Fix audio clock drift and adaptive jitter buffering

clients/apple/dotnet/VoiceCat.Mac/MacAudioBackend.cs

@@ -139,7 +139,7 @@ internal sealed class MacAudioBackend : IAudioDeviceBackend
 
     private sealed class Playback : IAudioPlayback
     {
-        private readonly PcmRing pcm = new(32_768);
+        private readonly AdaptivePcmBuffer pcm = new(2);
         private readonly short[] renderScratch = new short[8_192];
         private readonly AVAudioEngine engine = new();
         private readonly AVAudioFormat format;
@@ -187,6 +187,7 @@ internal sealed class MacAudioBackend : IAudioDeviceBackend
         }
 
         public void Write(ReadOnlySpan<short> stereoPcm) => pcm.TryWrite(stereoPcm);
+        public int BufferMilliseconds { get => pcm.BufferMilliseconds; set => pcm.BufferMilliseconds = value; }
 
         private unsafe int Render(IntPtr isSilence, IntPtr _, uint frameCount, IntPtr outputData)
         {
@@ -199,8 +200,6 @@ internal sealed class MacAudioBackend : IAudioDeviceBackend
                 int frames = checked((int)frameCount);
                 int requested = checked(frames * 2);
                 if (bufferCount != 2 || requested > renderScratch.Length) return Fail("Core Audio returned an invalid planar stereo playback buffer.");
-                Span<short> discard = stackalloc short[1_920];
-                while (pcm.Count > 11_520) pcm.Read(discard[..Math.Min(discard.Length, pcm.Count - 11_520)]);
                 Span<short> source = renderScratch.AsSpan(0, requested);
                 int read = pcm.Read(source);
                 source[read..].Clear();

clients/apple/dotnet/VoiceCat.Mac/MacSettings.cs

@@ -19,6 +19,7 @@ internal sealed class MacSettings
     internal bool AuxiliaryEnabled { get; set; }
     internal string? AuxiliaryDeviceId { get; set; }
     internal float AuxiliaryGain { get; set; } = 1f;
+    internal int AudioBufferMilliseconds { get; set; } = 40;
     internal bool EventSounds { get; set; } = true;
     internal bool SpokenEvents { get; set; }
     internal float EventVolume { get; set; } = 1f;
@@ -28,9 +29,9 @@ internal sealed class MacSettings
     internal void Load()
     {
         NSString[] keys = [(NSString)"voice.inputMode", (NSString)"voice.vadThreshold", (NSString)"voice.inputGain", (NSString)"voice.outputGain",
-            (NSString)"voice.pttKeyCode", (NSString)"voice.auxGain", (NSString)"feedback.sounds", (NSString)"feedback.volume"];
+            (NSString)"voice.pttKeyCode", (NSString)"voice.auxGain", (NSString)"voice.audioBufferMs", (NSString)"feedback.sounds", (NSString)"feedback.volume"];
         NSObject[] defaults = [NSNumber.FromInt32((int)AudioInputMode.VoiceActivation), NSNumber.FromFloat(0.05f),
-            NSNumber.FromFloat(1f), NSNumber.FromFloat(0.8f), NSNumber.FromInt32(0x60), NSNumber.FromFloat(1f),
+            NSNumber.FromFloat(1f), NSNumber.FromFloat(0.8f), NSNumber.FromInt32(0x60), NSNumber.FromFloat(1f), NSNumber.FromInt32(40),
             NSNumber.FromBoolean(true), NSNumber.FromFloat(1f)];
         values.RegisterDefaults(new NSDictionary<NSString, NSObject>(keys, defaults));
         InputMode = Enum.IsDefined(typeof(AudioInputMode), (int)values.IntForKey("voice.inputMode"))
@@ -46,6 +47,7 @@ internal sealed class MacSettings
         AuxiliaryEnabled = values.BoolForKey("voice.auxEnabled");
         AuxiliaryDeviceId = EmptyToNull(values.StringForKey("voice.auxDeviceUID"));
         AuxiliaryGain = Math.Clamp(values.FloatForKey("voice.auxGain"), 0f, 4f);
+        int audioBuffer = checked((int)values.IntForKey("voice.audioBufferMs")); AudioBufferMilliseconds = audioBuffer is 20 or 40 or 60 ? audioBuffer : 40;
         EventSounds = values.BoolForKey("feedback.sounds");
         SpokenEvents = values.BoolForKey("feedback.speech");
         EventVolume = Math.Clamp(values.FloatForKey("feedback.volume"), 0f, 1f);
@@ -59,7 +61,7 @@ internal sealed class MacSettings
         values.SetFloat(InputGain, "voice.inputGain"); values.SetFloat(OutputGain, "voice.outputGain");
         values.SetBool(InputNoiseReduction, "voice.inputNoiseReduction"); values.SetBool(StereoMicrophone, "voice.stereoMic");
         values.SetInt(PushToTalkKeyCode, "voice.pttKeyCode"); Set("voice.inputDevice", InputDeviceId); Set("voice.outputDevice", OutputDeviceId);
-        values.SetBool(AuxiliaryEnabled, "voice.auxEnabled"); Set("voice.auxDeviceUID", AuxiliaryDeviceId); values.SetFloat(AuxiliaryGain, "voice.auxGain");
+        values.SetBool(AuxiliaryEnabled, "voice.auxEnabled"); Set("voice.auxDeviceUID", AuxiliaryDeviceId); values.SetFloat(AuxiliaryGain, "voice.auxGain"); values.SetInt(AudioBufferMilliseconds, "voice.audioBufferMs");
         values.SetBool(EventSounds, "feedback.sounds"); values.SetBool(SpokenEvents, "feedback.speech"); values.SetFloat(EventVolume, "feedback.volume");
         values.SetBool(SelfTalkSounds, "feedback.selfTalk"); values.SetBool(PushToTalkSound, "feedback.ptt"); values.Synchronize();
     }

