MythEclipse/GMW
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

perf(voice): reduce transmit latency

Browse Source 
- FFmpeg: frame_duration 10ms, compression_level 0, nobuffer/low_delay flags, minimal probe
- Frontend: buffer 4096→1024 samples (~170ms→~42ms), faster base64 with chunking
- Remove processor.connect(destination) to avoid audio feedback loop

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
This commit is contained in:
MythEclipse
2026-06-09 01:07:35 +07:00
co-authored by Claude Opus 4.8
parent a4c2b93314
commit 7b74061788
2 changed files with 18 additions and 10 deletions
Download Patch File Download Diff File Expand all files Collapse all files
services/discord-gateway/src/modules/voice-recording/transmitter.ts
+9 -1
View File
@@ -50,8 +50,16 @@ export class VoiceTransmitter {
"-ar", "48000", // Output sample rate: 48kHz "-ar", "48000", // Output sample rate: 48kHz
"-ac", "2", // Output channels: stereo "-ac", "2", // Output channels: stereo
"-application", "lowdelay", // Low delay mode for real-time "-application", "lowdelay", // Low delay mode for real-time
"-frame_duration", "20", // 20ms frames "-frame_duration", "10", // 10ms frames (smaller = less latency)
"-packet_loss", "0", // No packet loss expected "-packet_loss", "0", // No packet loss expected
"-compression_level", "0", // Fastest compression (lowest CPU latency)
"-cutoff", "20000", // Audio cutoff frequency
"-vsync", "0", // No video sync
"-fflags", "nobuffer", // No buffering
"-flags", "low_delay", // Low delay processing
"-strict", "experimental",
"-analyzeduration", "0", // No analysis duration
"-probesize", "32", // Minimal probe size
"pipe:1", // Write to stdout "pipe:1", // Write to stdout
], { ], {
stdio: ["pipe", "pipe", "pipe"], stdio: ["pipe", "pipe", "pipe"],
services/frontend/src/shared/hooks/useAudioTransmit.ts
+9 -9
View File
@@ -58,10 +58,10 @@ export function useAudioTransmit(socketRef: {
const audioContext = new AudioContextCtor({ sampleRate: SAMPLE_RATE }); const audioContext = new AudioContextCtor({ sampleRate: SAMPLE_RATE });
audioContextRef.current = audioContext; audioContextRef.current = audioContext;
const source = audioContext.createMediaStreamSource(stream); const source = audioContext.createMediaStreamSource(stream);
const processor = audioContext.createScriptProcessor(4096, 1, 1); const processor = audioContext.createScriptProcessor(1024, 1, 1);
processorRef.current = processor; processorRef.current = processor;
source.connect(processor); source.connect(processor);
processor.connect(audioContext.destination); // Don't connect processor to destination (no monitoring feedback)
processor.onaudioprocess = (event) => { processor.onaudioprocess = (event) => {
if (!socketRef.current || socketRef.current.readyState !== WebSocket.OPEN) if (!socketRef.current || socketRef.current.readyState !== WebSocket.OPEN)
return; return;
@@ -70,15 +70,15 @@ export function useAudioTransmit(socketRef: {
for (let i = 0; i < inputData.length; i++) for (let i = 0; i < inputData.length; i++)
pcmData[i] = Math.max(-1, Math.min(1, inputData[i])) * 32767; pcmData[i] = Math.max(-1, Math.min(1, inputData[i])) * 32767;
// Convert to base64 // Fast base64 using Uint8Array + btoa
const bytes = new Uint8Array(pcmData.buffer); const uint8 = new Uint8Array(pcmData.buffer);
let binary = ''; let base64 = '';
for (let i = 0; i < bytes.length; i++) { const chunkSize = 8192;
binary += String.fromCharCode(bytes[i]); for (let i = 0; i < uint8.length; i += chunkSize) {
const chunk = uint8.subarray(i, i + chunkSize);
base64 += btoa(String.fromCharCode(...chunk));
} }
const base64 = btoa(binary);
// Send as JSON for backend to forward to Redis
socketRef.current.send(JSON.stringify({ socketRef.current.send(JSON.stringify({
type: 'voice_transmit', type: 'voice_transmit',
buffer: base64 buffer: base64