小白:我是标准的男中音,唱“李香兰”不输张学友,为什么我要变成女高音?

本文介绍,如何修改音频数据,控制音频的节奏、速率或音调。

大概的思路是这样的,先解码音频,得到pcm数据,再通过soundtouch来修改pcm数据,最后压缩为常见格式的音频。

对于音频编码格式之类的知识,可以参考之前同系列的文章。

先给出一个经过修改后的音频文件,可以听一下效果(如果这里可以上传并播放音频文件的话):

解码与编码部分,同样是FFmpeg的使用(之前多次介绍过了),得到pcm后再调用soundtouch。

演示demo的文件结构:
演示demo的目录结构

先上代码(change_pcm_pitch.cpp),之后再简单介绍soundtouch及它的调用:

extern "C" {
#include "ffmpeg/include/libavcodec/avcodec.h"
#include "ffmpeg/include/libavformat/avformat.h"
#include "ffmpeg/include/libswresample/swresample.h"
#include "ffmpeg/include/libavutil/samplefmt.h"
}
#include "SoundTouch.h"
using namespace soundtouch;

void change_pcm_pitch(const char* filepath) {
    av_register_all();
    av_log_set_level(AV_LOG_DEBUG);
    AVFormatContext* formatContext = avformat_alloc_context();
    AVCodecContext* codecContext = NULL;
    int status = 0;
    bool success = false;
    int audioindex = -1;
    status = avformat_open_input(&formatContext, filepath, NULL, NULL);
    if (status == 0) {
        status = avformat_find_stream_info(formatContext, NULL);
        if (status >= 0) {
            for (int i = 0; i < formatContext->nb_streams; i ++) {
                if (formatContext->streams[i]->codec->codec_type == AVMEDIA_TYPE_AUDIO) {
                    audioindex = i;
                    break;
                }   
            }
            if (audioindex > -1) {
                codecContext = formatContext->streams[audioindex]->codec;
                AVCodec* codec = avcodec_find_decoder(codecContext->codec_id);
                if (codec) {
                    status = avcodec_open2(codecContext, codec, NULL);
                    if (status == 0) {
                        success = true; 
                    }
                }
            }
        }
    }
    if (success) {
        av_dump_format(formatContext, 0, filepath, false);
        av_log(NULL, AV_LOG_DEBUG, "format and decoder sucessful, and now in decoding each frame\n");
        printf("sample_rate=%d, channels=%d\n", codecContext->sample_rate, codecContext->channels);
        SoundTouch* soundtouch = new SoundTouch();
        printf("soundtouch version=%s\n", soundtouch->getVersionString());
        soundtouch->setSampleRate(codecContext->sample_rate);
        soundtouch->setChannels(codecContext->channels);
        soundtouch->setTempo(0.5);  // tempo,播放节奏,1.0为正常节奏,大于1.0加快,小于1.0变慢,pcm的体积随之变化
        soundtouch->setRate(3.0);  // rate,播放速率,1.0为正常速度;单设置这个时,除了影响播放速度,还会影响到音调
        soundtouch->setPitch(0.5);   // pitch,音调,1.0为正常音调;这个设置并不会影响到时长
        AVFrame* frame = av_frame_alloc();
        SwrContext* swr = NULL;
        int gotframe = 0;
        char outfile[512] = {0};
        strcpy(outfile, filepath);
        strcat(outfile + strlen(outfile), ".pcm");
        FILE* file = fopen(outfile, "wb");
        if (file) {
            while (true) {
                AVPacket packet;
                av_init_packet(&packet);
                status = av_read_frame(formatContext, &packet);
                if (status < 0) {
                    if (status == AVERROR_EOF) {
                        av_log(NULL, AV_LOG_DEBUG, "read end for file\n");
                        break;
                    }
                    else {
                        av_packet_unref(&packet);
                    }
                }
                else {
                    if (packet.stream_index == audioindex) {
                        int srcCount = packet.size;
                        while (srcCount > 0) {
                            int decodedcount = avcodec_decode_audio4(codecContext, frame, &gotframe, &packet);
                            if (decodedcount < 0) {
                                av_log(NULL, AV_LOG_DEBUG, "decode failed, perhaps not enough data\n");
                                break;
                            }
                            if (gotframe > 0) {
                                // resample 
                                int targetchannel = 2;
                                int targetsrate = 44100;
                                int targetfmt = AV_SAMPLE_FMT_S16;
                                bool needresample = false;
                                if (av_frame_get_channels(frame) != targetchannel || frame->sample_rate != targetsrate || frame->format != targetfmt) {
                                    needresample = true;    
                                }
                                if (needresample) {
                                    if (swr == NULL) {