clients/apple/dotnet/VoiceCat.Mac/MainWindowController.cs

@@ -210,6 +210,8 @@ internal sealed class MainWindowController : NSWindowController
         if (settings.InputMode != AudioInputMode.PushToTalk) { client.Audio.PushToTalk = false; pushToTalkEngaged = false; }
         client.Audio.InputGain = settings.InputGain; client.Audio.OutputGain = settings.OutputGain;
         client.Audio.InputNoiseReduction = settings.InputNoiseReduction;
+        client.Audio.DeviceBufferMilliseconds = settings.AudioBufferMilliseconds;
+        if (playback is not null) playback.BufferMilliseconds = settings.AudioBufferMilliseconds;
         if (auxiliaryStreamId != 0) client.Audio.SetLocalGain(auxiliaryStreamId, settings.AuxiliaryGain);
     }
 
@@ -367,6 +369,7 @@ internal sealed class MainWindowController : NSWindowController
             activeOutputDevice = settings.OutputDeviceId ?? SelectedDevice(outputDevice);
             activeStereo = settings.StereoMicrophone;
             playback = audioBackend.OpenPlayback(activeOutputDevice);
+            playback.BufferMilliseconds = settings.AudioBufferMilliseconds;
             client.Audio.MixedPcm += PlayMixedPcm;
             StreamInfo stream = await client.StartStreamAsync(StreamKind.StreamMic, "Microphone", settings.StereoMicrophone ? 2 : 1);
             microphoneStreamId = stream.StreamId;

clients/apple/dotnet/VoiceCat.Mac/SettingsWindowController.cs

@@ -26,15 +26,18 @@ internal sealed class SettingsWindowController : NSWindowController
     private readonly NSButton pttSound = NSButton.CreateCheckbox("Push-to-talk cue", () => { });
     private readonly NSButton pttKey = new(new CGRect(220, 12, 200, 32)) { Title = "Change PTT key…" };
     private readonly NSSlider eventVolume = new(new CGRect(160, 505, 260, 24)) { MinValue = 0, MaxValue = 100 };
+    private readonly NSPopUpButton audioBuffer = new(new CGRect(160, 545, 260, 28), false);
 
