            {
+                Log?.Invoke($"STT: dropping stale reply (session {ses} != current {_session})");
+                return null;
+            }
+
+            return new TranscriptMessage(TranscriptType.Final, rest[(space + 1)..]);
+        }
+
+        return null;
+    }
+
+    public ValueTask DisposeAsync()
+    {
+        if (_disposed) return ValueTask.CompletedTask;
+        _disposed = true;
+        StopAsync();
+        return ValueTask.CompletedTask;
+    }
+}

Robovoice.Tts.LibPiper/LibPiperTtsEngine.cs

@@ -120,6 +120,71 @@ public sealed class LibPiperTtsEngine : ITtsEngine
         }
     }
 
+    public IEnumerable<AudioChunk> SynthesizeSync(string text)
+    {
+        ObjectDisposedException.ThrowIf(_disposed, this);
+        if (!_initialized)
+            throw new InvalidOperationException("Engine not initialized.");
+
+        var options = new PiperSynthesizeOptions
+        {
+            SpeakerId = 0,
+            LengthScale = _lengthScale,
+            NoiseScale = _noiseScale,
+            NoiseWScale = _noiseWScale,
+        };
+
+        byte[] textBytes = System.Text.Encoding.UTF8.GetBytes(EnsureTerminator(text) + "\0");
+        GCHandle textPin = GCHandle.Alloc(textBytes, GCHandleType.Pinned);
+        try
+        {
+            int startResult = PiperNative.piper_synthesize_start(
+                _synth, textPin.AddrOfPinnedObject(), in options);
+            if (startResult != PiperNative.PiperOk)
+                throw new InvalidOperationException($"piper_synthesize_start failed: {startResult}");
+
+            while (true)
+            {
+                PiperAudioChunk chunk = default;
+                int result = PiperNative.piper_synthesize_next(_synth, out chunk);
+
+                if (chunk.NumSamples > 0 && chunk.Samples != IntPtr.Zero)
+                {
+                    int numSamples = (int)chunk.NumSamples;
+                    float[] samples = new float[numSamples];
+                    Marshal.Copy(chunk.Samples, samples, 0, numSamples);
+
+                    if (chunk.SampleRate > 0)
+                        _sampleRate = chunk.SampleRate;
+
+                    yield return new AudioChunk(samples, _sampleRate);
+                }
+
+                if (result == PiperNative.PiperDone || chunk.IsLast)
+                    break;
+
+                if (result < 0)
+                    throw new InvalidOperationException($"piper_synthesize_next failed: {result}");
+            }
+        }
+        finally
+        {
+            textPin.Free();
+        }
+    }
+
+    public void SynthesizeDrain()
+    {
+        if (!_initialized || _synth == IntPtr.Zero)
+            return;
+
+        PiperAudioChunk chunk;
+        while (PiperNative.piper_synthesize_next(_synth, out chunk) != PiperNative.PiperDone)
+        {
+            if (chunk.IsLast) break;
+        }
+    }
+
     private static string EnsureTerminator(string text)
     {
         string trimmed = text.TrimEnd();

Robovoice.slnx

@@ -2,7 +2,6 @@
   <Project Path="Robovoice.App/Robovoice.App.csproj" />
   <Project Path="Robovoice.Core/Robovoice.Core.csproj" />
   <Project Path="Robovoice.Stt.File/Robovoice.Stt.File.csproj" />
-  <Project Path="Robovoice.Stt.Dhcp/Robovoice.Stt.Dhcp.csproj" />
+  <Project Path="Robovoice.Stt.Tcp/Robovoice.Stt.Tcp.csproj" />
   <Project Path="Robovoice.Tts.LibPiper/Robovoice.Tts.LibPiper.csproj" />
-  <Project Path="DhcpTunnelTest/Robovoice.DhcpTunnelTest.csproj" />
 </Solution>

