Minecraft/MinecraftConsolesProximityChat
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

- improve voice chat smoothness

Browse Source 
- fixed the position of the speaker indicator. constant aspect ratio added
- fixed voice chat functionality to work with the latest commits
- add support for cross compiling windows64 builds on linux
This commit is contained in:
haxi0
2026-03-23 00:58:13 +03:00
parent 8be27511bb
commit 784242fcf7
1320 changed files with 361482 additions and 116 deletions
Download Patch File Download Diff File Expand all files Collapse all files
Minecraft.Client/VoiceChatManager.cpp
+103 -20
View File
@@ -21,11 +21,16 @@ namespace
static const int VOICE_ACTIVATION_OPEN_RMS = 220;
static const int VOICE_ACTIVATION_CLOSE_PEAK = 600;
static const int VOICE_ACTIVATION_CLOSE_RMS = 160;
static const int VOICE_ACTIVATION_HOLD_FRAMES = 10;
static const int VOICE_ACTIVATION_HOLD_FRAMES = 16;
static const int PUSH_TO_TALK_GATE_PEAK = 450;
static const int PUSH_TO_TALK_GATE_RMS = 120;
static const int PUSH_TO_TALK_HOLD_FRAMES = 4;
static const float CAPTURE_HIGHPASS_ALPHA = 0.995f;
static const int JITTER_BUFFER_TARGET_PACKETS = 14; // 140ms target playout delay
static const int JITTER_BUFFER_MIN_PACKETS = 10; // 100ms to restart after starvation
static const int JITTER_BUFFER_MAX_PACKETS = 30; // cap added latency at 300ms
static const int MAX_CONCEAL_PACKETS = 12; // up to 120ms concealment per gap
static const float CONCEAL_DECAY = 0.992f;
static void RefreshDeviceLists(
std::vector<std::wstring> &playbackDevices,
@@ -70,6 +75,11 @@ namespace
return static_cast<short>(sample);
}
static int RingBufferedSamples(int writePos, int readPos, int capacity)
{
return (writePos - readPos + capacity) % capacity;
}
struct VoiceFrameMetrics
{
int peak;
@@ -154,8 +164,14 @@ void VoiceChatManager::onCaptureAudio(ma_device *pDevice, void *pOutput, const v
EnterCriticalSection(&mgr->m_captureLock);
for (unsigned int i = 0; i < frameCount; i++)
{
const int nextWritePos = (mgr->m_captureWritePos + 1) % CAPTURE_BUFFER_SIZE;
if (nextWritePos == mgr->m_captureReadPos)
{
// Capture overrun: drop oldest sample to preserve continuity.
mgr->m_captureReadPos = (mgr->m_captureReadPos + 1) % CAPTURE_BUFFER_SIZE;
}
mgr->m_captureBuffer[mgr->m_captureWritePos] = input[i];
mgr->m_captureWritePos = (mgr->m_captureWritePos + 1) % CAPTURE_BUFFER_SIZE;
mgr->m_captureWritePos = nextWritePos;
}
LeaveCriticalSection(&mgr->m_captureLock);
}
@@ -201,20 +217,67 @@ void VoiceChatManager::onPlaybackAudio(ma_device *pDevice, void *pOutput, const
volume *= (float)mgr->m_voiceChatVolumePercent / 100.0f;
if (volume <= 0.001f) continue;
// Mix this stream's audio into the output
const int jitterTargetSamples = JITTER_BUFFER_TARGET_PACKETS * FRAMES_PER_PACKET;
const int jitterRestartSamples = JITTER_BUFFER_MIN_PACKETS * FRAMES_PER_PACKET;
const int jitterMaxSamples = JITTER_BUFFER_MAX_PACKETS * FRAMES_PER_PACKET;
int bufferedSamples = RingBufferedSamples(stream.writePos, stream.readPos, RemoteVoiceStream::BUFFER_SIZE);
// Keep latency bounded if packets burst in faster than real-time.
if (bufferedSamples > jitterMaxSamples)
{
int drop = bufferedSamples - jitterMaxSamples;
stream.readPos = (stream.readPos + drop) % RemoteVoiceStream::BUFFER_SIZE;
bufferedSamples -= drop;
}
// Prime before playout to absorb network/tick jitter.
if (!stream.primed)
{
if (bufferedSamples < jitterTargetSamples)
{
continue;
}
stream.primed = true;
}
// Mix this stream's audio into the output.
for (unsigned int i = 0; i < frameCount; i++)
{
if (stream.readPos != stream.writePos)
bufferedSamples = RingBufferedSamples(stream.writePos, stream.readPos, RemoteVoiceStream::BUFFER_SIZE);
short sourceSample = 0;
if (bufferedSamples > 0)
{
int sample = (int)(stream.buffer[stream.readPos] * volume);
int mixed = output[i] + sample;
// Clamp to prevent clipping
if (mixed > 32767) mixed = 32767;
if (mixed < -32768) mixed = -32768;
output[i] = (short)mixed;
sourceSample = stream.buffer[stream.readPos];
stream.readPos = (stream.readPos + 1) % RemoteVoiceStream::BUFFER_SIZE;
stream.lastOutputSample = sourceSample;
stream.pendingConcealSamples = 0;
}
else if (stream.pendingConcealSamples > 0)
{
const int faded = static_cast<int>(stream.lastOutputSample * CONCEAL_DECAY);
stream.lastOutputSample = ClampSampleToShort(faded);
sourceSample = stream.lastOutputSample;
--stream.pendingConcealSamples;
}
else
{
// Starved: re-prime with a smaller threshold before resuming to avoid tremble.
