libobs: Correct raw output starting audio data

If the audio subsystem was buffered to any extent, the audio of a raw
output would start off at a negative offset, requiring each raw output
to implement a "prepare_audio" function (as seen in the FFmpeg output)
in order to ensure proper synchronization with video.  This did not
apply to encoded outputs because it was already being performed by the
obs-encoder code.

libobs/obs-internal.h

@@ -885,6 +885,14 @@ struct obs_output {
 	obs_service_t *service;
 	size_t mixer_mask;
 
+	struct circlebuf audio_buffer[MAX_AUDIO_MIXES][MAX_AV_PLANES];
+	uint64_t audio_start_ts;
+	uint64_t video_start_ts;
+	size_t audio_size;
+	size_t planes;
+	size_t sample_rate;
+	size_t total_audio_frames;
+
 	uint32_t scaled_width;
 	uint32_t scaled_height;
 
@@ -912,6 +920,8 @@ struct obs_output {
 	volatile bool delay_capturing;
 
 	char *last_error_message;
+
+	float audio_data[MAX_AUDIO_CHANNELS][AUDIO_OUTPUT_FRAMES];
 };
 
 static inline void do_output_signal(struct obs_output *output,

libobs/obs-output.c

@@ -169,6 +169,15 @@ static inline void free_packets(struct obs_output *output)
 	da_free(output->interleaved_packets);
 }
 
+static inline void clear_audio_buffers(obs_output_t *output)
+{
+	for (size_t i = 0; i < MAX_AUDIO_MIXES; i++) {
+		for (size_t j = 0; j < MAX_AV_PLANES; j++) {
+			circlebuf_free(&output->audio_buffer[i][j]);
+		}
+	}
+}
+
 void obs_output_destroy(obs_output_t *output)
 {
 	if (output) {
@@ -202,6 +211,8 @@ void obs_output_destroy(obs_output_t *output)
 			}
 		}
 
+		clear_audio_buffers(output);
+
 		os_event_destroy(output->stopping_event);
 		pthread_mutex_destroy(&output->caption_mutex);
 		pthread_mutex_destroy(&output->interleaved_mutex);
@@ -1524,19 +1535,89 @@ static void default_raw_video_callback(void *param, struct video_data *frame)
 	if (data_active(output))
 		output->info.raw_video(output->context.data, frame);
 	output->total_frames++;
+
+	if (!output->video_start_ts) {
+		output->video_start_ts = frame->timestamp;
+	}
+}
+
+static bool prepare_audio(struct obs_output *output,
+			  const struct audio_data *old, struct audio_data *new)
+{
+	*new = *old;
+
+	if (old->timestamp < output->video_start_ts) {
+		uint64_t duration = (uint64_t)old->frames * 1000000000 /
+				    (uint64_t)output->sample_rate;
+		uint64_t end_ts = (old->timestamp + duration);
+		uint64_t cutoff;
+
+		if (end_ts <= output->video_start_ts)
+			return false;
+
+		cutoff = output->video_start_ts - old->timestamp;
+		new->timestamp += cutoff;
+
+		cutoff = cutoff * (uint64_t)output->sample_rate / 1000000000;
+
+		for (size_t i = 0; i < output->planes; i++)
+			new->data[i] += output->audio_size *(uint32_t)cutoff;
+		new->frames -= (uint32_t)cutoff;
+	}
+
+	return true;
 }
 
 static void default_raw_audio_callback(void *param, size_t mix_idx,
-				       struct audio_data *frames)
+				       struct audio_data *in)
 {
 	struct obs_output *output = param;
+	struct audio_data out;
+	size_t frame_size_bytes;
+
 	if (!data_active(output))
 		return;
 
-	if (output->info.raw_audio2)
-		output->info.raw_audio2(output->context.data, mix_idx, frames);
-	else
-		output->info.raw_audio(output->context.data, frames);
+	/* -------------- */
+
+	if (!output->video_start_ts)
+		return;
+	if (!prepare_audio(output, in, &out))
+		return;
+	if (!output->audio_start_ts) {
+		output->audio_start_ts = out.timestamp;
+	}
+
+	frame_size_bytes = AUDIO_OUTPUT_FRAMES * output->audio_size;
+
+	for (size_t i = 0; i < output->planes; i++)
+		circlebuf_push_back(&output->audio_buffer[mix_idx][i],
+				    out.data[i],
+				    out.frames * output->audio_size);
+
+	/* -------------- */
+
+	while (output->audio_buffer[mix_idx][0].size > frame_size_bytes) {
+		for (size_t i = 0; i < output->planes; i++) {
+			circlebuf_pop_front(&output->audio_buffer[mix_idx][i],
+					    output->audio_data[i],
+					    frame_size_bytes);
+			out.data[i] = (uint8_t *)output->audio_data[i];
+		}
+
+		out.frames = AUDIO_OUTPUT_FRAMES;
+		out.timestamp = output->audio_start_ts +
+				audio_frames_to_ns(output->sample_rate,
+						   output->total_audio_frames);
+
+		output->total_audio_frames += AUDIO_OUTPUT_FRAMES;
+
+		if (output->info.raw_audio2)
+			output->info.raw_audio2(output->context.data, mix_idx,
+						&out);
+		else
+			output->info.raw_audio(output->context.data, &out);
+	}
 }
 
 static inline void start_audio_encoders(struct obs_output *output,
@@ -1797,6 +1878,36 @@ static bool begin_delayed_capture(obs_output_t *output)
 	return true;
 }
 
+static void reset_raw_output(obs_output_t *output)
+{
+	const struct audio_output_info *aoi =
+		audio_output_get_info(output->audio);
+	struct audio_convert_info conv = output->audio_conversion;
+	struct audio_convert_info info = {
+		aoi->samples_per_sec,
+		aoi->format,
+		aoi->speakers,
+	};
+
+	clear_audio_buffers(output);
+
+	if (output->audio_conversion_set) {
+		if (conv.samples_per_sec)
+			info.samples_per_sec = conv.samples_per_sec;
+		if (conv.format != AUDIO_FORMAT_UNKNOWN)
+			info.format = conv.format;
+		if (conv.speakers != SPEAKERS_UNKNOWN)
+			info.speakers = conv.speakers;
+	}
+
+	output->audio_start_ts = 0;
+	output->video_start_ts = 0;
+	output->sample_rate = info.samples_per_sec;
+	output->planes = get_audio_planes(info.format, info.speakers);
+	output->total_audio_frames = 0;
+	output->audio_size = get_audio_size(info.format, info.speakers, 1);
+}
+
 bool obs_output_begin_data_capture(obs_output_t *output, uint32_t flags)
 {
 	bool encoded, has_video, has_audio, has_service;
@@ -1812,6 +1923,10 @@ bool obs_output_begin_data_capture(obs_output_t *output, uint32_t flags)
 
 	output->total_frames = 0;
 
+	if ((output->info.flags & OBS_OUTPUT_ENCODED) == 0) {
+		reset_raw_output(output);
+	}
+
 	convert_flags(output, flags, &encoded, &has_video, &has_audio,
 		      &has_service);