Server/PROTOCOL.md

@@ -1,136 +0,0 @@
-# Robovoice DHCP Tunnel Protocol
-
-## Overview
-
-Robovoice communicates with a remote STT server by tunneling through the
-DHCP UDP ports (68→67). This exploits a common killswitch exception: VPN
-software (e.g. WireGuard) blocks all traffic except DHCP, which is allowed
-for network connectivity maintenance.
-
-```
-[Robovoice client] --broadcast UDP :68→:67--> [STT server]
-[Robovoice client] <--unicast UDP :67→:68-- [STT server]
-```
-
-The client broadcasts NOP heartbeats while PTT is held. The server starts
-recording on the first NOP and stops when it receives OFF or when 150ms
-pass with no NOPs.
-
-## Transport
-
-- **Protocol:** UDP (connectionless, unreliable)
-- **Client → Server:** broadcast, source port 68, dest port 67
-- **Server → Client:** unicast, source port 67, dest port 68
-- **Client binds:** to a specific LAN interface IP on port 68 (with
-  `SO_REUSEADDR` to coexist with the Windows DHCP service)
-- **No connection state** — purely fire-and-forget datagrams
-
-## Wire format
-
-All messages are plain text, newline-terminated (`\n`). Every message starts
-with the 6-byte magic `HKMSTR` to distinguish our traffic from real DHCP.
-
-### Client → Server
-
-**NOP (heartbeat while PTT held):**
-```
-HKMSTR <session> <nonce>\n
-```
-Sent every 50ms while PTT is held. The session is an incrementing unsigned
-integer that identifies the current PTT utterance (incremented on each PTT
-press). The nonce is an incrementing unsigned integer that makes each
-datagram unique. The server should echo the session back in replies. Both
-are discarded by the server for protocol logic — the server tracks liveness
-via "did anything arrive recently."
-
-**OFF (PTT released):**
-```
-HKMSTR:OFF <session> <nonce>\n
-```
-Sent once when PTT is released. This is the fast-stop signal. If lost, the
-150ms timeout acts as a backstop.
-
-### Server → Client
-
-The server echoes the session ID from the NOPs in all replies. The client
-drops any reply with a stale session ID.
-
-**Partial segment (completed VAD segment):**
-```
-HKMSTR:P <session> <text>\n
-```
-A completed, VAD-separated utterance segment. The client starts TTS
-synthesis immediately and buffers the audio output, but does **not** play
-it yet. Playback starts when `:F` arrives (or timeout).
-
-**Final (all done):**
-```
-HKMSTR:F <session> <text>\n
-```
-Signals that all segments have been sent. May be empty
-(`HKMSTR:F <session>\n`). Triggers playback of all buffered audio on the
-client. If `<text>` is non-empty, the client synthesizes it before playing.
-
-The purpose of this design is to minimize latency: TTS synthesis runs in
-parallel with recording, so by the time `:F` arrives, audio is already
-buffered and playback starts immediately.
-
-## Server state machine
-
-```
-                    ┌──────────────────────────────────────────┐
-                    │                                          │
-                    ▼                                          │
-              ┌──────────┐  first NOP   ┌──────────────┐       │
-              │  IDLE    │ ──────────► │  RECORDING   │       │
-              └──────────┘             └──────────────┘       │
-                                           │      │           │
-                                   OFF     │      │ 150ms     │
-                                   recv'd  │      │ silence   │
-                                           ▼      ▼           │
-                                       ┌─────────────┐        │
-                                       │  PROCESSING │        │
-                                       └─────────────┘        │
-                                            │                 │
-                                     send   │                 │
-                                     :P/:F  │                 │
-                                            ▼                 │
-                                       back to IDLE ──────────┘
-```
-
-- **IDLE → RECORDING:** first NOP received, start mic capture
-- **RECORDING:** VAD detects completed segments → send `HKMSTR:P <text>`
-- **RECORDING → PROCESSING:** OFF received, OR 150ms since last NOP
-- **PROCESSING → IDLE:** send remaining segments as `:P`, then `HKMSTR:F`
-
-## Client playback model
-
-1. `:P` arrives → start TTS synthesis immediately, buffer audio (don't play).
-   Reset 250ms segment timer.
-2. More `:P` arrive → keep synthesizing and buffering, reset timer each time.
-3. `:F` arrives → play all buffered audio immediately, cancel timer.
-4. If `:F` doesn't arrive within 250ms of the last `:P` → play buffered audio
-   early. If more `:P` arrive after early playback, synthesis continues and
-   new audio is appended to the output — not a failure.
-5. `:F` may be empty — it just signals "all segments sent, start/confirm playback."
-
-## Timing
-
-| Parameter | Value | Purpose |
-|-----------|-------|---------|
-| NOP interval | 50ms | Heartbeat frequency while PTT held |
-| Silence timeout | 150ms | Stop recording if no NOPs (3 missed = lost OFF) |
-| NOP bandwidth | ~20 msg/s × ~20 bytes | ~400 bytes/s — negligible |
-
-## Why this works
-
-1. **Outbound broadcast `:68→:67` to `255.255.255.255`** passes the
-   WireGuard WFP killswitch (DHCP exception matches this exact pattern)
-2. **Inbound `:67→:68`** has no address restriction in the WFP rule, so
-   unicast replies pass through
-3. **Binding to a specific interface IP** (not `0.0.0.0`) wins unicast
-   delivery over the Windows DHCP client service
-4. **NOP spam** ensures the ON message gets through even at 5% packet loss
-   (3 consecutive NOPs = ~0.01% drop probability)
-5. **150ms timeout** is the backstop for lost OFF — at 50ms intervals, 3
-   consecutive NOPs must all be lost to false-stop

rvsttd/Cargo.toml

@@ -0,0 +1,20 @@
+[package]
+name = "rvsttd"
+version = "0.1.0"
+edition = "2021"
+
+[[bin]]
+name = "rvsttd"
+path = "src/main.rs"
+
+[dependencies]
+anyhow = "1"
+cpal = "0.15"
+serde = { version = "1", features = ["derive"] }
+serde_json = "1"
+
+[build-dependencies]
+bindgen = "0.71"
+
+[profile.release]
+opt-level = 3

rvsttd/build.rs

@@ -0,0 +1,28 @@
+use std::path::PathBuf;
+
+fn main() {
+    let home = std::env::var("HOME").unwrap_or_else(|_| "/root".to_string());
+    let rvsttd_dir = PathBuf::from(&home).join(".rvsttd");
+    let lib_dir = rvsttd_dir.join("lib");
+    let include_dir = rvsttd_dir.join("include");
+    let header = include_dir.join("moonshine-c-api.h");
+
+    println!("cargo:rerun-if-changed={}", header.display());
+    println!("cargo:rustc-link-search=native={}", lib_dir.display());
+    println!("cargo:rustc-link-lib=dylib=moonshine");
+    println!("cargo:rustc-link-arg=-Wl,-rpath,{}", lib_dir.display());
+
+    let bindings = bindgen::Builder::default()
+        .header(header.to_str().unwrap())
+        .allowlist_function("moonshine_.*")
+        .allowlist_var("MOONSHINE_.*")
+        .allowlist_type("transcript.*|moonshine_option_t|speaker_span_t|transcript_word_t")
+        .derive_default(true)
+        .generate()
+        .expect("Unable to generate moonshine bindings");
+
+    let out_path = PathBuf::from(std::env::var("OUT_DIR").unwrap());
+    bindings
+        .write_to_file(out_path.join("moonshine_bindings.rs"))
+        .expect("Couldn't write bindings");
+}

