Fix recording truncation and rate mismatch causing static

The recording pipeline had two bugs that combined to produce
buzzing/static in recorded audio:

1. NormalizeStereoFrame truncated incoming audio frames to
   frameSize*AudioChannels samples. When the Opus decoder produced
   20ms frames (1920 stereo samples) but the recorder used a 10ms
   frameSize (960 samples), half the audio from every packet was
   silently dropped.

2. The recorder's run() loop dequeued one fixed-length frame per
   source per tick. After truncation, the remaining 10ms of each
   20ms packet was gone, so every other tick produced silence.
   This 50 Hz on/off pattern sounded like static.

Fixes:
- NormalizeStereoFrame no longer truncates; it only converts mono
  to stereo and preserves all audio data
- RecordAudioFrame now accepts an explicit stereo flag from callers
  instead of guessing from sample count (which failed for even-length
  mono data like 480-sample mic frames)
- run() accumulates variable-length frames per source and consumes
  them in fixed-size chunks, preserving any leftover for the next tick
This commit is contained in:
Brandon McGinty (deepseek)
2026-08-10 18:34:52 -04:00
committed by Brandon McGinty
parent cd003d7ac4
commit 90f8a1ca5f
3 changed files with 57 additions and 33 deletions
+4 -4
View File
@@ -45,7 +45,7 @@ type FilePlayer interface {
}
type Recorder interface {
RecordAudioFrame(source uint32, samples []int16)
RecordAudioFrame(source uint32, samples []int16, stereo bool)
}
const recorderOutgoingSource uint32 = ^uint32(0)
@@ -742,7 +742,7 @@ func (s *Stream) processAudioPacket(packet *gumble.AudioPacket, user *gumble.Use
}
}
if recorder != nil && recordPtr > 0 {
recorder.RecordAudioFrame(user.Session, recordBuffer[:recordPtr])
recorder.RecordAudioFrame(user.Session, recordBuffer[:recordPtr], true)
}
if len(emptyBufs) == 0 {
@@ -932,7 +932,7 @@ func (s *Stream) sourceRoutine(inputDevice *string, stop chan bool, done chan st
// Send stereo buffer when file is playing
outgoing <- gumble.AudioBuffer(outputBuffer)
if recorder := s.getRecorder(); recorder != nil {
recorder.RecordAudioFrame(recorderOutgoingSource, outputBuffer)
recorder.RecordAudioFrame(recorderOutgoingSource, outputBuffer, true)
}
} else if hasMicInput {
// Send mic when no file is playing. If the microphone is
@@ -949,7 +949,7 @@ func (s *Stream) sourceRoutine(inputDevice *string, stop chan bool, done chan st
}
outgoing <- gumble.AudioBuffer(outBuf)
if recorder := s.getRecorder(); recorder != nil {
recorder.RecordAudioFrame(recorderOutgoingSource, outBuf)
recorder.RecordAudioFrame(recorderOutgoingSource, outBuf, false)
}
}
}
+29 -25
View File
@@ -152,14 +152,14 @@ func (r *Recorder) Path() string {
return r.path
}
func (r *Recorder) RecordAudioFrame(source uint32, samples []int16) {
func (r *Recorder) RecordAudioFrame(source uint32, samples []int16, stereo bool) {
if r == nil || len(samples) == 0 {
return
}
if len(r.input) >= cap(r.input) {
return
}
frame := NormalizeStereoFrame(samples, r.frameSize)
frame := NormalizeStereoFrame(samples, stereo)
select {
case r.input <- sourceFrame{source: source, samples: frame}:
default:
@@ -192,31 +192,35 @@ func (r *Recorder) run() {
defer close(r.done)
ticker := time.NewTicker(r.interval)
defer ticker.Stop()
queues := make(map[uint32][][]int16)
frame := make([]int16, r.frameSize*gumble.AudioChannels)
// Per-source accumulated stereo samples. Incoming frames of any size are
// appended and then consumed in frameSize*AudioChannels chunks each tick.
