jeka/utun2
1
0 Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Раздели аудиопотоки и привяжи AEC к фактическому общему выходу

Browse Source
This commit is contained in:
evgeny committed 2026-10-02 10:22:34 +03:00
1 parent d7fb0c83f9
commit bd082a26bf
30 files changed
+1839 -412
Download Patch File Download Diff File Expand all files Collapse all files

No files matched your search

lib/miniaudio.h
+21 -10
View File
@@ -7565,6 +7565,7 @@ struct ma_context
ma_proc pa_stream_get_buffer_attr;
ma_proc pa_stream_set_buffer_attr;
ma_proc pa_stream_get_device_name;
ma_proc pa_stream_get_latency;
ma_proc pa_stream_set_write_callback;
ma_proc pa_stream_set_read_callback;
ma_proc pa_stream_set_suspended_callback;
@@ -30959,6 +30960,7 @@ typedef const ma_pa_sample_spec* (* ma_pa_stream_get_sample_spec_proc) (
typedef const ma_pa_channel_map* (* ma_pa_stream_get_channel_map_proc) (ma_pa_stream* s);
typedef const ma_pa_buffer_attr* (* ma_pa_stream_get_buffer_attr_proc) (ma_pa_stream* s);
typedef ma_pa_operation* (* ma_pa_stream_set_buffer_attr_proc) (ma_pa_stream* s, const ma_pa_buffer_attr* attr, ma_pa_stream_success_cb_t cb, void* userdata);
typedef int (* ma_pa_stream_get_latency_proc)(const ma_pa_stream* s, ma_uint64* usec, int* negative);
typedef const char* (* ma_pa_stream_get_device_name_proc) (const ma_pa_stream* s);
typedef void (* ma_pa_stream_set_write_callback_proc) (ma_pa_stream* s, ma_pa_stream_request_cb_t cb, void* userdata);
typedef void (* ma_pa_stream_set_read_callback_proc) (ma_pa_stream* s, ma_pa_stream_request_cb_t cb, void* userdata);
@@ -31840,6 +31842,9 @@ static void ma_device_on_read__pulse(ma_pa_stream* pStream, size_t byteCount, vo
ma_silence_pcm_frames(silence, capacity, pDevice->capture.internalFormat, pDevice->capture.internalChannels);
while (remaining > 0 && ma_device_get_state(pDevice) == ma_device_state_started) {
ma_uint32 n = (ma_uint32)ma_min(remaining, capacity);
#ifdef MA_AUDIO_CAPTURE_HOLE
MA_AUDIO_CAPTURE_HOLE(framesMapped - remaining);
#endif
ma_device_handle_backend_data_callback(pDevice, NULL, silence, n);
remaining -= n;
}
@@ -32144,7 +32149,7 @@ static ma_result ma_device_init__pulse(ma_device* pDevice, const ma_device_confi
if (pDescriptorCapture->sampleRate != 0) {
ss.rate = pDescriptorCapture->sampleRate;
}
streamFlags = MA_PA_STREAM_START_CORKED | MA_PA_STREAM_ADJUST_LATENCY;
streamFlags = MA_PA_STREAM_START_CORKED | MA_PA_STREAM_ADJUST_LATENCY | MA_PA_STREAM_AUTO_TIMING_UPDATE | MA_PA_STREAM_INTERPOLATE_TIMING;
if (ma_format_from_pulse(ss.format) == ma_format_unknown) {
if (ma_is_little_endian()) {
@@ -32303,7 +32308,7 @@ static ma_result ma_device_init__pulse(ma_device* pDevice, const ma_device_confi
ss.rate = pDescriptorPlayback->sampleRate;
}
streamFlags = MA_PA_STREAM_START_CORKED | MA_PA_STREAM_ADJUST_LATENCY;
streamFlags = MA_PA_STREAM_START_CORKED | MA_PA_STREAM_ADJUST_LATENCY | MA_PA_STREAM_AUTO_TIMING_UPDATE | MA_PA_STREAM_INTERPOLATE_TIMING;
if (ma_format_from_pulse(ss.format) == ma_format_unknown) {
if (ma_is_little_endian()) {
ss.format = MA_PA_SAMPLE_FLOAT32LE;
@@ -32729,6 +32734,7 @@ static ma_result ma_context_init__pulse(ma_context* pContext, const ma_context_c
pContext->pulse.pa_stream_get_channel_map = (ma_proc)ma_dlsym(ma_context_get_log(pContext), pContext->pulse.pulseSO, "pa_stream_get_channel_map");