rvsttd/setup.sh

@@ -0,0 +1,102 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+RVSTTD_DIR="${HOME}/.rvsttd"
+LIB_DIR="${RVSTTD_DIR}/lib"
+INCLUDE_DIR="${RVSTTD_DIR}/include"
+BUILD_DIR="${RVSTTD_DIR}/build"
+
+ARCH="linux-x86_64"
+
+# Fetch latest release tag from GitHub API
+VERSION=$(curl -s https://api.github.com/repos/moonshine-ai/moonshine/releases/latest | grep '"tag_name"' | sed -E 's/.*"([^"]+)".*/\1/')
+if [ -z "${VERSION}" ]; then
+    echo "ERROR: could not fetch latest release tag from GitHub API"
+    exit 1
+fi
+
+PREBUILT_URL="https://github.com/moonshine-ai/moonshine/releases/download/${VERSION}/moonshine-voice-${ARCH}.tar.gz"
+SOURCE_URL="https://github.com/moonshine-ai/moonshine/archive/refs/tags/${VERSION}.tar.gz"
+
+echo "=== rvsttd setup ==="
+echo "Latest Moonshine release: ${VERSION}"
+echo "Target: ${RVSTTD_DIR}"
+echo ""
+
+mkdir -p "${LIB_DIR}" "${INCLUDE_DIR}" "${BUILD_DIR}"
+
+# Step 1: Download prebuilt package (for libonnxruntime.so.1 + header)
+PREBUILT_TGZ="${BUILD_DIR}/moonshine-voice-${ARCH}.tar.gz"
+PREBUILT_EXTRACTED="${BUILD_DIR}/moonshine-voice-${ARCH}"
+
+if [ ! -f "${LIB_DIR}/libonnxruntime.so.1" ]; then
+    echo ">>> Downloading prebuilt package (for libonnxruntime.so.1)..."
+    curl -L -o "${PREBUILT_TGZ}" "${PREBUILT_URL}"
+    mkdir -p "${PREBUILT_EXTRACTED}"
+    tar xzf "${PREBUILT_TGZ}" -C "${PREBUILT_EXTRACTED}" --strip-components=1
+
+    # Copy ONNX Runtime (prebuilt is fine — it has no glibc issue)
+    cp "${PREBUILT_EXTRACTED}/lib/libonnxruntime.so.1" "${LIB_DIR}/"
+    echo "    Installed libonnxruntime.so.1"
+
+    # Copy the header (it's the same in source and prebuilt)
+    cp "${PREBUILT_EXTRACTED}/include/moonshine-c-api.h" "${INCLUDE_DIR}/"
+    echo "    Installed moonshine-c-api.h"
+else
+    echo ">>> libonnxruntime.so.1 already present, skipping prebuilt download"
+fi
+
+# Step 2: Download source
+SOURCE_TGZ="${BUILD_DIR}/moonshine-source.tar.gz"
+SOURCE_DIR="${BUILD_DIR}/moonshine-source"
+
+if [ ! -d "${SOURCE_DIR}" ]; then
+    echo ">>> Downloading Moonshine source ${VERSION}..."
+    curl -L -o "${SOURCE_TGZ}" "${SOURCE_URL}"
+    mkdir -p "${SOURCE_DIR}"
+    tar xzf "${SOURCE_TGZ}" -C "${SOURCE_DIR}" --strip-components=1
+fi
+
+# Step 3: Build libmoonshine.so from source
+CMAKE_BUILD="${BUILD_DIR}/cmake-build"
+
+if [ ! -f "${LIB_DIR}/libmoonshine.so" ]; then
+    echo ">>> Building libmoonshine.so from source..."
+
+    # Point CMake at the prebuilt ONNX Runtime
+    ORT_LIB_DIR="${LIB_DIR}"
+    ORT_INCLUDE_DIR="${SOURCE_DIR}/core/third-party/onnxruntime/include"
+
+    # Patch CMake minimum version for older distros (Debian 11 ships 3.18)
+    find "${SOURCE_DIR}" -name CMakeLists.txt -exec sed -i 's/cmake_minimum_required(VERSION 3\.22\.1)/cmake_minimum_required(VERSION 3.18.4)/' {} +
+
+    # Remove -Werror (fails on third-party headers with older compilers)
+    sed -i 's/-Werror//' "${SOURCE_DIR}/core/CMakeLists.txt"
+
+    mkdir -p "${CMAKE_BUILD}"
+    cd "${CMAKE_BUILD}"
+
+    cmake "${SOURCE_DIR}/core" \
+        -DCMAKE_BUILD_TYPE=Release \
+        -DONNXRUNTIME_LIB_PATH="${ORT_LIB_DIR}/libonnxruntime.so.1" \
+        -DMOONSHINE_TTS_BUILD_ONNX=ON \
+        -DCMAKE_CXX_FLAGS="-I${SOURCE_DIR}/core/moonshine-tts/src"
+
+    make -j"$(nproc)" moonshine
+
+    cp "${CMAKE_BUILD}/libmoonshine.so" "${LIB_DIR}/"
+    cp "${SOURCE_DIR}/core/moonshine-c-api.h" "${INCLUDE_DIR}/"
+    echo "    Installed libmoonshine.so (built from source)"
+else
+    echo ">>> libmoonshine.so already present, skipping build"
+fi
+
+echo ""
+echo "=== Setup complete ==="
+echo "Library:  ${LIB_DIR}/libmoonshine.so"
+echo "Library:  ${LIB_DIR}/libonnxruntime.so.1"
+echo "Header:   ${INCLUDE_DIR}/moonshine-c-api.h"
+echo ""
+echo "Next: cd rvsttd && cargo build --release"
+echo "Then:  ./target/release/rvsttd fetch    # downloads the model"
+echo "Then:  ./target/release/rvsttd          # starts the server"

