Read audio data from pulseaudio as it's available and buffer it. Fixes audio recording on pulseaudio (and some pipewire configs)
This commit is contained in:
parent
5ba4c05953
commit
0059724fdc
@ -20,8 +20,6 @@
|
|||||||
|
|
||||||
typedef struct {
|
typedef struct {
|
||||||
void *handle;
|
void *handle;
|
||||||
void *buffer;
|
|
||||||
int buffer_size;
|
|
||||||
unsigned int frames;
|
unsigned int frames;
|
||||||
} SoundDevice;
|
} SoundDevice;
|
||||||
|
|
||||||
|
15
src/main.cpp
15
src/main.cpp
@ -1470,10 +1470,16 @@ int main(int argc, char **argv) {
|
|||||||
|
|
||||||
int64_t pts = 0;
|
int64_t pts = 0;
|
||||||
const double target_audio_hz = 1.0 / (double)audio_track.codec_context->sample_rate;
|
const double target_audio_hz = 1.0 / (double)audio_track.codec_context->sample_rate;
|
||||||
|
double received_audio_time = clock_get_monotonic_seconds();
|
||||||
|
|
||||||
while(running) {
|
while(running) {
|
||||||
void *sound_buffer;
|
void *sound_buffer;
|
||||||
int sound_buffer_size = sound_device_read_next_chunk(&audio_track.sound_device, &sound_buffer);
|
int sound_buffer_size = sound_device_read_next_chunk(&audio_track.sound_device, &sound_buffer);
|
||||||
|
const bool got_audio_data = sound_buffer_size >= 0;
|
||||||
|
|
||||||
|
const double this_audio_frame_time = clock_get_monotonic_seconds();
|
||||||
|
if(got_audio_data)
|
||||||
|
received_audio_time = this_audio_frame_time;
|
||||||
|
|
||||||
int ret = av_frame_make_writable(audio_track.frame);
|
int ret = av_frame_make_writable(audio_track.frame);
|
||||||
if (ret < 0) {
|
if (ret < 0) {
|
||||||
@ -1481,15 +1487,14 @@ int main(int argc, char **argv) {
|
|||||||
break;
|
break;
|
||||||
}
|
}
|
||||||
|
|
||||||
const double this_audio_frame_time = clock_get_monotonic_seconds();
|
const int64_t num_missing_frames = std::round((this_audio_frame_time - received_audio_time) / target_audio_hz / (int64_t)audio_track.frame->nb_samples);
|
||||||
const int64_t expected_frames = std::round((this_audio_frame_time - start_time_pts) / target_audio_hz);
|
|
||||||
const int64_t num_missing_frames = std::max(0L, (expected_frames - pts) / audio_track.frame->nb_samples);
|
|
||||||
// Jesus is there a better way to do this? I JUST WANT TO KEEP VIDEO AND AUDIO SYNCED HOLY FUCK I WANT TO KILL MYSELF NOW.
|
// Jesus is there a better way to do this? I JUST WANT TO KEEP VIDEO AND AUDIO SYNCED HOLY FUCK I WANT TO KILL MYSELF NOW.
|
||||||
// THIS PIECE OF SHIT WANTS EMPTY FRAMES OTHERWISE VIDEO PLAYS TOO FAST TO KEEP UP WITH AUDIO OR THE AUDIO PLAYS TOO EARLY.
|
// THIS PIECE OF SHIT WANTS EMPTY FRAMES OTHERWISE VIDEO PLAYS TOO FAST TO KEEP UP WITH AUDIO OR THE AUDIO PLAYS TOO EARLY.
|
||||||
// BUT WE CANT USE DELAYS TO GIVE DUMMY DATA BECAUSE PULSEAUDIO MIGHT GIVE AUDIO A BIG DELAYED!!!
|
// BUT WE CANT USE DELAYS TO GIVE DUMMY DATA BECAUSE PULSEAUDIO MIGHT GIVE AUDIO A BIG DELAYED!!!