pContext->pulse.pa_stream_get_buffer_attr = (ma_proc)ma_dlsym(ma_context_get_log(pContext), pContext->pulse.pulseSO, "pa_stream_get_buffer_attr");
pContext->pulse.pa_stream_set_buffer_attr = (ma_proc)ma_dlsym(ma_context_get_log(pContext), pContext->pulse.pulseSO, "pa_stream_set_buffer_attr");
pContext->pulse.pa_stream_get_latency = (ma_proc)ma_dlsym(ma_context_get_log(pContext), pContext->pulse.pulseSO, "pa_stream_get_latency");
pContext->pulse.pa_stream_get_device_name = (ma_proc)ma_dlsym(ma_context_get_log(pContext), pContext->pulse.pulseSO, "pa_stream_get_device_name");
pContext->pulse.pa_stream_set_write_callback = (ma_proc)ma_dlsym(ma_context_get_log(pContext), pContext->pulse.pulseSO, "pa_stream_set_write_callback");
pContext->pulse.pa_stream_set_read_callback = (ma_proc)ma_dlsym(ma_context_get_log(pContext), pContext->pulse.pulseSO, "pa_stream_set_read_callback");
@@ -32795,6 +32801,7 @@ static ma_result ma_context_init__pulse(ma_context* pContext, const ma_context_c
ma_pa_stream_get_channel_map_proc _pa_stream_get_channel_map = pa_stream_get_channel_map;
ma_pa_stream_get_buffer_attr_proc _pa_stream_get_buffer_attr = pa_stream_get_buffer_attr;
ma_pa_stream_set_buffer_attr_proc _pa_stream_set_buffer_attr = pa_stream_set_buffer_attr;
ma_pa_stream_get_latency_proc _pa_stream_get_latency = pa_stream_get_latency;
ma_pa_stream_get_device_name_proc _pa_stream_get_device_name = pa_stream_get_device_name;
ma_pa_stream_set_write_callback_proc _pa_stream_set_write_callback = pa_stream_set_write_callback;
ma_pa_stream_set_read_callback_proc _pa_stream_set_read_callback = pa_stream_set_read_callback;
@@ -32860,6 +32867,7 @@ static ma_result ma_context_init__pulse(ma_context* pContext, const ma_context_c
pContext->pulse.pa_stream_get_channel_map = (ma_proc)_pa_stream_get_channel_map;
pContext->pulse.pa_stream_get_buffer_attr = (ma_proc)_pa_stream_get_buffer_attr;
pContext->pulse.pa_stream_set_buffer_attr = (ma_proc)_pa_stream_set_buffer_attr;
pContext->pulse.pa_stream_get_latency = (ma_proc)_pa_stream_get_latency;
pContext->pulse.pa_stream_get_device_name = (ma_proc)_pa_stream_get_device_name;
pContext->pulse.pa_stream_set_write_callback = (ma_proc)_pa_stream_set_write_callback;
pContext->pulse.pa_stream_set_read_callback = (ma_proc)_pa_stream_set_read_callback;
@@ -44639,6 +44647,14 @@ MA_API ma_result ma_device_handle_backend_data_callback(ma_device* pDevice, void
if (frameCount == 0) {
return MA_INVALID_ARGS;
}
if (((pDevice->type == ma_device_type_capture || pDevice->type == ma_device_type_loopback) && pInput == NULL) ||
(pDevice->type == ma_device_type_playback && pOutput == NULL)) {
return MA_INVALID_ARGS;
}
#ifdef MA_AUDIO_TIMING_BEGIN
MA_AUDIO_TIMING_BEGIN(pDevice, frameCount, pInput != NULL);
#endif
if (pDevice->type == ma_device_type_duplex) {
if (pInput != NULL) {
@@ -44650,22 +44666,17 @@ MA_API ma_result ma_device_handle_backend_data_callback(ma_device* pDevice, void
}
} else {
if (pDevice->type == ma_device_type_capture || pDevice->type == ma_device_type_loopback) {
if (pInput == NULL) {
return MA_INVALID_ARGS;
}
ma_device__send_frames_to_client(pDevice, frameCount, pInput);
}
if (pDevice->type == ma_device_type_playback) {