stream.primed = false;
const int refillBuffered = RingBufferedSamples(stream.writePos, stream.readPos, RemoteVoiceStream::BUFFER_SIZE);
if (refillBuffered < jitterRestartSamples)
{
break;
}
stream.primed = true;
}
if (sourceSample != 0)
{
int sample = static_cast<int>(sourceSample * volume);
int mixed = output[i] + sample;
output[i] = ClampSampleToShort(mixed);
}
}
}
@@ -266,6 +329,8 @@ void VoiceChatManager::init()
captureConfig.capture.channels = CHANNELS;
captureConfig.capture.pDeviceID = GetSelectedDeviceId(m_selectedCaptureDevice, pCaptureInfos, captureCount);
captureConfig.sampleRate = SAMPLE_RATE;
captureConfig.periodSizeInFrames = FRAMES_PER_PACKET;
captureConfig.periods = 3;
captureConfig.dataCallback = onCaptureAudio;
captureConfig.pUserData = this;
@@ -290,6 +355,8 @@ void VoiceChatManager::init()
playbackConfig.playback.channels = CHANNELS;
playbackConfig.playback.pDeviceID = GetSelectedDeviceId(m_selectedPlaybackDevice, pPlaybackInfos, playbackCount);
playbackConfig.sampleRate = SAMPLE_RATE;
playbackConfig.periodSizeInFrames = FRAMES_PER_PACKET;
playbackConfig.periods = 3;
playbackConfig.dataCallback = onPlaybackAudio;
playbackConfig.pUserData = this;
@@ -599,7 +666,8 @@ void VoiceChatManager::tick(Minecraft *minecraft)
bool shouldSend = false;
if (m_voiceInputMode == VOICE_INPUT_PUSH_TO_TALK)
{
shouldSend = m_isPushToTalkActive && (pushToTalkSpeechDetected || m_pushToTalkHoldFrames > 0);
// PTT should be stable while held; avoid extra per-frame speech gating cuts.
shouldSend = m_isPushToTalkActive;
}
else
{
@@ -689,16 +757,16 @@ void VoiceChatManager::receiveVoiceData(int playerId, unsigned short sequence, d
return; // stale packet
}
// Packet loss concealment: for small gaps, insert silence frames.
// Track short packet gaps; playback callback will synthesize a smooth decay.
const int missingPackets = static_cast<int>(delta) - 1;
if (missingPackets > 0)
{
const int concealPackets = (missingPackets > 3) ? 3 : missingPackets;
const int concealSamples = concealPackets * FRAMES_PER_PACKET;
for (int i = 0; i < concealSamples; ++i)
const int concealPackets = (missingPackets > MAX_CONCEAL_PACKETS) ? MAX_CONCEAL_PACKETS : missingPackets;
stream.pendingConcealSamples += concealPackets * FRAMES_PER_PACKET;
const int maxConcealSamples = MAX_CONCEAL_PACKETS * FRAMES_PER_PACKET;
if (stream.pendingConcealSamples > maxConcealSamples)
{
stream.buffer[stream.writePos] = 0;
stream.writePos = (stream.writePos + 1) % RemoteVoiceStream::BUFFER_SIZE;
stream.pendingConcealSamples = maxConcealSamples;
}
}
}
@@ -708,8 +776,23 @@ void VoiceChatManager::receiveVoiceData(int playerId, unsigned short sequence, d
for (int i = 0; i < sampleCount; i++)
{
const int nextWritePos = (stream.writePos + 1) % RemoteVoiceStream::BUFFER_SIZE;
if (nextWritePos == stream.readPos)
{
// Playback overrun: drop oldest sample to keep stream moving.
stream.readPos = (stream.readPos + 1) % RemoteVoiceStream::BUFFER_SIZE;
}
stream.buffer[stream.writePos] = samples[i];
stream.writePos = (stream.writePos + 1) % RemoteVoiceStream::BUFFER_SIZE;
stream.writePos = nextWritePos;
}
// Keep queue depth bounded to avoid growing latency after network bursts.
const int jitterMaxSamples = JITTER_BUFFER_MAX_PACKETS * FRAMES_PER_PACKET;
int bufferedSamples = RingBufferedSamples(stream.writePos, stream.readPos, RemoteVoiceStream::BUFFER_SIZE);
if (bufferedSamples > jitterMaxSamples)
{
const int drop = bufferedSamples - jitterMaxSamples;
stream.readPos = (stream.readPos + drop) % RemoteVoiceStream::BUFFER_SIZE;
}
LeaveCriticalSection(&m_playbackLock);