rvsttd/src/main.rs

@@ -0,0 +1,782 @@
+use anyhow::{anyhow, Result};
+use cpal::traits::{DeviceTrait, HostTrait, StreamTrait};
+use cpal::{SampleFormat, SampleRate};
+use serde::Deserialize;
+use std::collections::HashSet;
+use std::ffi::CStr;
+use std::io::{BufRead, BufReader, Write};
+use std::net::{TcpListener, TcpStream};
+use std::path::PathBuf;
+use std::sync::atomic::{AtomicBool, Ordering};
+use std::sync::mpsc;
+use std::sync::{Arc, Mutex};
+use std::thread;
+use std::time::{Duration, SystemTime, UNIX_EPOCH};
+
+include!(concat!(env!("OUT_DIR"), "/moonshine_bindings.rs"));
+
+const SAMPLE_RATE: i32 = 16000;
+const HEADER_VERSION: i32 = 30000;
+const ARCH: u32 = 5; // MOONSHINE_MODEL_ARCH_MEDIUM_STREAMING
+const BIND_ADDR: &str = "127.0.0.1:6996";
+const MAX_TEXT_BYTES: usize = 1380;
+
+// ─── helpers ──────────────────────────────────────────────────────────────
+
+fn ts() -> String {
+    let now = SystemTime::now()
+        .duration_since(UNIX_EPOCH)
+        .unwrap_or_default();
+    let secs = now.as_secs() % 86400;
+    let h = secs / 3600;
+    let m = (secs % 3600) / 60;
+    let s = secs % 60;
+    let ms = now.subsec_millis();
+    format!("{:02}:{:02}:{:02}.{:03}", h, m, s, ms)
+}
+
+fn log(msg: &str) {
+    eprintln!("[{}] {}", ts(), msg);
+}
+
+fn err_str(code: i32) -> String {
+    unsafe {
+        let s = moonshine_error_to_string(code);
+        if s.is_null() {
+            format!("error {}", code)
+        } else {
+            CStr::from_ptr(s).to_string_lossy().into_owned()
+        }
+    }
+}
+
+fn truncate_to_word(text: &str, max_bytes: usize) -> &str {
+    if text.len() <= max_bytes {
+        return text;
+    }
+    let cut = &text[..max_bytes.min(text.len())];
+    match cut.rfind(' ') {
+        Some(pos) => &text[..pos],
+        None => cut,
+    }
+}
+
+fn line_text(line: &transcript_line_t) -> String {
+    if line.text.is_null() {
+        return String::new();
+    }
+    unsafe { CStr::from_ptr(line.text) }
+        .to_string_lossy()
+        .into_owned()
+}
+
+fn unix_now() -> u64 {
+    SystemTime::now()
+        .duration_since(UNIX_EPOCH)
+        .unwrap_or_default()
+        .as_secs()
+}
+
+// ─── debug session recorder ──────────────────────────────────────────────
+
+struct DebugRecorder {
+    dir: PathBuf,
+    audio: Vec<f32>,
+    log_lines: Vec<String>,
+}
+
+impl DebugRecorder {
+    fn new(base_dir: &PathBuf) -> Self {
+        let dir = base_dir.join(unix_now().to_string());
+        std::fs::create_dir_all(&dir).ok();
+        Self {
+            dir,
+            audio: Vec::new(),
+            log_lines: Vec::new(),
+        }
+    }
+
+    fn log_event(&mut self, event: &str) {
+        self.log_lines.push(format!("[{}] {}", ts(), event));
+    }
+
+    fn log_segment(&mut self, line: &transcript_line_t, prefix: &str) {
+        let text = line_text(line);
+        let event = format!(
+            "SEGMENT id={} prefix={} is_complete={} start={:.3}s duration={:.3}s text=\"{}\"",
+            line.id, prefix, line.is_complete, line.start_time, line.duration, text
+        );
+        self.log_event(&event);
+    }
+
+    fn add_audio(&mut self, samples: &[f32]) {
+        self.audio.extend_from_slice(samples);
+    }
+
+    fn save(self) {
+        // Write log
+        let log_path = self.dir.join("session.log");
+        match std::fs::File::create(&log_path) {
+            Ok(mut f) => {
+                for line in &self.log_lines {
+                    writeln!(f, "{}", line).ok();
+                }
+            }
+            Err(e) => log(&format!("debug: failed to write log: {}", e)),
+        }
+
+        // Write WAV
+        let wav_path = self.dir.join("audio.wav");
+        match write_wav(&wav_path, &self.audio, SAMPLE_RATE as u32) {
+            Ok(()) => {
+                let secs = self.audio.len() as f64 / SAMPLE_RATE as f64;
+                log(&format!("debug: saved {} ({:.1}s, {} samples)", self.dir.display(), secs, self.audio.len()));
+            }
+            Err(e) => log(&format!("debug: failed to write wav: {}", e)),
+        }
+    }
+}
+
+fn write_wav(path: &PathBuf, samples: &[f32], sample_rate: u32) -> Result<()> {
+    let num_samples = samples.len() as u32;
+    let data_size = num_samples * 4; // f32 = 4 bytes
+    let file_size = 44 + data_size;
+
+    let mut f = std::fs::File::create(path)?;
+    // RIFF header
+    f.write_all(b"RIFF")?;
+    f.write_all(&(file_size - 8).to_le_bytes())?;
+    f.write_all(b"WAVE")?;
+    // fmt chunk
+    f.write_all(b"fmt ")?;
+    f.write_all(&16u32.to_le_bytes())?;       // chunk size
+    f.write_all(&3u16.to_le_bytes())?;         // IEEE float
+    f.write_all(&1u16.to_le_bytes())?;         // mono
+    f.write_all(&sample_rate.to_le_bytes())?;
+    f.write_all(&(sample_rate * 4).to_le_bytes())?; // byte rate
+    f.write_all(&4u16.to_le_bytes())?;         // block align
+    f.write_all(&32u16.to_le_bytes())?;        // bits per sample
+    // data chunk
+    f.write_all(b"data")?;
+    f.write_all(&data_size.to_le_bytes())?;
+
+    // Convert f32 samples to little-endian bytes
+    let mut bytes = Vec::with_capacity(data_size as usize);
+    for &s in samples {