if (pOutput == NULL) {
return MA_INVALID_ARGS;
}
ma_device__read_frames_from_client(pDevice, frameCount, pOutput);
}
}
#ifdef MA_AUDIO_TIMING_END
MA_AUDIO_TIMING_END();
#endif
return MA_SUCCESS;
}
lib/speex_aec.c
+24 -14
View File
@@ -24,6 +24,7 @@
struct speex_aec {
SpeexEchoState* st;
int mic_channels, render_channels;
int frame_samples; /* сэмплов в кадре (960 @48кГц) */
int depth; /* глубина линии задержки в кадрах (>=1) */
int16_t* line; /* кольцо depth*frame_samples */
@@ -42,17 +43,25 @@ static void aec_enqueue(speex_aec_t* a, const int16_t* frame) {
a->count--;
a->overruns++;
}
memcpy(a->line + (size_t)((a->head + a->count) % a->depth) * a->frame_samples,
frame, (size_t)a->frame_samples * sizeof(int16_t));
memcpy(a->line + (size_t)((a->head + a->count) % a->depth) * a->frame_samples * a->render_channels,
frame, (size_t)a->frame_samples * a->render_channels * sizeof(int16_t));
a->count++;
}
speex_aec_t* speex_aec_create(int sample_rate, int frame_samples, int filter_samples, int delay_frames) {
return speex_aec_create_mc(sample_rate, frame_samples, filter_samples, delay_frames, 1, 1);
}
/* Независимые каналы референса: противофазное стерео не усредняется. */
speex_aec_t* speex_aec_create_mc(int sample_rate, int frame_samples, int filter_samples, int delay_frames,
int mic_channels, int render_channels) {
speex_aec_t* a;
int rate;
if (sample_rate <= 0 || frame_samples <= 0 || filter_samples < frame_samples ||
(delay_frames > 0 && delay_frames > INT_MAX / frame_samples)) {
mic_channels < 1 || mic_channels > 2 || render_channels < 1 || render_channels > 2 ||
frame_samples > INT_MAX / render_channels || frame_samples > INT_MAX / mic_channels ||
(delay_frames > 0 && delay_frames > INT_MAX / frame_samples / render_channels)) {
DEBUG_ERROR(DEBUG_CATEGORY_AEC, "%s: bad args rate=%d frame=%d filter=%d delay=%d",
AEC_ID, sample_rate, frame_samples, filter_samples, delay_frames);
return NULL;
@@ -64,7 +73,7 @@ speex_aec_t* speex_aec_create(int sample_rate, int frame_samples, int filter_sam
return NULL;
}
a->st = speex_echo_state_init(frame_samples, filter_samples);
a->st = speex_echo_state_init_mc(frame_samples, filter_samples, mic_channels, render_channels);
if (!a->st) {
DEBUG_ERROR(DEBUG_CATEGORY_AEC, "%s: echo state allocation failed", AEC_ID);
u_free(a);
@@ -78,10 +87,11 @@ speex_aec_t* speex_aec_create(int sample_rate, int frame_samples, int filter_sam
return NULL;
}
a->mic_channels = mic_channels; a->render_channels = render_channels;
a->frame_samples = frame_samples;
a->depth = delay_frames > 0 ? delay_frames : 1;
a->line = (int16_t*)u_calloc((uint32_t)(a->depth * frame_samples), sizeof(int16_t));
a->play_acc = (int16_t*)u_calloc((uint32_t)frame_samples, sizeof(int16_t));
a->line = (int16_t*)u_calloc((uint32_t)(a->depth * frame_samples * render_channels), sizeof(int16_t));
a->play_acc = (int16_t*)u_calloc((uint32_t)(frame_samples * render_channels), sizeof(int16_t));
if (!a->line || !a->play_acc) {
DEBUG_ERROR(DEBUG_CATEGORY_AEC, "%s: OOM for buffers", AEC_ID);
if (a->line) u_free(a->line);
@@ -91,8 +101,8 @@ speex_aec_t* speex_aec_create(int sample_rate, int frame_samples, int filter_sam
return NULL;
}
DEBUG_INFO(DEBUG_CATEGORY_AEC, "%s: created rate=%d frame=%d filter=%d delay=%d frames",
