One step closer to sending audio

examples/HelloWorld/src/Program.cs

@@ -3,6 +3,7 @@
 using System.Globalization;
 using System.IO;
 using System.Threading.Tasks;
+using DSharpPlus.VoiceLink.AudioCodecs;
 using DSharpPlus.VoiceLink.Enums;
 using Microsoft.Extensions.DependencyInjection;
 using Serilog;
@@ -62,14 +63,26 @@ public static async Task Main()
             clientBuilder.ConfigureLogging(builder => builder.AddSerilog());
 
             DiscordClient client = clientBuilder.Build();
-            VoiceLinkExtension voiceLinkExtension = client.UseVoiceLink();
+            VoiceLinkExtension voiceLinkExtension = client.UseVoiceLink(new()
+            {
+                AudioCodecFactory = (serviceProvider) => new OpusAudioCodec()
+            });
+
             voiceLinkExtension.UserSpeaking += async (sender, e) =>
             {
                 FileStream fileStream = File.Open($"test/{e.Member.Id}.pcm", FileMode.OpenOrCreate, FileAccess.ReadWrite, FileShare.ReadWrite);
                 await e.VoiceUser.AudioStream.CopyToAsync(fileStream);
                 await fileStream.DisposeAsync();
             };
 
+            voiceLinkExtension.ConnectionCreated += async (sender, e) =>
+            {
+                FileStream fileStream = File.Open("res/test.opus", FileMode.Open, FileAccess.Read, FileShare.Read);
+                _ = fileStream.CopyToAsync(e.Connection.AudioInput.AsStream(true));
+                await e.Connection.StartSpeakingAsync();
+                await fileStream.DisposeAsync();
+            };
+
             client.GuildDownloadCompleted += async (sender, e) =>
             {
                 if (!ulong.TryParse(Environment.GetEnvironmentVariable("DISCORD_GUILD"), out ulong guildId))

src/DSharpPlus.VoiceLink/AudioCodecs/IAudioCodec.cs

@@ -1,11 +1,13 @@
 using System;
+using System.Buffers;
 
 namespace DSharpPlus.VoiceLink.AudioCodecs
 {
     public delegate IAudioCodec AudioCodecFactory(IServiceProvider serviceProvider);
     public interface IAudioCodec
     {
         public int GetMaxBufferSize();
-        public int Decode(bool hasPacketLoss, ReadOnlySpan<byte> input, Span<byte> output);
+        public int EncodeOpus(ReadOnlySequence<byte> input, Span<byte> output);
+        public int DecodeOpus(bool hasPacketLoss, ReadOnlySpan<byte> input, Span<byte> output);
     }
 }

src/DSharpPlus.VoiceLink/AudioCodecs/OpusAudioCodec.cs

@@ -1,18 +1,33 @@
 using System;
+using System.Buffers;
 
 namespace DSharpPlus.VoiceLink.AudioCodecs
 {
     public class OpusAudioCodec : IAudioCodec
     {
-        private const int CHANNELS = 2;
-        private const int MAX_FRAME_SIZE = 5760;
-        private const int MAX_BUFFER_SIZE = MAX_FRAME_SIZE * 2 * CHANNELS;
+        public const int CHANNELS = 2;
+        public const int MAX_FRAME_SIZE = 5760;
+        public const int MAX_BUFFER_SIZE = MAX_FRAME_SIZE * 2 * CHANNELS;
+        public static readonly byte[] SilenceFrame = [0xF8, 0xFF, 0xFE];
 
         public int GetMaxBufferSize() => MAX_BUFFER_SIZE;
-        public int Decode(bool hasPacketLoss, ReadOnlySpan<byte> input, Span<byte> output)
+        public int DecodeOpus(bool hasPacketLoss, ReadOnlySpan<byte> input, Span<byte> output)
         {
+            if (hasPacketLoss)
+            {
+                SilenceFrame.AsSpan().CopyTo(output);
+                return SilenceFrame.Length;
+            }
+
             input.CopyTo(output);
             return input.Length;
         }
+
+        public int EncodeOpus(ReadOnlySequence<byte> input, Span<byte> output)
+        {
+            int frameSize = (int)Math.Min(input.Length, MAX_FRAME_SIZE);
+            input.Slice(0, frameSize).CopyTo(output);
+            return frameSize;
+        }
     }
 }

src/DSharpPlus.VoiceLink/AudioCodecs/Pcm16BitAudioCodec.cs

@@ -1,4 +1,6 @@
 using System;
+using System.Buffers;
+using System.Runtime.CompilerServices;
 using DSharpPlus.VoiceLink.Opus;
 
 namespace DSharpPlus.VoiceLink.AudioCodecs
@@ -17,23 +19,42 @@ public class Pcm16BitAudioCodec : IAudioCodec
         // 960 samples
         private const int FRAME_SIZE = (int)(SAMPLE_RATE * FRAME_DURATION);
 