|
||||||
if(num_missing_frames >= 5) {
|
if(num_missing_frames >= 5 || (num_missing_frames > 0 && got_audio_data)) {
|
||||||
// TODO:
|
// TODO:
|
||||||
//audio_track.frame->data[0] = empty_audio;
|
//audio_track.frame->data[0] = empty_audio;
|
||||||
|
received_audio_time = this_audio_frame_time;
|
||||||
swr_convert(swr, &audio_track.frame->data[0], audio_track.frame->nb_samples, (const uint8_t**)&empty_audio, audio_track.sound_device.frames);
|
swr_convert(swr, &audio_track.frame->data[0], audio_track.frame->nb_samples, (const uint8_t**)&empty_audio, audio_track.sound_device.frames);
|
||||||
// TODO: Check if duplicate frame can be saved just by writing it with a different pts instead of sending it again
|
// TODO: Check if duplicate frame can be saved just by writing it with a different pts instead of sending it again
|
||||||
for(int i = 0; i < num_missing_frames; ++i) {
|
for(int i = 0; i < num_missing_frames; ++i) {
|
||||||
@ -1504,7 +1509,7 @@ int main(int argc, char **argv) {
|
|||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
|
||||||
if(sound_buffer_size >= 0) {
|
if(got_audio_data) {
|
||||||
// TODO: Instead of converting audio, get float audio from alsa. Or does alsa do conversion internally to get this format?
|
// TODO: Instead of converting audio, get float audio from alsa. Or does alsa do conversion internally to get this format?
|
||||||
swr_convert(swr, &audio_track.frame->data[0], audio_track.frame->nb_samples, (const uint8_t**)&sound_buffer, audio_track.sound_device.frames);
|
swr_convert(swr, &audio_track.frame->data[0], audio_track.frame->nb_samples, (const uint8_t**)&sound_buffer, audio_track.sound_device.frames);
|
||||||
|
|
||||||
|
153
src/sound.cpp
153
src/sound.cpp
@ -21,14 +21,38 @@
|
|||||||
#include <stdio.h>
|
#include <stdio.h>
|
||||||
#include <string.h>
|
#include <string.h>
|
||||||
#include <cmath>
|
#include <cmath>
|
||||||
|
#include <time.h>
|
||||||
|
|
||||||
#include <pulse/pulseaudio.h>
|
#include <pulse/pulseaudio.h>
|
||||||
#include <pulse/mainloop.h>
|
#include <pulse/mainloop.h>
|
||||||
#include <pulse/xmalloc.h>
|
#include <pulse/xmalloc.h>
|
||||||
#include <pulse/error.h>
|
#include <pulse/error.h>
|
||||||
|
|
||||||
|
#define CHECK_DEAD_GOTO(p, rerror, label) \
|
||||||
|
do { \
|
||||||
|
if (!(p)->context || !PA_CONTEXT_IS_GOOD(pa_context_get_state((p)->context)) || \
|
||||||
|
!(p)->stream || !PA_STREAM_IS_GOOD(pa_stream_get_state((p)->stream))) { \
|
||||||
|
if (((p)->context && pa_context_get_state((p)->context) == PA_CONTEXT_FAILED) || \
|
||||||
|
((p)->stream && pa_stream_get_state((p)->stream) == PA_STREAM_FAILED)) { \
|
||||||
|
if (rerror) \
|
||||||
|
*(rerror) = pa_context_errno((p)->context); \
|
||||||
|
} else \
|
||||||
|
if (rerror) \
|
||||||
|
*(rerror) = PA_ERR_BADSTATE; \
|
||||||
|
goto label; \
|
||||||
|
} \
|
||||||
|
} while(false);
|
||||||
|
|
||||||
static int sound_device_index = 0;
|
static int sound_device_index = 0;
|
||||||
|
|
||||||
|
static double clock_get_monotonic_seconds() {
|
||||||
|
struct timespec ts;
|
||||||
|
ts.tv_sec = 0;
|
||||||
|
ts.tv_nsec = 0;
|
||||||
|
clock_gettime(CLOCK_MONOTONIC, &ts);
|
||||||
|
return (double)ts.tv_sec + (double)ts.tv_nsec * 0.000000001;
|
||||||
|
}
|
||||||
|
|
||||||
struct pa_handle {
|
struct pa_handle {
|
||||||
pa_context *context;
|
pa_context *context;
|
||||||
pa_stream *stream;
|
pa_stream *stream;
|
||||||
@ -37,6 +61,9 @@ struct pa_handle {
|
|||||||
const void *read_data;
|
const void *read_data;
|
||||||
size_t read_index, read_length;
|
size_t read_index, read_length;
|
||||||
|
|
||||||
|
uint8_t *output_data;
|
||||||
