音频重采样到双声道
代码未经测试，只是偶尔看到，做个备份，ffmpeg也自带有转音频声道的
extern "C" {
#include <libavcodec/avcodec.h>
#include <libavformat/avformat.h>
#include <libswresample/swresample.h>
}
// 这里是PortAudio的头文件
#include <portaudio.h>
#include <assert.h>
#include <iostream>
struct AudioContext {
AVCodecContext* codecContext;
SwrContext* swrContext;
ReSampleContext* resamplerContext;
};
static
void audio_copy(AudioContext *context, AVFrame *dst, AVFrame* src)
{
int nb_sample;
int dst_buf_size;
int out_channels;
int bytes_per_sample = 0;
dst->linesize[0] = src->linesize[0];
*dst = *src;
dst->data[0] = NULL;
dst->type = 0;
/* 备注: FFMIN(codecContext->channels, 2); 会有问题, 因为swr_alloc_set_opts的out_channel_layout参数. */
out_channels = context->codecContext->channels;
bytes_per_sample = av_get_bytes_per_sample(context->codecContext->sample_fmt);
/* 备注: 由于 src->linesize[0] 可能是错误的, 所以计算得到的nb_sample会不正确, 直接使用src->nb_samples即可. */
nb_sample = src->nb_samples;/* src->linesize[0] / codecContext->channels / bytes_per_sample; */
bytes_per_sample = av_get_bytes_per_sample(AV_SAMPLE_FMT_S16);
dst_buf_size = nb_sample * bytes_per_sample * out_channels;
dst->data[0] = (uint8_t*) av_malloc(dst_buf_size);
assert(dst->data[0]);
avcodec_fill_audio_frame(dst, out_channels, AV_SAMPLE_FMT_S16, dst->data[0], dst_buf_size, 0);
/* 重采样到AV_SAMPLE_FMT_S16格式. */
if (context->codecContext->sample_fmt != AV_SAMPLE_FMT_S16)
{
if (!context->swrContext)
{
uint64_t in_channel_layout = av_get_default_channel_layout(context->codecContext->channels);
uint64_t out_channel_layout = av_get_default_channel_layout(out_channels);
context->swrContext = swr_alloc_set_opts(NULL,
out_channel_layout, AV_SAMPLE_FMT_S16, context->codecContext->sample_rate,
in_channel_layout, context->codecContext->sample_fmt, context->codecContext->sample_rate,
0, NULL);
swr_init(context->swrContext);
}
if (context->swrContext)
{
int ret, out_count;
out_count = dst_buf_size / out_channels / av_get_bytes_per_sample(AV_SAMPLE_FMT_S16);
ret = swr_convert(context->swrContext, dst->data, out_count, const_cast<const uint8_t**>(src->data), nb_sample);
if (ret < 0)
assert(0);
src->linesize[0] = dst->linesize[0] = ret * av_get_bytes_per_sample(AV_SAMPLE_FMT_S16) * out_channels;
memcpy(src->data[0], dst->data[0], src->linesize[0]);
}
}
/* 重采样到双声道. */
if (context->codecContext->channels > 2)
{
if (!context->resamplerContext)
{
context->resamplerContext = av_audio_resample_init(
FFMIN(2, context->codecContext->channels),
context->codecContext->channels, context->codecContext->sample_rate,
context->codecContext->sample_rate, AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_S16,
16, 10, 0, 0.f);
}
if (context->resamplerContext)
{
int samples = src->linesize[0] / (av_get_bytes_per_sample(AV_SAMPLE_FMT_S16) * context->codecContext->channels);
dst->linesize[0] = audio_resample(context->resamplerContext,
(short *) dst->data[0], (short *) src->data[0], samples) * 4;
}
}
else
{
dst->linesize[0] = dst->linesize[0];
memcpy(dst->data[0], src->data[0], dst->linesize[0]);
}
}
int main(int argc, char* argv[])
{
// 将要打开的音频文件(视频文件也可以支持).
const char* filename = argc > 1 ? argv[1] : "1.mp3";
// 初始化libavformat,并注册所有的模块
av_register_all();
// 这里一定要设置成NULL, 或者调用avformat_alloc_context分配内存, 否则可能崩溃.
AVFormatContext *formatContext = NULL;
// 打开输入文件.
if( avformat_open_input(&formatContext, filename, NULL, NULL) < 0) {
std::cerr << "cannot open file" << std::endl;
return -1;
}
// 探测文件里面的音视频流信息.
if( avformat_find_stream_info(formatContext, NULL) < 0) {
std::cerr << "cannot find stream info" << std::endl;
return -1;
}
// 输出来看看.
av_dump_format(formatContext, 0, 0, 0);
// 找到音频流的索引(如果是视频的话,可能存在多个流).
int audioIndex;
if((audioIndex = av_find_best_stream(formatContext, AVMEDIA_TYPE_AUDIO, 0, 0, NULL, 0)) < 0) {
std::cerr << "cannot find audio stream" << std::endl;
return -1;
}
AVCodecContext *codecContext = formatContext->streams[audioIndex]->codec;
// 找到解码器.
AVCodec *codec = avcodec_find_decoder(codecContext->codec_id);
if(codec == NULL) {
std::cerr << "cannot find decoder for " << codecContext->codec_name << std::endl;
}
// 打开解码器.
if( avcodec_open2(codecContext, codec, NULL) < 0) {
std::cerr << "cannot open decoder" << std::endl;
return -1;
}
// AVPacket是解码前的数据, AVFrame是解码后的数据.
AVPacket packet;
AVFrame *frame = avcodec_alloc_frame();
int got;
AudioContext context;
context.resamplerContext = NULL;
context.swrContext = NULL;
context.codecContext = codecContext;
// 下面是初始化PortAudio, 用PortAudio的Blocking API比较简单.
PaStream *stream;
Pa_Initialize();
Pa_OpenDefaultStream(&stream, 0, codecContext->channels,
paInt16, codecContext->sample_rate,
1024, NULL, NULL);
Pa_StartStream(stream);
codecContext->sample_fmt;
int size = 4092;
uint8_t* buffer = new uint8_t[size * 2];
int channels = codecContext->channels;
while(true) {
// 从文件中读取一帧.
if(av_read_frame(formatContext, &packet) < 0) {
// 文件读完了.
break;
}
// 解码.
if( avcodec_decode_audio4(codecContext, frame, &got, &packet) < 0) {
std::cerr << "cannot decode" << std::endl;
// 偶尔会出错,一般都可以原谅的...
// break;
}
// 解码出来了一帧
if(got) {
// 因为frame->data[0]表示的是左声道LLL....,frame->data[1]表示右声道RRR...
// 而PortAudio要求的是LRLRLR....这样的数据排布, 所以这里用循环重新将数据复制到buffer中
AVFrame *dst = avcodec_alloc_frame();
audio_copy(&context, dst, frame);
Pa_WriteStream(stream, reinterpret_cast<int16_t*>(dst->data[0]), dst->nb_samples);
}
}
delete buffer;
return 0;
}