Sloppily implement audio decoders

src/DSharpPlus.VoiceLink/AudioDecoders/IAudioDecoder.cs

@@ -0,0 +1,11 @@
+using System;
+
+namespace DSharpPlus.VoiceLink.AudioDecoders
+{
+    public delegate IAudioDecoder AudioDecoderFactory(IServiceProvider serviceProvider);
+    public interface IAudioDecoder
+    {
+        public int GetMaxBufferSize();
+        public int Decode(bool hasPacketLoss, ReadOnlySpan<byte> input, Span<byte> output);
+    }
+}

src/DSharpPlus.VoiceLink/AudioDecoders/OpusAudioDecoder.cs

@@ -0,0 +1,18 @@
+using System;
+
+namespace DSharpPlus.VoiceLink.AudioDecoders
+{
+    public class OpusAudioDecoder : IAudioDecoder
+    {
+        private const int CHANNELS = 2;
+        private const int MAX_FRAME_SIZE = 5760;
+        private const int MAX_BUFFER_SIZE = MAX_FRAME_SIZE * 2 * CHANNELS;
+
+        public int GetMaxBufferSize() => MAX_BUFFER_SIZE;
+        public int Decode(bool hasPacketLoss, ReadOnlySpan<byte> input, Span<byte> output)
+        {
+            input.CopyTo(output);
+            return input.Length;
+        }
+    }
+}

src/DSharpPlus.VoiceLink/AudioDecoders/Pcm16BitAudioDecoder.cs

@@ -0,0 +1,32 @@
+using System;
+using DSharpPlus.VoiceLink.Opus;
+
+namespace DSharpPlus.VoiceLink.AudioDecoders
+{
+    public class Pcm16BitAudioDecoder : IAudioDecoder
+    {
+        // 48 kHz
+        private const int SAMPLE_RATE = 48000;
+
+        // 20 milliseconds
+        private const double FRAME_DURATION = 0.020;
+
+        // 960 samples
+        private const int FRAME_SIZE = (int)(SAMPLE_RATE * FRAME_DURATION);
+
+        // Stereo audio + opus PCM units are 16 bits
+        private const int BUFFER_SIZE = FRAME_SIZE * 2 * sizeof(short);
+
+        /// <inheritdoc/>
+        public int GetMaxBufferSize() => BUFFER_SIZE;
+
+        private OpusDecoder _opusDecoder { get; init; } = OpusDecoder.Create(OpusSampleRate.Opus48000Hz, 2);
+
+        /// <inheritdoc/>
+        public int Decode(bool hasPacketLoss, ReadOnlySpan<byte> input, Span<byte> output)
+        {
+            _opusDecoder.Decode(input, output, FRAME_SIZE, hasPacketLoss);
+            return BUFFER_SIZE;
+        }
+    }
+}

src/DSharpPlus.VoiceLink/VoiceLinkConfiguration.cs

@@ -1,4 +1,5 @@
 using System;
+using DSharpPlus.VoiceLink.AudioDecoders;
 using DSharpPlus.VoiceLink.VoiceEncrypters;
 using Microsoft.Extensions.DependencyInjection;
 
@@ -11,5 +12,6 @@ public sealed record VoiceLinkConfiguration
         private IServiceProvider? _serviceProvider;
         public int MaxHeartbeatQueueSize { get; set; } = 5;
         public IVoiceEncrypter VoiceEncrypter { get; set; } = new XSalsa20Poly1305();
+        public AudioDecoderFactory AudioDecoderFactory { get; set; } = _ => new Pcm16BitAudioDecoder();
     }
 }

src/DSharpPlus.VoiceLink/VoiceLinkConnection.VoiceGatewayHandlers.cs

@@ -62,7 +62,7 @@ private static async ValueTask ReadyAsync(VoiceLinkConnection connection, ReadRe
 
             // Insert our SSRC code
             connection._logger.LogDebug("Connection {GuildId}: Bot's SSRC code is {Ssrc}.", connection.Guild.Id, voiceReadyPayload.Ssrc);
-            connection._speakers.Add(voiceReadyPayload.Ssrc, new(connection, voiceReadyPayload.Ssrc, connection.Member));
+            connection._speakers.Add(voiceReadyPayload.Ssrc, new(connection, voiceReadyPayload.Ssrc, connection.Member, connection._audioDecoderFactory(connection.Extension.Configuration.ServiceProvider)));
 
             // Setup UDP while also doing ip discovery
             connection._logger.LogDebug("Connection {GuildId}: Setting up UDP, sending ip discovery...", connection.Guild.Id);
@@ -174,7 +174,7 @@ private static async ValueTask SpeakingAsync(VoiceLinkConnection connection, Rea
             // When we receive the speaking payload, we update the user's member object.
             if (!connection._speakers.TryGetValue(voiceSpeakingPayload.Ssrc, out VoiceLinkUser? voiceLinkUser))
             {
-                voiceLinkUser = new(connection, voiceSpeakingPayload.Ssrc, await connection.Guild.GetMemberAsync(voiceSpeakingPayload.UserId));
+                voiceLinkUser = new(connection, voiceSpeakingPayload.Ssrc, await connection.Guild.GetMemberAsync(voiceSpeakingPayload.UserId), connection._audioDecoderFactory(connection.Extension.Configuration.ServiceProvider));
                 connection._speakers.TryAdd(voiceSpeakingPayload.Ssrc, voiceLinkUser);
             }
             else

src/DSharpPlus.VoiceLink/VoiceLinkConnection.cs

@@ -14,6 +14,7 @@
 using DSharpPlus.Entities;
 using DSharpPlus.EventArgs;
 using DSharpPlus.Net.Abstractions;
+using DSharpPlus.VoiceLink.AudioDecoders;
 using DSharpPlus.VoiceLink.Commands;
 using DSharpPlus.VoiceLink.Enums;
 using DSharpPlus.VoiceLink.Payloads;
@@ -42,6 +43,7 @@ public sealed partial class VoiceLinkConnection
         private CancellationTokenSource _cancellationTokenSource { get; init; } = new();
         private Dictionary<uint, VoiceLinkUser> _speakers { get; init; } = [];
         private IVoiceEncrypter _voiceEncrypter { get; init; }
+        private AudioDecoderFactory _audioDecoderFactory { get; init; }
         private byte[] _secretKey { get; set; } = [];
         private Pipe _audioPipe { get; init; } = new();
 