-        // 20 milliseconds of audio data, 3840 bytes
+        // 20 milliseconds of audio data, 1920 bytes
         private const int SINGLE_CHANNEL_BUFFER_SIZE = FRAME_SIZE * BYTES_PER_SAMPLE;
 
         public int Channels { get; init; }
+        public int BufferSize { get; init; }
+        private OpusEncoder _opusEncoder { get; init; }
         private OpusDecoder _opusDecoder { get; init; }
 
         public Pcm16BitAudioCodec(int channels = 2)
         {
             Channels = channels;
+            BufferSize = SINGLE_CHANNEL_BUFFER_SIZE * Channels;
             _opusDecoder = OpusDecoder.Create(OpusSampleRate.Opus48000Hz, channels);
         }
 
         /// <inheritdoc/>
-        public int GetMaxBufferSize() => SINGLE_CHANNEL_BUFFER_SIZE * Channels;
+        public int GetMaxBufferSize() => BufferSize;
 
         /// <inheritdoc/>
-        public int Decode(bool hasPacketLoss, ReadOnlySpan<byte> input, Span<byte> output)
+        public int EncodeOpus(ReadOnlySequence<byte> input, Span<byte> output)
+        {
+            int sliceSize = Math.Min((int)input.Length, BufferSize);
+            byte[] buffer = ArrayPool<byte>.Shared.Rent(sliceSize);
+            try
+            {
+                input.Slice(0, sliceSize).CopyTo(buffer);
+                return _opusEncoder.Encode(Unsafe.As<byte[], short[]>(ref buffer), FRAME_SIZE, ref output);
+            }
+            finally
+            {
+                ArrayPool<byte>.Shared.Return(buffer);
+            }
+        }
+
+        /// <inheritdoc/>
+        public int DecodeOpus(bool hasPacketLoss, ReadOnlySpan<byte> input, Span<byte> output)
         {
             _opusDecoder.Decode(input, output, FRAME_SIZE, hasPacketLoss);
             return output.Length;

src/DSharpPlus.VoiceLink/Opus/OpusEncoder.cs

@@ -48,7 +48,6 @@ public unsafe int Encode(ReadOnlySpan<short> pcm, int frameSize, ref Span<byte>
             }
 
             // Trim the data to the encoded length
-            data = data[..encodedLength];
             return encodedLength;
         }
 
@@ -73,7 +72,6 @@ public unsafe int EncodeFloat(ReadOnlySpan<float> pcm, int frameSize, ref Span<b
             }
 
             // Trim the data to the encoded length
-            data = data[..encodedLength];
             return encodedLength;
         }
 

src/DSharpPlus.VoiceLink/Payloads/VoiceSpeakingPayload.cs

@@ -4,8 +4,9 @@ namespace DSharpPlus.VoiceLink.Payloads
 {
     public sealed record VoiceSpeakingPayload
     {
-        public required ulong UserId { get; init; }
+        public ulong UserId { get; init; }
         public required uint Ssrc { get; init; }
         public required VoiceSpeakingIndicators Speaking { get; init; }
+        public required int Delay { get; init; }
     }
 }

src/DSharpPlus.VoiceLink/Rtp/RtcpUtilities.cs

@@ -9,7 +9,7 @@ public static class RtcpUtilities
         /// </summary>
         /// <param name="source">The data to reference.</param>
         /// <returns>Whether the data contains a valid RTCP header.</returns>
-        public static bool IsRtcpReceiverReport(ReadOnlySpan<byte> source) => source.Length >= 8 && source[1] == 201;
+        public static bool HasRtcpReceiverReport(ReadOnlySpan<byte> source) => source.Length >= 8 && source[1] == 201;
 
         public static RtcpHeader DecodeHeader(ReadOnlySpan<byte> source)
         {

src/DSharpPlus.VoiceLink/Rtp/RtpUtilities.cs

@@ -19,7 +19,7 @@ public static class RtpUtilities
         /// </summary>
         /// <param name="source">The data to reference.</param>
         /// <returns>Whether the data contains a valid RTP header.</returns>
-        public static bool IsRtpHeader(ReadOnlySpan<byte> source) => source.Length >= 12 && (source[0] == Version || source[0] == VersionWithExtension) && source[1] == DiscordPayloadType;
+        public static bool HasRtpHeader(ReadOnlySpan<byte> source) => source.Length >= 12 && (source[0] == Version || source[0] == VersionWithExtension) && source[1] == DiscordPayloadType;
 