|
size_t output_index, output_length;
|
||||||
|
|
||||||
int operation_success;
|
int operation_success;
|
||||||
};
|
};
|
||||||
|
|
||||||
@ -54,6 +81,11 @@ static void pa_sound_device_free(pa_handle *s) {
|
|||||||
if (s->mainloop)
|
if (s->mainloop)
|
||||||
pa_mainloop_free(s->mainloop);
|
pa_mainloop_free(s->mainloop);
|
||||||
|
|
||||||
|
if (s->output_data) {
|
||||||
|
free(s->output_data);
|
||||||
|
s->output_data = NULL;
|
||||||
|
}
|
||||||
|
|
||||||
pa_xfree(s);
|
pa_xfree(s);
|
||||||
}
|
}
|
||||||
|
|
||||||
@ -68,6 +100,21 @@ static pa_handle* pa_sound_device_new(const char *server,
|
|||||||
int error = PA_ERR_INTERNAL, r;
|
int error = PA_ERR_INTERNAL, r;
|
||||||
|
|
||||||
p = pa_xnew0(pa_handle, 1);
|
p = pa_xnew0(pa_handle, 1);
|
||||||
|
p->read_data = NULL;
|
||||||
|
p->read_length = 0;
|
||||||
|
p->read_index = 0;
|
||||||
|
|
||||||
|
const int buffer_size = attr->maxlength;
|
||||||
|
void *buffer = malloc(buffer_size);
|
||||||
|
if(!buffer) {
|
||||||
|
fprintf(stderr, "failed to allocate buffer for audio\n");
|
||||||
|
*rerror = -1;
|
||||||
|
return NULL;
|
||||||
|
}
|
||||||
|
|
||||||
|
p->output_data = (uint8_t*)buffer;
|
||||||
|
p->output_length = buffer_size;
|
||||||
|
p->output_index = 0;
|
||||||
|
|
||||||
if (!(p->mainloop = pa_mainloop_new()))
|
if (!(p->mainloop = pa_mainloop_new()))
|
||||||
goto fail;
|
goto fail;
|
||||||
@ -130,30 +177,86 @@ fail:
|
|||||||
return NULL;
|
return NULL;
|
||||||
}
|
}
|
||||||
|
|
||||||
// Returns a negative value on failure or if no data is available at the moment
|
// Returns a negative value on failure or if |p->output_length| data is not available within the time frame specified by the sample rate
|
||||||
static int pa_sound_device_read(pa_handle *p, void *data, size_t length) {
|
static int pa_sound_device_read(pa_handle *p) {
|
||||||
assert(p);
|
assert(p);
|
||||||
|
|
||||||
const int64_t timeout_ms = std::round((1000.0 / (double)pa_stream_get_sample_spec(p->stream)->rate) * 1000.0);
|
const int64_t timeout_ms = std::round((1000.0 / (double)pa_stream_get_sample_spec(p->stream)->rate) * 1000.0);
|
||||||
pa_mainloop_prepare(p->mainloop, timeout_ms * 1000);
|
const double start_time = clock_get_monotonic_seconds();
|
||||||
pa_mainloop_poll(p->mainloop);
|
|
||||||
pa_mainloop_dispatch(p->mainloop);
|
|
||||||
|
|
||||||
if(pa_stream_readable_size(p->stream) < length)
|
bool success = false;
|
||||||
return -1;
|
int r = 0;
|
||||||
|
int *rerror = &r;
|
||||||
|
CHECK_DEAD_GOTO(p, rerror, fail);
|
||||||
|
|
||||||
int r = pa_stream_peek(p->stream, &p->read_data, &p->read_length);
|
while (p->output_index < p->output_length) {
|
||||||
if(r != 0)
|
if((clock_get_monotonic_seconds() - start_time) * 1000 >= timeout_ms)
|
||||||
return -1;
|
return -1;
|
||||||
|
|
||||||
if(p->read_length < length || !p->read_data) {
|
if(p->read_data) {
|
||||||
pa_stream_drop(p->stream);
|
assert(p->output_index == 0);
|
||||||
return -1;
|
memcpy(p->output_data, (const uint8_t*)p->read_data + p->read_index, p->read_length);
|
||||||
|
p->output_index += p->read_length;
|
||||||
|
p->read_data = NULL;
|
||||||
|
p->read_length = 0;
|
||||||
|
p->read_index = 0;
|
||||||
|
|
||||||
|
if(pa_stream_drop(p->stream) != 0)
|
||||||
|
goto fail;
|
||||||
|
}
|
||||||
|
|
||||||
|
pa_mainloop_prepare(p->mainloop, 1 * 1000); // 1 ms
|
||||||
|
pa_mainloop_poll(p->mainloop);
|
||||||
|
pa_mainloop_dispatch(p->mainloop);
|
||||||
|
|
||||||
|
if(pa_stream_peek(p->stream, &p->read_data, &p->read_length) < 0)
|
||||||
|
goto fail;
|
||||||
|
|
||||||
|
if(!p->read_data && p->read_length == 0)
|
||||||
|
continue;