     internal SettingsWindowController(MainWindowController owner, MacSettings settings, MacAudioBackend audio) : base(
-        new NSWindow(new CGRect(0, 0, 460, 620), NSWindowStyle.Titled | NSWindowStyle.Closable,
+        new NSWindow(new CGRect(0, 0, 460, 660), NSWindowStyle.Titled | NSWindowStyle.Closable,
             NSBackingStore.Buffered, false))
     {
         this.owner = owner; this.settings = settings; this.audio = audio;
         Window!.Title = "VoiceCat Settings"; Window.Center();
         NSView view = Window.ContentView!;
         mode.AddItems(["Voice activation", "Push to talk", "Always on"]);
+        audioBuffer.AddItems(["Low latency (20 ms)", "Balanced (40 ms)", "Stable (60 ms)"]);
+        Add(view, "Audio buffering", audioBuffer, 555);
         Add(view, "Sound volume", eventVolume, 515);
         Add(view, "Input mode", mode, 465); Add(view, "VAD sensitivity", vad, 425); Add(view, "Microphone volume", inputGain, 385);
         noiseReduction.Frame = new CGRect(160, 340, 260, 24); stereo.Frame = new CGRect(160, 315, 260, 24);
@@ -46,7 +49,7 @@ internal sealed class SettingsWindowController : NSWindowController
         selfTalk.Frame = new CGRect(265, 45, 180, 24); pttSound.Frame = new CGRect(25, 18, 180, 24);
         view.AddSubview(sounds); view.AddSubview(speech); view.AddSubview(selfTalk); view.AddSubview(pttSound); view.AddSubview(pttKey);
         foreach (NSControl control in new NSControl[] { mode, vad, inputGain, noiseReduction, stereo, input, output,
-            outputGain, auxiliary, auxiliaryDevice, auxiliaryGain, sounds, speech, selfTalk, pttSound, eventVolume }) control.Activated += Changed;
+            outputGain, auxiliary, auxiliaryDevice, auxiliaryGain, sounds, speech, selfTalk, pttSound, eventVolume, audioBuffer }) control.Activated += Changed;
         pttKey.Activated += CapturePushToTalkKey;
         Populate(input, true, settings.InputDeviceId); Populate(output, false, settings.OutputDeviceId); Populate(auxiliaryDevice, true, settings.AuxiliaryDeviceId);
         LoadValues(); SetAccessibility();
@@ -76,6 +79,7 @@ internal sealed class SettingsWindowController : NSWindowController
         eventVolume.DoubleValue = settings.EventVolume * 100;
         selfTalk.State = State(settings.SelfTalkSounds); pttSound.State = State(settings.PushToTalkSound);
         pttKey.Title = $"PTT key code: {settings.PushToTalkKeyCode}";
+        audioBuffer.SelectItem(settings.AudioBufferMilliseconds switch { 20 => 0, 60 => 2, _ => 1 });
         UpdateEnabled();
     }
 
@@ -102,6 +106,7 @@ internal sealed class SettingsWindowController : NSWindowController
         settings.AuxiliaryGain = (float)auxiliaryGain.DoubleValue / 100; settings.EventSounds = On(sounds);
         settings.EventVolume = (float)eventVolume.DoubleValue / 100; settings.SpokenEvents = On(speech);
         settings.SelfTalkSounds = On(selfTalk); settings.PushToTalkSound = On(pttSound);
+        settings.AudioBufferMilliseconds = audioBuffer.IndexOfSelectedItem switch { 0 => 20, 2 => 60, _ => 40 };
         settings.Save(); UpdateEnabled(); await owner.ApplySettingsAsync();
     }
 