+        bytes.extend_from_slice(&s.to_le_bytes());
+    }
+    f.write_all(&bytes)?;
+    Ok(())
+}
+
+// ─── shared state ─────────────────────────────────────────────────────────
+
+struct Shared {
+    writer: Mutex<TcpStream>,
+    session_id: u64,
+    transcriber_handle: i32,
+    debug_dir: Option<PathBuf>,
+}
+
+impl Shared {
+    fn send_msg(&self, prefix: &str, text: &str) {
+        let text = truncate_to_word(text, MAX_TEXT_BYTES);
+        let line = if text.is_empty() {
+            format!("{} {}\n", prefix, self.session_id)
+        } else {
+            format!("{} {} {}\n", prefix, self.session_id, text)
+        };
+        let mut writer = self.writer.lock().unwrap();
+        match writer.write_all(line.as_bytes()) {
+            Ok(_) => log(&format!("TX {} {} {}", prefix, self.session_id, text)),
+            Err(e) => log(&format!("TX failed: {}", e)),
+        }
+    }
+}
+
+// ─── session ──────────────────────────────────────────────────────────────
+
+struct Session {
+    shared: Arc<Shared>,
+    stop_signal: Arc<AtomicBool>,
+    aborted: Arc<AtomicBool>,
+    transcriber: thread::JoinHandle<()>,
+    cpal_stream: Option<cpal::Stream>,
+    stream_handle: i32,
+}
+
+impl Session {
+    fn stop(mut self) {
+        // Signal transcriber to exit main loop, then wait for it to drain
+        // trailing audio + final flush. cpal stream stays alive during drain.
+        self.stop_signal.store(true, Ordering::SeqCst);
+        self.transcriber.join().ok();
+        // Now safe to kill ALSA — transcriber is done
+        self.cpal_stream.take();
+        unsafe { moonshine_free_stream(self.shared.transcriber_handle, self.stream_handle) };
+    }
+
+    fn abort(mut self) {
+        self.aborted.store(true, Ordering::SeqCst);
+        self.stop_signal.store(true, Ordering::SeqCst);
+        self.transcriber.join().ok();
+        self.cpal_stream.take();
+        unsafe { moonshine_free_stream(self.shared.transcriber_handle, self.stream_handle) };
+    }
+}
+
+fn start_session(shared: Arc<Shared>) -> Option<Session> {
+    let stream_handle = unsafe { moonshine_create_stream(shared.transcriber_handle, 0) };
+    if stream_handle < 0 {
+        log(&format!("create_stream failed: {}", err_str(stream_handle)));
+        return None;
+    }
+    let rc = unsafe { moonshine_start_stream(shared.transcriber_handle, stream_handle) };
+    if rc != 0 {
+        log(&format!("start_stream failed: {}", err_str(rc)));
+        unsafe { moonshine_free_stream(shared.transcriber_handle, stream_handle) };
+        return None;
+    }
+
+    let (tx, rx) = mpsc::channel::<Vec<f32>>();
+    let stop_signal = Arc::new(AtomicBool::new(false));
+    let aborted = Arc::new(AtomicBool::new(false));
+
+    let cpal_stream = match start_cpal(tx) {
+        Ok(s) => Some(s),
+        Err(e) => {
+            log(&format!("cpal failed: {}", e));
+            unsafe { moonshine_free_stream(shared.transcriber_handle, stream_handle) };
+            return None;
+        }
+    };
+
+    let shared_clone = shared.clone();
+    let stop_signal_clone = stop_signal.clone();
+    let aborted_clone = aborted.clone();
+
+    let transcriber = thread::spawn(move || {
+        transcriber_loop(shared_clone, rx, stop_signal_clone, aborted_clone, stream_handle);
+    });
+
+    Some(Session {
+        shared,
+        stop_signal,
+        aborted,
+        transcriber,
+        cpal_stream,
+        stream_handle,
+    })
+}
+
+fn transcriber_loop(
+    shared: Arc<Shared>,
+    rx: mpsc::Receiver<Vec<f32>>,
+    stop_signal: Arc<AtomicBool>,
+    aborted: Arc<AtomicBool>,
+    stream_handle: i32,
+) {
+    let handle = shared.transcriber_handle;
+    let mut sent_ids: HashSet<u64> = HashSet::new();
+
+    // Debug recorder (if enabled)
+    let mut recorder = shared.debug_dir.as_ref().map(|_| DebugRecorder::new(
+        &shared.debug_dir.as_ref().unwrap().join(shared.session_id.to_string()),
+    ));
+
+    if let Some(ref mut r) = recorder {
+        r.log_event(&format!("session {} started", shared.session_id));
+    }
+
+    // Main loop: process audio until stop_signal
+    while !stop_signal.load(Ordering::SeqCst) {
+        match rx.recv_timeout(Duration::from_millis(100)) {
+            Ok(chunk) => {
+                if let Some(ref mut r) = recorder {
+                    r.add_audio(&chunk);
+                }
+
+                unsafe {
+                    moonshine_transcribe_add_audio_to_stream(
+                        handle, stream_handle,
+                        chunk.as_ptr(), chunk.len() as u64,
+                        SAMPLE_RATE, 0,
+                    );
+                }
+
+                let mut t_ptr: *mut transcript_t = std::ptr::null_mut();
+                let rc = unsafe { moonshine_transcribe_stream(handle, stream_handle, 0, &mut t_ptr) };
+                if rc != 0 || t_ptr.is_null() {
+                    continue;
+                }
+
+                if let Some(ref mut r) = recorder {
+                    log_transcript_lines(r, t_ptr);
+                }
+
+                send_new_segments(&shared, t_ptr, &mut sent_ids, "P");
+            }