|
||||||
|
|
||||||
|
if(!p->read_data && p->read_length > 0) {
|
||||||
|
// There is a hole in the stream :( drop it. Maybe we should generate silence instead? TODO
|
||||||
|
if(pa_stream_drop(p->stream) != 0)
|
||||||
|
goto fail;
|
||||||
|
continue;
|
||||||
|
}
|
||||||
|
|
||||||
|
if(p->read_length <= 0) {
|
||||||
|
CHECK_DEAD_GOTO(p, rerror, fail);
|
||||||
|
continue;
|
||||||
|
}
|
||||||
|
|
||||||
|
const size_t space_free_in_output_buffer = p->output_length - p->output_index;
|
||||||
|
if(space_free_in_output_buffer < p->read_length) {
|
||||||
|
assert(p->read_index == 0);
|
||||||
|
memcpy(p->output_data + p->output_index, p->read_data, space_free_in_output_buffer);
|
||||||
|
p->output_index = 0;
|
||||||
|
p->read_index += space_free_in_output_buffer;
|
||||||
|
p->read_length -= space_free_in_output_buffer;
|
||||||
|
break;
|
||||||
|
} else {
|
||||||
|
assert(p->read_index == 0);
|
||||||
|
memcpy(p->output_data + p->output_index, p->read_data, p->read_length);
|
||||||
|
p->output_index += p->read_length;
|
||||||
|
p->read_data = NULL;
|
||||||
|
p->read_length = 0;
|
||||||
|
p->read_index = 0;
|
||||||
|
|
||||||
|
if(pa_stream_drop(p->stream) != 0)
|
||||||
|
goto fail;
|
||||||
|
|
||||||
|
if(p->output_index == p->output_length) {
|
||||||
|
p->output_index = 0;
|
||||||
|
break;
|
||||||
|
}
|
||||||
|
}
|
||||||
}
|
}
|
||||||
|
|
||||||
memcpy(data, p->read_data, length);
|
success = true;
|
||||||
pa_stream_drop(p->stream);
|
|
||||||
return 0;
|
fail:
|
||||||
|
return success ? 0 : -1;
|
||||||
}
|
}
|
||||||
|
|
||||||
int sound_device_get_by_name(SoundDevice *device, const char *name, unsigned int num_channels, unsigned int period_frame_size) {
|
int sound_device_get_by_name(SoundDevice *device, const char *name, unsigned int num_channels, unsigned int period_frame_size) {
|
||||||
@ -181,33 +284,21 @@ int sound_device_get_by_name(SoundDevice *device, const char *name, unsigned int
|
|||||||
return -1;
|
return -1;
|
||||||
}
|
}
|
||||||
|
|
||||||
int buffer_size = buffer_attr.maxlength;
|
|
||||||
void *buffer = malloc(buffer_size);
|
|
||||||
if(!buffer) {
|
|
||||||
fprintf(stderr, "failed to allocate buffer for audio\n");
|
|
||||||
pa_sound_device_free(handle);
|
|
||||||
return -1;
|
|
||||||
}
|
|
||||||
|
|
||||||
fprintf(stderr, "Using pulseaudio\n");
|
|
||||||
|
|
||||||
device->handle = handle;
|
device->handle = handle;
|
||||||
device->buffer = buffer;
|
|
||||||
device->buffer_size = buffer_size;
|
|
||||||
device->frames = period_frame_size;
|
device->frames = period_frame_size;
|
||||||
return 0;
|
return 0;
|
||||||
}
|
}
|
||||||
|
|
||||||
void sound_device_close(SoundDevice *device) {
|
void sound_device_close(SoundDevice *device) {
|
||||||
pa_sound_device_free((pa_handle*)device->handle);
|
pa_sound_device_free((pa_handle*)device->handle);
|
||||||
free(device->buffer);
|
|
||||||
}
|
}
|
||||||
|
|
||||||
int sound_device_read_next_chunk(SoundDevice *device, void **buffer) {
|
int sound_device_read_next_chunk(SoundDevice *device, void **buffer) {
|
||||||
if(pa_sound_device_read((pa_handle*)device->handle, device->buffer, device->buffer_size) < 0) {
|
pa_handle *pa = (pa_handle*)device->handle;
|
||||||
|
if(pa_sound_device_read(pa) < 0) {
|
||||||
//fprintf(stderr, "pa_simple_read() failed: %s\n", pa_strerror(error));
|
//fprintf(stderr, "pa_simple_read() failed: %s\n", pa_strerror(error));
|
||||||
return -1;
|
return -1;
|
||||||
}
|
}
|
||||||
*buffer = device->buffer;
|
*buffer = pa->output_data;
|
||||||
return device->frames;
|
return device->frames;
|
||||||
}
|
}
|
Loading…
Reference in New Issue
Block a user