@@ -113,6 +118,7 @@ internal sealed class SettingsWindowController : NSWindowController
         ((INSAccessibility)output).AccessibilityLabel = "Output device"; ((INSAccessibility)outputGain).AccessibilityLabel = "Output volume";
         ((INSAccessibility)auxiliaryDevice).AccessibilityLabel = "Auxiliary input device"; ((INSAccessibility)auxiliaryGain).AccessibilityLabel = "Auxiliary input volume";
         ((INSAccessibility)eventVolume).AccessibilityLabel = "Event sound volume";
+        ((INSAccessibility)audioBuffer).AccessibilityLabel = "Audio buffering";
     }
     private static bool On(NSButton button) => button.State == NSCellStateValue.On;
     private static NSCellStateValue State(bool value) => value ? NSCellStateValue.On : NSCellStateValue.Off;

clients/apple/dotnet/VoiceCat.iOS/AppModel.cs

@@ -266,6 +266,8 @@ internal sealed class AppModel
         owner.Audio.InputMode = settings.InputMode; owner.Audio.VadThreshold = settings.VadThreshold;
         owner.Audio.InputGain = settings.InputGain; owner.Audio.OutputGain = settings.OutputGain;
         owner.Audio.InputNoiseReduction = settings.InputNoiseReduction;
+        owner.Audio.DeviceBufferMilliseconds = settings.AudioBufferMilliseconds;
+        IosAudioEngine.Shared.BufferMilliseconds = settings.AudioBufferMilliseconds;
     }
 
     private void PollAudio()

clients/apple/dotnet/VoiceCat.iOS/IosAudioEngine.cs

@@ -8,7 +8,7 @@ namespace VoiceCat.iOS;
 internal sealed class IosAudioEngine
 {
     internal static IosAudioEngine Shared { get; } = new();
-    private readonly PcmRing playbackRing = new(131_072);
+    private readonly AdaptivePcmBuffer playbackRing = new(2, capacityFrames: 65_536);
     private readonly short[] renderScratch = new short[16_384];
     private AVAudioEngine? engine;
     private AVAudioSourceNode? source;
@@ -28,6 +28,7 @@ internal sealed class IosAudioEngine
     private bool inputProvided;
     private bool tapInstalled;
     internal bool IsConnected { get; private set; }
+    internal int BufferMilliseconds { get => playbackRing.BufferMilliseconds; set => playbackRing.BufferMilliseconds = value; }
 
     private IosAudioEngine() { microphoneWorker = PumpMicrophoneAsync(microphoneStop.Token); }
 

clients/apple/dotnet/VoiceCat.iOS/IosSettings.cs

@@ -17,13 +17,14 @@ internal sealed class IosSettings
     internal float EventVolume { get; set; } = 1f;
     internal bool SelfTalkSounds { get; set; }
     internal bool PushToTalkSound { get; set; }
+    internal int AudioBufferMilliseconds { get; set; } = 40;
 
     internal void Load()
     {
         NSString[] keys = [(NSString)"voice.inputMode", (NSString)"voice.vadThreshold", (NSString)"voice.inputGain",
-            (NSString)"voice.outputGain", (NSString)"feedback.sounds", (NSString)"feedback.volume"];
+            (NSString)"voice.outputGain", (NSString)"voice.audioBufferMs", (NSString)"feedback.sounds", (NSString)"feedback.volume"];
         NSObject[] defaults = [NSNumber.FromInt32((int)AudioInputMode.VoiceActivation), NSNumber.FromFloat(0.025f),
-            NSNumber.FromFloat(1f), NSNumber.FromFloat(1f), NSNumber.FromBoolean(true), NSNumber.FromFloat(1f)];
+            NSNumber.FromFloat(1f), NSNumber.FromFloat(1f), NSNumber.FromInt32(40), NSNumber.FromBoolean(true), NSNumber.FromFloat(1f)];
         values.RegisterDefaults(new NSDictionary<NSString, NSObject>(keys, defaults));
         int inputMode = checked((int)values.IntForKey("voice.inputMode"));
         InputMode = Enum.IsDefined(typeof(AudioInputMode), inputMode) ? (AudioInputMode)inputMode : AudioInputMode.VoiceActivation;
@@ -31,6 +32,7 @@ internal sealed class IosSettings
         InputGain = Math.Clamp(values.FloatForKey("voice.inputGain"), 0f, 4f);
         OutputGain = Math.Clamp(values.FloatForKey("voice.outputGain"), 0f, 1f);
         InputNoiseReduction = values.BoolForKey("voice.inputNoiseReduction");
+        int audioBuffer = checked((int)values.IntForKey("voice.audioBufferMs")); AudioBufferMilliseconds = audioBuffer is 20 or 40 or 60 ? audioBuffer : 40;
         EventSounds = values.BoolForKey("feedback.sounds");
         SpokenEvents = values.BoolForKey("feedback.speech");
         EventVolume = Math.Clamp(values.FloatForKey("feedback.volume"), 0f, 1f);
@@ -42,6 +44,7 @@ internal sealed class IosSettings
     {
         values.SetInt((int)InputMode, "voice.inputMode"); values.SetFloat(VadThreshold, "voice.vadThreshold");
         values.SetFloat(InputGain, "voice.inputGain"); values.SetFloat(OutputGain, "voice.outputGain");
+        values.SetInt(AudioBufferMilliseconds, "voice.audioBufferMs");
         values.SetBool(InputNoiseReduction, "voice.inputNoiseReduction"); values.SetBool(EventSounds, "feedback.sounds");
         values.SetBool(SpokenEvents, "feedback.speech"); values.SetFloat(EventVolume, "feedback.volume");
         values.SetBool(SelfTalkSounds, "feedback.selfTalk"); values.SetBool(PushToTalkSound, "feedback.ptt"); values.Synchronize();