queues := make(map[uint32][]int16)
chunkSize := r.frameSize * gumble.AudioChannels
chunk := make([]int16, chunkSize)
for {
select {
case <-r.stop:
r.closeEncoder()
return
case item := <-r.input:
queues[item.source] = append(queues[item.source], item.samples)
queues[item.source] = append(queues[item.source], item.samples...)
case <-ticker.C:
clear(frame)
for source, queue := range queues {
if len(queue) == 0 {
clear(chunk)
for source, buffer := range queues {
if len(buffer) == 0 {
delete(queues, source)
continue
}
mix(frame, queue[0])
queue = queue[1:]
if len(queue) == 0 {
// Mix one chunk worth of samples from this source.
if len(buffer) <= chunkSize {
mix(chunk, buffer)
delete(queues, source)
} else {
queues[source] = queue
mix(chunk, buffer[:chunkSize])
queues[source] = buffer[chunkSize:]
}
}
if err := writePCM(r.stdin, frame); err != nil {
if err := writePCM(r.stdin, chunk); err != nil {
r.setError(err)
r.closeEncoder()
return
@@ -248,19 +252,19 @@ func (r *Recorder) setError(err error) {
}
}
func NormalizeStereoFrame(samples []int16, frameSize int) []int16 {
out := make([]int16, frameSize*gumble.AudioChannels)
if len(samples) >= frameSize*gumble.AudioChannels && len(samples)%gumble.AudioChannels == 0 {
copy(out, samples[:frameSize*gumble.AudioChannels])
return out
// NormalizeStereoFrame ensures samples are in stereo interleaved format.
// If stereo is true the samples are returned as-is (already interleaved).
// Mono input is duplicated to both channels. The returned slice preserves
// all input audio without truncation.
func NormalizeStereoFrame(samples []int16, stereo bool) []int16 {
if stereo {
return samples
}
limit := frameSize
if len(samples) < limit {
limit = len(samples)
}
for i := 0; i < limit; i++ {
out[i*2] = samples[i]
out[i*2+1] = samples[i]
// Convert mono to stereo by duplicating each sample.
out := make([]int16, len(samples)*gumble.AudioChannels)
for i, s := range samples {
out[i*2] = s
out[i*2+1] = s
}
return out
}
+24 -4
View File
@@ -93,16 +93,36 @@ func TestReservePathPreventsConcurrentRecordingCollisions(t *testing.T) {
}
func TestNormalizeStereoFrame(t *testing.T) {
mono := NormalizeStereoFrame([]int16{1, -2}, 3)
wantMono := []int16{1, 1, -2, -2, 0, 0}
// Mono input duplicating each sample to both channels.
mono := NormalizeStereoFrame([]int16{1, -2, 3}, false)
wantMono := []int16{1, 1, -2, -2, 3, 3}
if len(mono) != len(wantMono) {
t.Fatalf("mono len = %d, want %d", len(mono), len(wantMono))
}
for i := range wantMono {
if mono[i] != wantMono[i] {
t.Fatalf("mono[%d] = %d, want %d", i, mono[i], wantMono[i])
}
}
stereo := NormalizeStereoFrame([]int16{1, 2, 3, 4, 5, 6}, 2)
wantStereo := []int16{1, 2, 3, 4}
// Even-length mono must not be mistaken for stereo.
monoEven := NormalizeStereoFrame([]int16{1, -2}, false)
wantMonoEven := []int16{1, 1, -2, -2}
if len(monoEven) != len(wantMonoEven) {
t.Fatalf("monoEven len = %d, want %d", len(monoEven), len(wantMonoEven))
}
for i := range wantMonoEven {
if monoEven[i] != wantMonoEven[i] {
t.Fatalf("monoEven[%d] = %d, want %d", i, monoEven[i], wantMonoEven[i])
}
}
// Stereo input passes through unchanged.
stereo := NormalizeStereoFrame([]int16{1, 2, 3, 4, 5, 6}, true)
wantStereo := []int16{1, 2, 3, 4, 5, 6}
if len(stereo) != len(wantStereo) {
t.Fatalf("stereo len = %d, want %d", len(stereo), len(wantStereo))
}
for i := range wantStereo {
if stereo[i] != wantStereo[i] {
t.Fatalf("stereo[%d] = %d, want %d", i, stereo[i], wantStereo[i])