+            Err(mpsc::RecvTimeoutError::Timeout) => continue,
+            Err(mpsc::RecvTimeoutError::Disconnected) => break,
+        }
+    }
+
+    // Drain trailing audio from ALSA buffer (cpal stream still alive).
+    // Fixed 100ms window — cpal delivers every ~50ms (800 samples @ 16kHz),
+    // so this captures 1-2 more callbacks worth of trailing audio.
+    // Hold one chunk back so we can apply a fade-out to the very last one.
+    let drain_deadline = std::time::Instant::now() + Duration::from_millis(100);
+    let mut held_chunk: Option<Vec<f32>> = None;
+    while std::time::Instant::now() < drain_deadline {
+        match rx.recv_timeout(drain_deadline - std::time::Instant::now()) {
+            Ok(chunk) => {
+                // Feed previously held chunk to Moonshine + recorder (no fade)
+                if let Some(prev) = held_chunk.take() {
+                    if let Some(ref mut r) = recorder {
+                        r.add_audio(&prev);
+                    }
+                    unsafe {
+                        moonshine_transcribe_add_audio_to_stream(
+                            handle, stream_handle,
+                            prev.as_ptr(), prev.len() as u64,
+                            SAMPLE_RATE, 0,
+                        );
+                    }
+                }
+                held_chunk = Some(chunk);
+            }
+            Err(mpsc::RecvTimeoutError::Timeout) => break,
+            Err(mpsc::RecvTimeoutError::Disconnected) => break,
+        }
+    }
+
+    // Apply 5ms fade-out to the last chunk, then feed to both Moonshine and recorder
+    if let Some(mut chunk) = held_chunk.take() {
+        let fade_samples = (SAMPLE_RATE as usize * 25) / 1000; // 25ms
+        if chunk.len() > fade_samples {
+            let start = chunk.len() - fade_samples;
+            for i in 0..fade_samples {
+                let t = 1.0 - (i as f32 / fade_samples as f32);
+                chunk[start + i] *= t;
+            }
+        }
+
+        if let Some(ref mut r) = recorder {
+            r.add_audio(&chunk);
+        }
+        unsafe {
+            moonshine_transcribe_add_audio_to_stream(
+                handle, stream_handle,
+                chunk.as_ptr(), chunk.len() as u64,
+                SAMPLE_RATE, 0,
+            );
+        }
+    }
+
+    // If aborted (new session took over), skip final flush entirely
+    if aborted.load(Ordering::SeqCst) {
+        unsafe { moonshine_stop_stream(handle, stream_handle) };
+        log(&format!("Session {} aborted, skipping final flush", shared.session_id));
+        if let Some(mut r) = recorder {
+            r.log_event("aborted (new session took over)");
+            r.save();
+        }
+        return;
+    }
+
+    // Final flush
+    unsafe { moonshine_stop_stream(handle, stream_handle) };
+    let mut t_ptr: *mut transcript_t = std::ptr::null_mut();
+    let rc = unsafe { moonshine_transcribe_stream(handle, stream_handle, 0, &mut t_ptr) };
+
+    if rc == 0 && !t_ptr.is_null() {
+        let t = unsafe { &*t_ptr };
+        let mut new_segments: Vec<String> = Vec::new();
+
+        if let Some(ref mut r) = recorder {
+            log_transcript_lines(r, t_ptr);
+        }
+
+        for i in 0..t.line_count as usize {
+            let line = unsafe { &*t.lines.add(i) };
+            if line.text.is_null() || line.is_complete == 0 {
+                continue;
+            }
+            if !sent_ids.insert(line.id) {
+                continue;
+            }
+            let text = line_text(line);
+            if !text.is_empty() {
+                new_segments.push(text);
+            }
+        }
+
+        if new_segments.is_empty() {
+            shared.send_msg("F", "");
+        } else {
+            let last = new_segments.len() - 1;
+            for (i, text) in new_segments.iter().enumerate() {
+                let prefix = if i == last { "F" } else { "P" };
+                shared.send_msg(prefix, text);
+                if let Some(ref mut r) = recorder {
+                    r.log_event(&format!("TX {} \"{}\"", prefix, text));
+                }
+            }
+        }
+    } else {
+        shared.send_msg("F", "");
+        if let Some(ref mut r) = recorder {
+            r.log_event("TX F (empty)");
+        }
+    }
+
+    if let Some(mut r) = recorder {
+        r.log_event("session ended");
+        r.save();
+    }
+}
+
+fn log_transcript_lines(recorder: &mut DebugRecorder, t_ptr: *const transcript_t) {
+    let t = unsafe { &*t_ptr };
+    for i in 0..t.line_count as usize {
+        let line = unsafe { &*t.lines.add(i) };
+        if line.text.is_null() {
+            continue;
+        }
+        recorder.log_segment(line, if line.is_complete != 0 { "complete" } else { "partial" });
+    }
+}
+
+fn send_new_segments(
+    shared: &Shared,
+    t_ptr: *const transcript_t,
+    sent_ids: &mut HashSet<u64>,
+    prefix: &str,
+) {
+    let t = unsafe { &*t_ptr };
+
+    for i in 0..t.line_count as usize {
+        let line = unsafe { &*t.lines.add(i) };
+        if line.text.is_null() || line.is_complete == 0 {
+            continue;
+        }
+        if !sent_ids.insert(line.id) {
+            continue;