clients/apple/dotnet/VoiceCat.iOS/SettingsController.cs

@@ -14,23 +14,23 @@ internal sealed class SettingsController : UITableViewController
     public override void ViewDidLoad() { base.ViewDidLoad(); TableView.RegisterClassForCellReuse(typeof(UITableViewCell), "setting"); }
     private void Reload() => TableView.ReloadData();
     public override nint NumberOfSections(UITableView tableView) => 5;
-    public override nint RowsInSection(UITableView tableView, nint section) => section switch { 0 => 4, 1 => 4, 2 => 5, 3 => model.Client?.Permissions is { } p && (p.IsAdmin || p.CanAdminAccounts) ? 1 : 0, _ => 2 };
+    public override nint RowsInSection(UITableView tableView, nint section) => section switch { 0 => 5, 1 => 4, 2 => 5, 3 => model.Client?.Permissions is { } p && (p.IsAdmin || p.CanAdminAccounts) ? 1 : 0, _ => 2 };
     public override string? TitleForHeader(UITableView tableView, nint section) => section switch { 0 => "Audio", 1 => "Voice", 2 => "Notifications", 3 => "Administration", _ => "Server" };
     public override UITableViewCell GetCell(UITableView tableView, NSIndexPath path)
     {
         UITableViewCell cell = tableView.DequeueReusableCell("setting", path); cell.AccessoryView = null; cell.Accessory = UITableViewCellAccessory.None;
         string title = path.Section switch
         {
-            0 => path.Row switch { 0 => "Audio preset", 1 => "Speaker output", 2 => "Advanced audio", _ => model.ScreenSharing ? "Stop sharing screen audio" : "Share screen audio" },
+            0 => path.Row switch { 0 => "Audio preset", 1 => "Speaker output", 2 => "Advanced audio", 3 => "Audio buffering", _ => model.ScreenSharing ? "Stop sharing screen audio" : "Share screen audio" },
             1 => path.Row switch { 0 => "Input mode", 1 => $"VAD threshold: {model.Settings.VadThreshold:0.000}", 2 => $"Microphone volume: {model.Settings.InputGain:P0}", _ => "Microphone noise reduction" },
             2 => path.Row switch { 0 => "Event sounds", 1 => $"Sound volume: {model.Settings.EventVolume:P0}", 2 => "Speak events", 3 => "Own voice activity sounds", _ => "Push-to-talk cue" },
             3 => "Manage accounts", _ => path.Row == 0 ? "Disconnect" : "VoiceCat 0.0.1"
         };
-        var content = cell.DefaultContentConfiguration; content.Text = title; content.SecondaryText = path.Section == 0 && path.Row == 0 ? IosAudioRouter.Shared.Preset.ToString() : path.Section == 1 && path.Row == 0 ? model.Settings.InputMode.ToString() : path.Section == 0 && path.Row == 3 && model.ScreenSharing ? "Sharing" : null; cell.ContentConfiguration = content; cell.AccessibilityLabel = title;
+        var content = cell.DefaultContentConfiguration; content.Text = title; content.SecondaryText = path.Section == 0 && path.Row == 0 ? IosAudioRouter.Shared.Preset.ToString() : path.Section == 1 && path.Row == 0 ? model.Settings.InputMode.ToString() : path.Section == 0 && path.Row == 3 ? $"{model.Settings.AudioBufferMilliseconds} ms" : path.Section == 0 && path.Row == 4 && model.ScreenSharing ? "Sharing" : null; cell.ContentConfiguration = content; cell.AccessibilityLabel = title + (content.SecondaryText is null ? "" : ", " + content.SecondaryText);
