Skip to content

Commit

Permalink
resample raw audio
Browse files Browse the repository at this point in the history
  • Loading branch information
@maxnut
maxnut committed Nov 25, 2024
1 parent a143e7e commit 0bbadf2
Showing 1 changed file with 73 additions and 12 deletions.
85 changes: 73 additions & 12 deletions src/audio_mixer.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -123,6 +123,62 @@ std::vector<float> readAudioFile(const char *filename, int targetSampleRate, AVS
return audioFrames;
}

std::vector<float> resampleAudio(const std::vector<float>& inputAudio, int inputSampleRate, int targetSampleRate) {
SwrContext *swrCtx = nullptr;
int ret;
AVChannelLayout ch_layout;
av_channel_layout_from_string(&ch_layout, "2 channels");

ret = swr_alloc_set_opts2(&swrCtx, &ch_layout, AV_SAMPLE_FMT_FLT,
targetSampleRate, &ch_layout, AV_SAMPLE_FMT_FLT,
inputSampleRate, 0, nullptr);

if (ret < 0) {
char errbuf[AV_ERROR_MAX_STRING_SIZE];
av_strerror(ret, errbuf, AV_ERROR_MAX_STRING_SIZE);
geode::log::error("Failed to set up swr context: {}", errbuf);
return {};
}
ret = swr_init(swrCtx);
if (ret < 0) {
char errbuf[AV_ERROR_MAX_STRING_SIZE];
av_strerror(ret, errbuf, AV_ERROR_MAX_STRING_SIZE);
geode::log::error("Failed to initialize swr context: {}", errbuf);
return {};
}

const int chunkSize = 4096;
const int numChannels = 2;

int maxOutputSamples = av_rescale_rnd(chunkSize, targetSampleRate, inputSampleRate, AV_ROUND_UP);
std::vector<float> outputAudio;
std::vector<float> outputChunk(maxOutputSamples * numChannels);

const uint8_t* inData[1] = { nullptr };
uint8_t* outData[1] = { reinterpret_cast<uint8_t*>(outputChunk.data()) };

for (size_t i = 0; i < inputAudio.size(); i += chunkSize * numChannels) {
size_t currentChunkSize = std::min((size_t)(chunkSize * numChannels), inputAudio.size() - i);

inData[0] = reinterpret_cast<const uint8_t*>(&inputAudio[i]);
int inputSamples = currentChunkSize / numChannels;

int resampledSamples = swr_convert(swrCtx, outData, maxOutputSamples, inData, inputSamples);
if (ret < 0) {
char errbuf[AV_ERROR_MAX_STRING_SIZE];
av_strerror(ret, errbuf, AV_ERROR_MAX_STRING_SIZE);
geode::log::error("Failed to convert audio frame: {}", errbuf);
return {};
}

outputAudio.insert(outputAudio.end(), outputChunk.begin(), outputChunk.begin() + resampledSamples * numChannels);
}

swr_free(&swrCtx);

return outputAudio;
}

namespace ffmpeg {
void AudioMixer::mixVideoAudio(std::filesystem::path videoFile, std::filesystem::path audioFile, std::filesystem::path outputMp4File) {
constexpr int frameSize = 1024;
Expand All @@ -143,9 +199,7 @@ namespace ffmpeg {
}

void AudioMixer::mixVideoRaw(std::filesystem::path videoFile, const std::vector<float>& raw, std::filesystem::path outputMp4File, uint32_t sampleRate) {
const int frameSize = 1024;

geode::log::debug("raw size {}", raw.size());
const int frameSize = 1024;

AVFormatContext* videoFormatContext = nullptr;
if (avformat_open_input(&videoFormatContext, videoFile.string().c_str(), nullptr, nullptr) < 0) {
Expand Down Expand Up @@ -187,6 +241,17 @@ namespace ffmpeg {

constexpr int channels = 2;

avformat_find_stream_info(videoFormatContext, nullptr);
auto duration = static_cast<double>(videoFormatContext->duration) / AV_TIME_BASE;
auto newSampleRate = raw.size() / duration / channels;

std::vector<float> resampled = resampleAudio(raw, newSampleRate, 44100);

if(resampled.empty()) {
geode::log::error("Failed to resample audio.");
return;
}

outputAudioStream->codecpar->codec_tag = 0;
outputAudioStream->codecpar->codec_type = AVMEDIA_TYPE_AUDIO;
outputAudioStream->codecpar->sample_rate = sampleRate;
Expand Down Expand Up @@ -242,9 +307,7 @@ namespace ffmpeg {
avformat_close_input(&videoFormatContext);
return;
}

geode::log::debug("1 timebase {} {} {} {}", videoFormatContext->streams[videoStreamIndex]->time_base.num, videoFormatContext->streams[videoStreamIndex]->time_base.den, outputVideoStream->time_base.num, outputVideoStream->time_base.den);


AVPacket packet;
while (true) {
if (av_read_frame(videoFormatContext, &packet) >= 0) {
Expand All @@ -268,8 +331,8 @@ namespace ffmpeg {
audioFrame->format = AV_SAMPLE_FMT_FLTP;
audioFrame->ch_layout = AV_CHANNEL_LAYOUT_STEREO;

for (size_t i = 0; i < raw.size(); i += frameSize * channels) {
int samplesToEncode = std::min(frameSize, static_cast<int>((raw.size() - i) / channels));
for (size_t i = 0; i < resampled.size(); i += frameSize * channels) {
int samplesToEncode = std::min(frameSize, static_cast<int>((resampled.size() - i) / channels));

audioFrame->nb_samples = samplesToEncode;
audioFrame->pts = pts;
Expand All @@ -282,8 +345,8 @@ namespace ffmpeg {
}

for (int j = 0; j < samplesToEncode; ++j) {
reinterpret_cast<float*>(audioFrame->data[0])[j] = raw[i + j * channels];
reinterpret_cast<float*>(audioFrame->data[1])[j] = raw[i + j * channels + 1];
reinterpret_cast<float*>(audioFrame->data[0])[j] = resampled[i + j * channels];
reinterpret_cast<float*>(audioFrame->data[1])[j] = resampled[i + j * channels + 1];
}

if (avcodec_send_frame(audio_codec_context_encoder, audioFrame) < 0) {
Expand All @@ -296,8 +359,6 @@ namespace ffmpeg {
audioPacket.data = nullptr;
audioPacket.size = 0;

geode::log::debug("2 timebase {} {} {} {}", audio_codec_context_encoder->time_base.num, audio_codec_context_encoder->time_base.den, outputAudioStream->time_base.num, outputAudioStream->time_base.den);

while (true) {
int ret = avcodec_receive_packet(audio_codec_context_encoder, &audioPacket);
if (ret == 0) {
Expand Down

0 comments on commit 0bbadf2

Please sign in to comment.