AEC_ID, sample_rate, frame_samples, filter_samples, a->depth);
DEBUG_INFO(DEBUG_CATEGORY_AEC, "%s: created rate=%d frame=%d filter=%d delay=%d frames mic=%d render=%d",
AEC_ID, sample_rate, frame_samples, filter_samples, a->depth, mic_channels, render_channels);
return a;
}
@@ -117,7 +127,7 @@ void speex_aec_feed_playback(speex_aec_t* a, const int16_t* pcm, int count) {
int fs, take;
if (!a || !pcm || count <= 0) return;
fs = a->frame_samples;
fs = a->frame_samples * a->render_channels;
while (count > 0) {
take = fs - a->play_len;
@@ -139,23 +149,23 @@ int speex_aec_process_capture(speex_aec_t* a, const int16_t* pcm, int count, int
if (!a || !pcm || !out || count <= 0) return 0;
fs = a->frame_samples;
if (count != fs) {
DEBUG_ERROR(DEBUG_CATEGORY_AEC, "%s: capture requires one frame samples=%d expected=%d", AEC_ID, count, fs);
if (count != fs * a->mic_channels) {
DEBUG_ERROR(DEBUG_CATEGORY_AEC, "%s: capture requires one frame samples=%d expected=%d", AEC_ID, count, fs * a->mic_channels);
return 0;
}
if (a->count == a->depth) {
const int16_t* ref = a->line + (size_t)a->head * fs;
const int16_t* ref = a->line + (size_t)a->head * fs * a->render_channels;
speex_echo_cancellation(a->st, pcm, ref, out);
a->head = (a->head + 1) % a->depth;
a->count--;
} else {
/* линия не наполнена (старт или рендер отстаёт): passthrough без канселлера */
memcpy(out, pcm, (size_t)fs * sizeof(int16_t));
memcpy(out, pcm, (size_t)count * sizeof(int16_t));
a->underruns++;
}
return fs;
return count;
}
int speex_aec_delay_fill(const speex_aec_t* a) {
lib/speex_aec.h
+8 -3
View File
@@ -2,7 +2,7 @@
* speex_aec.h — акустическое эхоподавление (AEC), C-обёртка над SpeexDSP mdf.c.
*
* Подавляет эхо дальнего конца (то, что играем в динамик) в сигнале микрофона
* перед кодированием. Работает на int16 PCM, один канал, частота 48000 (можно
* перед кодированием. Работает на interleaved int16 PCM, 1/2 канала, частота 48000 (можно
* 8000/16000/32000/48000), кадр 20 мс (960 сэмплов @48 кГц).
*
* Модель использования (duplex-контур звонка):
@@ -44,6 +44,11 @@ typedef struct speex_aec speex_aec_t;
*/
speex_aec_t* speex_aec_create(int sample_rate, int frame_samples, int filter_samples, int delay_frames);
/* Многоканальный AEC: frame_samples на канал, count в feed/process — все interleaved отсчёты.
* mic_channels и render_channels — 1/2; возвращаемый count захвата равен frame_samples*mic_channels. */
speex_aec_t* speex_aec_create_mc(int sample_rate, int frame_samples, int filter_samples, int delay_frames,
int mic_channels, int render_channels);
/* Освободить канселлер. NULL безопасен. */
void speex_aec_destroy(speex_aec_t* aec);
@@ -57,8 +62,8 @@ void speex_aec_reset(speex_aec_t* aec);
void speex_aec_feed_playback(speex_aec_t* aec, const int16_t* pcm, int count);
/**
* Обработать ровно один кадр захвата (count == frame_samples).
* Возвращает frame_samples при успехе, 0 при ошибке. out вмещает полный кадр
* Обработать ровно один кадр захвата (count == frame_samples*mic_channels).
* Возвращает count при успехе, 0 при ошибке. out вмещает полный interleaved кадр
* и не должен алиаситься с pcm. Накопитель аппаратных чанков принадлежит I/O-движку.
*/
int speex_aec_process_capture(speex_aec_t* aec, const int16_t* pcm, int count, int16_t* out);