Fix audio clock drift and adaptive jitter buffering

clients/windows/VoiceCat.App/Audio/WasapiAudioBackend.cs

@@ -34,7 +34,7 @@ public sealed class WasapiAudioBackend : IAudioDeviceBackend
     private sealed class Playback : IAudioPlayback
     {
         private readonly string? deviceId;
-        private readonly PcmRing pcm = new(32768);
+        private readonly AdaptivePcmBuffer pcm = new(2);
         private readonly ManualResetEventSlim initialized = new(false);
         private readonly Thread thread;
         private volatile bool running = true;
@@ -47,6 +47,7 @@ public sealed class WasapiAudioBackend : IAudioDeviceBackend
             thread.Start();
             if (!initialized.Wait(5000) || !ready) { Dispose(); throw new InvalidOperationException("WASAPI playback could not start."); }
         }
+        public int BufferMilliseconds { get => pcm.BufferMilliseconds; set => pcm.BufferMilliseconds = value; }
         public void Write(ReadOnlySpan<short> stereoPcm) => pcm.TryWrite(stereoPcm);
         private unsafe void Work()
         {
@@ -68,7 +69,6 @@ public sealed class WasapiAudioBackend : IAudioDeviceBackend
                 render = (IAudioRenderClient)output;
                 if (client.Start() < 0) return;
                 ready = true; initialized.Set();
-                Span<short> discard = stackalloc short[1920];
                 while (running)
                 {
                     bufferReady.WaitOne(20); // Scheduling wait is outside the buffer-fill cycle.
@@ -77,8 +77,6 @@ public sealed class WasapiAudioBackend : IAudioDeviceBackend
                     if (frames == 0) continue;
                     if (render.GetBuffer(frames, out nint buffer) < 0) break;
                     var destination = new Span<short>((void*)buffer, checked((int)frames * 2));
-                    // Keep queued playback bounded to ~120 ms, then fill underflow with silence.
-                    while (pcm.Count > 11520) pcm.Read(discard);
                     int count = pcm.Read(destination); destination[count..].Clear();
                     if (render.ReleaseBuffer(frames, 0) < 0) break;
                 }

clients/windows/VoiceCat.App/Forms/AudioSettingsForm.cs

@@ -34,6 +34,7 @@ public sealed class AudioSettingsForm : Form
     private readonly bool _origAuxEnabled;
     private readonly string? _origAuxDeviceId;
     private readonly int _origAuxGain;
+    private readonly int _origAudioBufferMilliseconds;
 
     private readonly ComboBox _cboDevice;
     private readonly Button _btnRefresh;
@@ -54,6 +55,7 @@ public sealed class AudioSettingsForm : Form
     private readonly Button _btnAuxRefresh;
     private readonly Label _lblAuxGain;
     private readonly TrackBar _trkAuxGain;
+    private readonly ComboBox _cboBuffer;
 
     private Keys _pttKey;
 
@@ -80,6 +82,7 @@ public sealed class AudioSettingsForm : Form
         _origAuxEnabled = settings.AuxEnabled;
         _origAuxDeviceId = settings.AuxDeviceId;
         _origAuxGain = settings.AuxGain;
+        _origAudioBufferMilliseconds = settings.AudioBufferMilliseconds;
 
         Text = "Audio settings";
         FormBorderStyle = FormBorderStyle.FixedDialog;
@@ -87,7 +90,7 @@ public sealed class AudioSettingsForm : Form
         MinimizeBox = false;
         StartPosition = FormStartPosition.CenterParent;
         AutoScaleMode = AutoScaleMode.Font;
-        ClientSize = new Size(420, 639);
+        ClientSize = new Size(420, 690);
 
         // ── Device row ────────────────────────────────────────────────────────
         var lblDevice = new Label
@@ -319,19 +322,34 @@ public sealed class AudioSettingsForm : Form
             "Volume of the aux input stream. 100 is unity gain; range 0–400 percent.";
         _trkAuxGain.Scroll += TrkAuxGain_Scroll;
 