         if (path.Section == 0 && path.Row == 1) cell.AccessoryView = Toggle(IosAudioRouter.Shared.ForceSpeaker, "Speaker output", (_, _) => IosAudioRouter.Shared.SetForceSpeaker(((UISwitch)cell.AccessoryView!).On));
         else if (path.Section == 1 && path.Row == 3) cell.AccessoryView = Toggle(model.Settings.InputNoiseReduction, title, (_, _) => { model.Settings.InputNoiseReduction = ((UISwitch)cell.AccessoryView!).On; model.ApplyVoiceSettings(); });
         else if (path.Section == 2 && path.Row is 0 or 2 or 3 or 4) { bool value = path.Row switch { 0 => model.Settings.EventSounds, 2 => model.Settings.SpokenEvents, 3 => model.Settings.SelfTalkSounds, _ => model.Settings.PushToTalkSound }; int row = path.Row; cell.AccessoryView = Toggle(value, title, (_, _) => { bool on = ((UISwitch)cell.AccessoryView!).On; if (row == 0) model.Settings.EventSounds = on; else if (row == 2) model.Settings.SpokenEvents = on; else if (row == 3) model.Settings.SelfTalkSounds = on; else model.Settings.PushToTalkSound = on; model.Save(); }); }
-        else if (path.Section == 0 && path.Row == 2 || path.Section == 3) cell.Accessory = UITableViewCellAccessory.DisclosureIndicator;
+        else if (path.Section == 0 && path.Row is 2 or 3 || path.Section == 3) cell.Accessory = UITableViewCellAccessory.DisclosureIndicator;
         return cell;
     }
     public override async void RowSelected(UITableView tableView, NSIndexPath path)
@@ -39,7 +39,8 @@ internal sealed class SettingsController : UITableViewController
         {
             if (path.Section == 0 && path.Row == 0) Choice("Audio preset", Enum.GetValues<IosAudioPreset>().Select(value => value.ToString()).ToArray(), index => IosAudioRouter.Shared.SelectPreset(Enum.GetValues<IosAudioPreset>()[index]));
             else if (path.Section == 0 && path.Row == 2) NavigationController?.PushViewController(new AdvancedAudioController(), true);
-            else if (path.Section == 0 && path.Row == 3) ShowScreenSharing();
+            else if (path.Section == 0 && path.Row == 3) Choice("Audio buffering", ["Low latency (20 ms)", "Balanced (40 ms)", "Stable (60 ms)"], index => { model.Settings.AudioBufferMilliseconds = index switch { 0 => 20, 2 => 60, _ => 40 }; model.ApplyVoiceSettings(); });
+            else if (path.Section == 0 && path.Row == 4) ShowScreenSharing();
             else if (path.Section == 1 && path.Row == 0) Choice("Input mode", ["Voice activation", "Push to talk", "Always on"], index => { model.Settings.InputMode = (AudioInputMode)index; model.ApplyVoiceSettings(); });
             else if (path.Section == 1 && path.Row == 1) Slider("Voice activation threshold", 0.001f, 0.1f, model.Settings.VadThreshold, value => { model.Settings.VadThreshold = value; model.ApplyVoiceSettings(); });
             else if (path.Section == 1 && path.Row == 2) Slider("Microphone volume", 0, 4, model.Settings.InputGain, value => { model.Settings.InputGain = value; model.ApplyVoiceSettings(); });