+        }
+        let text = line_text(line);
+        if text.is_empty() {
+            continue;
+        }
+        shared.send_msg(prefix, &text);
+    }
+}
+
+// ─── cpal ─────────────────────────────────────────────────────────────────
+
+fn start_cpal(
+    tx: mpsc::Sender<Vec<f32>>,
+) -> Result<cpal::Stream> {
+    let host = cpal::default_host();
+    let dev = host
+        .default_input_device()
+        .ok_or_else(|| anyhow!("no input device"))?;
+
+    let supported = dev
+        .supported_input_configs()?
+        .filter(|c| c.channels() <= 2 && c.min_sample_rate().0 <= 16000)
+        .min_by_key(|c| match c.sample_format() {
+            SampleFormat::F32 => 0,
+            SampleFormat::I16 => 1,
+            SampleFormat::U8 => 2,
+            _ => 99,
+        })
+        .ok_or_else(|| anyhow!("no suitable input config"))?;
+
+    let fmt = supported.sample_format();
+    let mut config = supported.with_max_sample_rate().config();
+    if config.channels > 1 {
+        config.channels = 1;
+    }
+    config.sample_rate = SampleRate(16000);
+    config.buffer_size = cpal::BufferSize::Fixed(800);
+
+    let err_fn = |e: cpal::StreamError| log(&format!("cpal error: {}", e));
+
+    let stream = match fmt {
+        SampleFormat::F32 => dev.build_input_stream(
+            &config,
+            move |data: &[f32], _: &_| {
+                let _ = tx.send(data.to_vec());
+            },
+            err_fn,
+            None,
+        )?,
+        SampleFormat::I16 => dev.build_input_stream(
+            &config,
+            move |data: &[i16], _: &_| {
+                let _ = tx.send(data.iter().map(|&x| x as f32 / 32768.0).collect());
+            },
+            err_fn,
+            None,
+        )?,
+        SampleFormat::U8 => dev.build_input_stream(
+            &config,
+            move |data: &[u8], _: &_| {
+                let _ = tx.send(data.iter().map(|&x| (x as f32 - 128.0) / 128.0).collect());
+            },
+            err_fn,
+            None,
+        )?,
+        _ => return Err(anyhow!("unsupported sample format {:?}", fmt)),
+    };
+
+    stream.play()?;
+    Ok(stream)
+}
+
+// ─── model fetch ──────────────────────────────────────────────────────────
+
+#[derive(Deserialize)]
+struct Manifest {
+    groups: Vec<ManifestGroup>,
+}
+
+#[derive(Deserialize)]
+struct ManifestGroup {
+    #[allow(dead_code)]
+    base_url: String,
+    files: Vec<ManifestFile>,
+}
+
+#[derive(Deserialize)]
+struct ManifestFile {
+    name: String,
+    url: String,
+    size: Option<u64>,
+}
+
+fn fetch_model(model_dir: &str) -> Result<()> {
+    let dest = PathBuf::from(model_dir);
+    log(&format!("Fetching medium-streaming-en model to {}", dest.display()));
+
+    let lang = std::ffi::CString::new("en").unwrap();
+    let opt_name = std::ffi::CString::new("model_arch").unwrap();
+    let opt_value = std::ffi::CString::new("5").unwrap();
+    let mut options = [moonshine_option_t {
+        name: opt_name.as_ptr(),
+        value: opt_value.as_ptr(),
+    }];
+
+    let mut json_ptr: *mut i8 = std::ptr::null_mut();
+    let rc = unsafe {
+        moonshine_get_stt_dependencies(
+            lang.as_ptr(),
+            options.as_mut_ptr(),
+            options.len() as u64,
+            &mut json_ptr,
+        )
+    };
+
+    if rc != 0 || json_ptr.is_null() {
+        return Err(anyhow!("moonshine_get_stt_dependencies failed: {}", err_str(rc)));
+    }
+
+    let json_str = unsafe { CStr::from_ptr(json_ptr) }
+        .to_string_lossy()
+        .into_owned();
+    unsafe { moonshine_free_buffer(json_ptr as *mut std::ffi::c_void) };
+
+    let manifest: Manifest = serde_json::from_str(&json_str)?;
+
+    std::fs::create_dir_all(&dest)?;
+
+    let mut total_files = 0;
+    let mut total_bytes: u64 = 0;
+
+    for group in &manifest.groups {
+        for file in &group.files {
+            let dest_path = dest.join(&file.name);
+            if dest_path.exists() {
+                log(&format!("  SKIP {} (already exists)", file.name));
+                continue;
+            }
+
+            if let Some(parent) = dest_path.parent() {
+                std::fs::create_dir_all(parent)?;
+            }
+
+            log(&format!("  GET {}", file.url));
+
+            if let Some(expected) = file.size {
+                log(&format!("      {} bytes", expected));
+                total_bytes += expected;
+            }
+
+            let status = std::process::Command::new("curl")
+                .arg("-sSL")
+                .arg("-o")
+                .arg(&dest_path)
+                .arg(&file.url)
+                .status()?;
+
+            if !status.success() {
+                return Err(anyhow!("curl failed for {}", file.url));
+            }
+            total_files += 1;
+        }
+    }
+
+    log(&format!("Done: {} files, ~{} MB", total_files, total_bytes / (1024 * 1024)));
+    log(&format!("Model directory: {}", dest.display()));
+    Ok(())
+}
+
+// ─── main ─────────────────────────────────────────────────────────────────
+
+fn main() -> Result<()> {
+    let mut args = std::env::args().skip(1);
+
+    let home = std::env::var("HOME").unwrap_or_else(|_| "/root".to_string());