@@ -62,6 +64,7 @@ public VoiceLinkConnection(VoiceLinkExtension extension, DiscordChannel channel,
             Channel = channel;
             _logger = extension.Configuration.ServiceProvider.GetRequiredService<ILogger<VoiceLinkConnection>>();
             _voiceEncrypter = extension.Configuration.VoiceEncrypter;
+            _audioDecoderFactory = extension.Configuration.AudioDecoderFactory;
         }
 
         public async ValueTask DisconnectAsync()
@@ -373,7 +376,7 @@ private void HandleRtcpReceiverReportPacket(byte[] buffer)
                 // We're explicitly passing a null member, however the dev should never expect this to
                 // be null as the speaking event should always fire once we receive both the user and the ssrc.
                 // TL;DR, this is to ensure we never lose any audio data.
-                voiceLinkUser = new(this, rtpHeader.Ssrc, null!);
+                voiceLinkUser = new(this, rtpHeader.Ssrc, null!, _audioDecoderFactory(Extension.Configuration.ServiceProvider), rtpHeader.Sequence);
                 _speakers.Add(rtpHeader.Ssrc, voiceLinkUser);
             }
 
@@ -402,27 +405,23 @@ private void HandleRtcpReceiverReportPacket(byte[] buffer)
                 decryptedAudio = decryptedAudio[(4 + (4 * extensionLength))..];
             }
 
+            // TODO: Handle FEC (Forward Error Correction) aka packet loss.
+            // * https://tools.ietf.org/html/rfc5109
+            bool hasPacketLoss = voiceLinkUser.UpdateSequence(rtpHeader.Sequence);
+
             // Decode the audio
             try
             {
-                // Calculate the frame size and buffer size
-                const int sampleRate = 48000; // 48 kHz
-                const double frameDuration = 0.020; // 20 milliseconds
-                const int frameSize = (int)(sampleRate * frameDuration); // 960 samples
-                const int bufferSize = frameSize * 2 * sizeof(short); // Stereo audio + opus PCM units are 16 bits
-
-                // TODO: Handle FEC (Forward Error Correction) aka packet loss.
-                // * https://tools.ietf.org/html/rfc5109
-                bool hasPacketLoss = voiceLinkUser.UpdateSequence(rtpHeader.Sequence);
+                int maxBufferSize = voiceLinkUser.AudioDecoder.GetMaxBufferSize();
 
                 // Allocate the buffer for the PCM data
-                Span<byte> audioBuffer = voiceLinkUser._audioPipe.Writer.GetSpan(bufferSize);
+                Span<byte> audioBuffer = voiceLinkUser._audioPipe.Writer.GetSpan(maxBufferSize);
 
                 // Decode the Opus packet
-                voiceLinkUser._opusDecoder.Decode(decryptedAudio, audioBuffer, frameSize, hasPacketLoss);
+                int writtenBytes = voiceLinkUser.AudioDecoder.Decode(hasPacketLoss, decryptedAudio, audioBuffer);
 
                 // Write the audio to the pipe
-                voiceLinkUser._audioPipe.Writer.Advance(bufferSize);
+                voiceLinkUser._audioPipe.Writer.Advance(writtenBytes);
             }
             catch (Exception error)
             {

src/DSharpPlus.VoiceLink/VoiceLinkUser.cs

@@ -2,8 +2,8 @@
 using System.IO;
 using System.IO.Pipelines;
 using DSharpPlus.Entities;
+using DSharpPlus.VoiceLink.AudioDecoders;
 using DSharpPlus.VoiceLink.Enums;
-using DSharpPlus.VoiceLink.Opus;
 
 namespace DSharpPlus.VoiceLink
 {
@@ -13,18 +13,19 @@ public sealed record VoiceLinkUser
         public uint Ssrc { get; init; }
         public DiscordMember Member { get; internal set; }
         public VoiceSpeakingIndicators VoiceIndication { get; internal set; } = VoiceSpeakingIndicators.None;
+        public IAudioDecoder AudioDecoder { get; init; }
         public PipeReader AudioPipe => _audioPipe.Reader;
         public Stream AudioStream => _audioPipe.Reader.AsStream(true);
 
         internal Pipe _audioPipe { get; init; } = new();
-        internal OpusDecoder _opusDecoder { get; init; } = OpusDecoder.Create(OpusSampleRate.Opus48000Hz, 2);
         internal ushort _lastSequence;
 
-        public VoiceLinkUser(VoiceLinkConnection connection, uint ssrc, DiscordMember member, ushort sequence = 0)
+        public VoiceLinkUser(VoiceLinkConnection connection, uint ssrc, DiscordMember member, IAudioDecoder audioDecoder, ushort sequence = 0)
         {
             Connection = connection ?? throw new ArgumentNullException(nameof(connection));
             Ssrc = ssrc;
             Member = member;
+            AudioDecoder = audioDecoder ?? throw new ArgumentNullException(nameof(audioDecoder));
             _lastSequence = sequence;
         }