         /// <summary>
         /// Encodes a RTP header into the given buffer.
@@ -29,7 +29,7 @@ public static class RtpUtilities
         /// <param name="ssrc">The srrc of the audio frame.</param>
         /// <param name="target">Which buffer to write to.</param>
         /// <exception cref="ArgumentException">The target buffer must have a minimum of 12 bytes for the RTP header to fit.</exception>
-        public static void EncodeHeader(ushort sequence, uint timestamp, uint ssrc, Span<byte> target)
+        public static void EncodeHeader(RtpHeader header, Span<byte> target)
         {
             if (target.Length < 12)
             {
@@ -39,9 +39,9 @@ public static void EncodeHeader(ushort sequence, uint timestamp, uint ssrc, Span
             target.Clear();
             target[0] = Version;
             target[1] = DiscordPayloadType;
-            BinaryPrimitives.WriteUInt16BigEndian(target[2..4], sequence);
-            BinaryPrimitives.WriteUInt32BigEndian(target[4..8], timestamp);
-            BinaryPrimitives.WriteUInt32BigEndian(target[8..12], ssrc);
+            BinaryPrimitives.WriteUInt16BigEndian(target[2..4], header.Sequence);
+            BinaryPrimitives.WriteUInt32BigEndian(target[4..8], header.Timestamp);
+            BinaryPrimitives.WriteUInt32BigEndian(target[8..12], header.Ssrc);
         }
 
         /// <summary>

src/DSharpPlus.VoiceLink/VoiceLinkConfiguration.cs

@@ -1,3 +1,4 @@
+using System;
 using DSharpPlus.VoiceLink.AudioCodecs;
 using DSharpPlus.VoiceLink.VoiceEncryptionCiphers;
 
@@ -11,13 +12,18 @@ public sealed record VoiceLinkConfiguration
         public int MaxHeartbeatQueueSize { get; set; } = 5;
 
         /// <summary>
-        ///
+        /// Which voice encryption cipher to use for voice data encryption/decryption.
         /// </summary>
         public IVoiceEncryptionCipher VoiceEncryptionCipher { get; set; } = new XSalsa20Poly1305EncryptionCipher();
 
         /// <summary>
         /// A delegate which creates a new audio codec instance. The audio codec is responsible for encoding and decoding audio data into the user's desired format.
         /// </summary>
         public AudioCodecFactory AudioCodecFactory { get; set; } = _ => new Pcm16BitAudioCodec();
+
+        /// <summary>
+        /// When <see cref="VoiceLinkConnection.StartSpeakingAsync"/> should timeout after attempting to read <see cref="VoiceLinkConnection.AudioInput"/> for too long.
+        /// </summary>
+        public TimeSpan SpeakingTimeout { get; set; } = TimeSpan.FromMilliseconds(200);
     }
 }

src/DSharpPlus.VoiceLink/VoiceLinkConnection.VoiceGatewayHandlers.cs

@@ -61,8 +61,9 @@ private static async ValueTask ReadyAsync(VoiceLinkConnection connection, ReadRe
             VoiceReadyPayload voiceReadyPayload = connection._websocketPipe.Reader.Parse<VoiceGatewayDispatch<VoiceReadyPayload>>(result).Data;
 
             // Insert our SSRC code
-            connection._logger.LogDebug("Connection {GuildId}: Bot's SSRC code is {Ssrc}.", connection.Guild.Id, voiceReadyPayload.Ssrc);
+            connection._ssrc = voiceReadyPayload.Ssrc;
             connection._speakers.Add(voiceReadyPayload.Ssrc, new(connection, voiceReadyPayload.Ssrc, connection.Member, connection._audioDecoderFactory(connection.Extension.Client.ServiceProvider)));
+            connection._logger.LogDebug("Connection {GuildId}: Bot's SSRC code is {Ssrc}.", connection.Guild.Id, voiceReadyPayload.Ssrc);
 
             // Setup UDP while also doing ip discovery
             connection._logger.LogDebug("Connection {GuildId}: Setting up UDP, sending ip discovery...", connection.Guild.Id);