+    let default_model = format!("{}/.rvsttd/model", home);
+
+    let first = args.next();
+    if first.as_deref() == Some("fetch") {
+        let model_dir = args.next().unwrap_or_else(|| default_model.clone());
+        return fetch_model(&model_dir);
+    }
+
+    let mut args = first.into_iter().chain(args);
+    let mut model_dir = default_model;
+    let mut debug = false;
+
+    while let Some(a) = args.next() {
+        match a.as_str() {
+            "--model-dir" | "-m" => {
+                model_dir = args.next().unwrap_or(model_dir);
+            }
+            "--debug" => {
+                debug = true;
+            }
+            "--help" | "-h" => {
+                println!("Usage: rvsttd [--model-dir DIR] [--debug]");
+                println!("       rvsttd fetch [DIR]");
+                println!("Listens on TCP {}", BIND_ADDR);
+                println!("Model: medium-streaming (Moonshine)");
+                println!("--debug: save session audio + transcript log to ~/.rvsttd/debug/");
+                println!("'rvsttd fetch' downloads the English medium-streaming model");
+                return Ok(());
+            }
+            _ => return Err(anyhow!("unknown arg: {}", a)),
+        }
+    }
+
+    let debug_dir = if debug {
+        let d = PathBuf::from(&home).join(".rvsttd").join("debug");
+        std::fs::create_dir_all(&d).ok();
+        log(&format!("Debug mode enabled — sessions saved to {}", d.display()));
+        Some(d)
+    } else {
+        None
+    };
+
+    let model_path = std::fs::canonicalize(&model_dir)
+        .unwrap_or_else(|_| std::path::PathBuf::from(&model_dir));
+
+    log(&format!("Loading model from {}...", model_path.display()));
+    let c_dir = std::ffi::CString::new(model_path.to_str().unwrap()).unwrap();
+    let transcriber_handle = unsafe {
+        moonshine_load_transcriber_from_files(
+            c_dir.as_ptr(),
+            ARCH,
+            std::ptr::null(),
+            0,
+            HEADER_VERSION,
+        )
+    };
+    if transcriber_handle < 0 {
+        return Err(anyhow!("failed to load model: {}", err_str(transcriber_handle)));
+    }
+    log(&format!("Model loaded (handle {})", transcriber_handle));
+
+    let listener = TcpListener::bind(BIND_ADDR)?;
+    log(&format!("STT server listening on TCP {}", BIND_ADDR));
+
+    let mut current_session: Option<Session> = None;
+
+    for stream in listener.incoming() {
+        let stream = match stream {
+            Ok(s) => s,
+            Err(e) => {
+                log(&format!("accept failed: {}", e));
+                continue;
+            }
+        };
+
+        stream.set_nodelay(true).ok();
+        log(&format!("Client connected: {}", stream.peer_addr().map(|a| a.to_string()).unwrap_or_else(|_| "?".to_string())));
+
+        let writer_stream = stream.try_clone()?;
+        let reader = BufReader::new(stream);
+
+        for line in reader.lines() {
+            let line = match line {
+                Ok(l) => l,
+                Err(_) => break,
+            };
+
+            let line = line.trim();
+            log(&format!("RX {}", line));
+
+            // ON <session>
+            if let Some(rest) = line.strip_prefix("ON ") {
+                let new_session_id: u64 = rest.parse().unwrap_or(0);
+
+                if let Some(s) = current_session.take() {
+                    log(&format!("Aborting session {} for new session {}", s.shared.session_id, new_session_id));
+                    s.abort();
+                }
+
+                let shared = Arc::new(Shared {
+                    writer: Mutex::new(writer_stream.try_clone()?),
+                    session_id: new_session_id,
+                    transcriber_handle,
+                    debug_dir: debug_dir.clone(),
+                });
+
+                log(&format!("PTT on session {}", new_session_id));
+                match start_session(shared) {
+                    Some(s) => current_session = Some(s),
+                    None => log("Failed to start session"),
+                }
+            }
+            // OFF <session>
+            else if let Some(rest) = line.strip_prefix("OFF ") {
+                let off_session: u64 = rest.parse().unwrap_or(0);
+
+                if let Some(s) = current_session.as_ref() {
+                    if s.shared.session_id == off_session {
+                        log(&format!("OFF session {}", off_session));
+                        if let Some(s) = current_session.take() {
+                            s.stop();
+                        }
+                    } else {
+                        log(&format!("OFF session {} (stale, current={}), ignoring", off_session, s.shared.session_id));
+                    }
+                } else {
+                    log(&format!("OFF session {} (no active session), ignoring", off_session));
+                }
+            }
+            else {
+                log(&format!("Unknown command: {}", line));
+            }
+        }
+
+        log("Client disconnected");
+
+        if let Some(s) = current_session.take() {
+            s.abort();
+        }
+    }
+
+    Ok(())
+}