+        var lblBuffer = new Label { Text = "Audio &buffering:", Location = new Point(12, 590), AutoSize = true };
+        _cboBuffer = new ComboBox
+        {
+            Location = new Point(130, 586), Width = 190, DropDownStyle = ComboBoxStyle.DropDownList,
+            AccessibleName = "Audio buffering",
+            AccessibleDescription = "Local capture and playback buffering. Lower values reduce latency; higher values tolerate more scheduling jitter."
+        };
+        _cboBuffer.Items.AddRange(["Low latency (20 ms)", "Balanced (40 ms)", "Stable (60 ms)"]);
+        _cboBuffer.SelectedIndex = settings.AudioBufferMilliseconds switch { 20 => 0, 60 => 2, _ => 1 };
+        _cboBuffer.SelectedIndexChanged += (_, _) =>
+        {
+            _settings.AudioBufferMilliseconds = _cboBuffer.SelectedIndex switch { 0 => 20, 2 => 60, _ => 40 };
+            _client.SetAudioBufferMilliseconds(_settings.AudioBufferMilliseconds);
+        };
+
         // ── OK / Cancel ───────────────────────────────────────────────────────
         var btnOk = new Button
         {
             Text = "&OK",
             DialogResult = DialogResult.OK,
-            Location = new Point(228, 600),
+            Location = new Point(228, 642),
             Size = new Size(80, 27),
         };
         var btnCancel = new Button
         {
             Text = "&Cancel",
             DialogResult = DialogResult.Cancel,
-            Location = new Point(316, 600),
+            Location = new Point(316, 642),
             Size = new Size(80, 27),
         };
 
@@ -350,6 +368,7 @@ public sealed class AudioSettingsForm : Form
             _radioPtt, _lblPttKey, _btnChangePtt, _chkSystemWidePtt, _radioAlwaysOn,
             lblGain, _trkGain, _chkNoiseReduction, _chkStereoMic,
             _chkAux, _lblAuxDevice, _cboAuxDevice, _btnAuxRefresh, _lblAuxGain, _trkAuxGain,
+            lblBuffer, _cboBuffer,
             btnOk, btnCancel,
         ]);
 
@@ -534,6 +553,8 @@ public sealed class AudioSettingsForm : Form
         _settings.StereoMic = _origStereoMic;
         _settings.PttKey = (int)_origPttKey;
         _settings.SystemWidePtt = _origSystemWidePtt;
+        _settings.AudioBufferMilliseconds = _origAudioBufferMilliseconds;
+        _client.SetAudioBufferMilliseconds(_origAudioBufferMilliseconds);
 
         if (_micStreamId != 0)
         {

clients/windows/VoiceCat.App/Forms/MainForm.cs

@@ -109,8 +109,11 @@ public partial class MainForm : Form
         BootstrapFromServer();
     }
 
-    private void ApplyPersistedVoiceSettings() =>
+    private void ApplyPersistedVoiceSettings()
+    {
         _pttKey = (Keys)_voiceSettings.PttKey;
+        _client.SetAudioBufferMilliseconds(_voiceSettings.AudioBufferMilliseconds);
+    }
 
     // ── Startup ──────────────────────────────────────────────────────────────
 

clients/windows/VoiceCat.App/Models/VoiceSettings.cs

@@ -59,6 +59,9 @@ public sealed class VoiceSettings
     /// to the captured PCM before feeding (the core's input gain is mic-only and global).</summary>
     public int AuxGain { get; set; } = 100;
 
+    /// <summary>Local capture/playback safety buffer in milliseconds: 20, 40, or 60.</summary>
+    public int AudioBufferMilliseconds { get; set; } = 40;
+
     private static readonly JsonSerializerOptions JsonOptions = new() { WriteIndented = true };
 
     private static string AppDataDir => Path.Combine(
@@ -72,7 +75,9 @@ public sealed class VoiceSettings
         {
             if (!File.Exists(FilePath)) return new VoiceSettings();
             string json = File.ReadAllText(FilePath);
-            return JsonSerializer.Deserialize<VoiceSettings>(json) ?? new VoiceSettings();
+            VoiceSettings result = JsonSerializer.Deserialize<VoiceSettings>(json) ?? new VoiceSettings();
+            if (result.AudioBufferMilliseconds is not (20 or 40 or 60)) result.AudioBufferMilliseconds = 40;
+            return result;
         }
         catch
         {