                                        uint64_t in_channel_layout = av_get_default_channel_layout(av_frame_get_channels(frame));
                                        uint64_t out_channel_layout = av_get_default_channel_layout(targetchannel);
                                        int inSamplerate = frame->sample_rate;
                                        swr = swr_alloc_set_opts(NULL,
                                                out_channel_layout, (enum AVSampleFormat )AV_SAMPLE_FMT_S16, targetsrate,
                                                in_channel_layout, (enum AVSampleFormat)frame->format, inSamplerate, 0, NULL);
                                        int ret = swr_init(swr);
                                        if (ret != 0) {
                                            printf("swr_init failed: ret=%d\n", ret);
                                        }
                                    }
                                    if (swr) {
                                        if (frame->extended_data && frame->data[0] && frame->linesize[0] > 0) {
                                            int out_size = av_samples_get_buffer_size(NULL, targetchannel, frame->nb_samples, (enum AVSampleFormat)targetfmt, 0);
                                            void* out_buffer = av_malloc(out_size);
                                            if (out_buffer) {
                                                int convertSamples = swr_convert(swr, (uint8_t**)(&out_buffer), frame->nb_samples, 
                                                        (const uint8_t**)frame->extended_data, frame->nb_samples);
                                                int len = convertSamples * targetchannel * av_get_bytes_per_sample((enum AVSampleFormat)targetfmt);
                                                int samplecount = convertSamples;
                                                soundtouch->putSamples((SAMPLETYPE*)out_buffer, samplecount);
                                                int bufsize = samplecount * frame->channels * sizeof(short);
                                                unsigned char* buf = (unsigned char*)malloc(bufsize);
                                                int gotsamplecount = soundtouch->receiveSamples((SAMPLETYPE*)buf, samplecount);
                                                printf("soundtouch receiveSamples after resample:gotsamplecount=%d bufsize=%d sizeof(SAMPLETYPE)=%lu\n", gotsamplecount, bufsize, sizeof(SAMPLETYPE));
                                                if (gotsamplecount) {
                                                    fwrite(buf, gotsamplecount * frame->channels * sizeof(short), 1, file);
                                                }
                                                free(buf);  
                                                av_free(out_buffer);
                                            }
                                        }
                                    }
                                }
                                else {
                                    int samplecount = frame->nb_samples;
                                    soundtouch->putSamples((SAMPLETYPE*)frame->data[0], samplecount);
                                    int bufsize = samplecount * frame->channels * sizeof(short);
                                    unsigned char* buf = (unsigned char*)malloc(bufsize);
                                    int gotsamplecount = soundtouch->receiveSamples((SAMPLETYPE*)buf, samplecount);
                                    printf("soundtouch receiveSamples:gotsamplecount=%d bufsize=%d sizeof(SAMPLETYPE)=%lu\n", gotsamplecount, bufsize, sizeof(SAMPLETYPE));
                                    if (gotsamplecount) {
                                        fwrite(buf, gotsamplecount * frame->channels * sizeof(short), 1, file);
                                    }
                                    free(buf);  
                                }
                            }
                            srcCount -= decodedcount;
                        }
                    }
                }
                av_packet_unref(&packet);
            }   
            fclose(file);
        }
        av_frame_free(&frame);
        delete soundtouch;
        if (swr) {
            swr_free(&swr);
        }
    }
    avformat_free_context(formatContext);
}

// 保证这个ffmepg支持mp3编码即可(使用lamemp3),当然也可以编码成其它格式
// 我有多个不同特性的ffmpeg,这里指定一个能编码mp3的ffmpeg
const char* FFMPEGEXE = "/usr/local/Cellar/ffmpeg/2.6.2/bin/ffmpeg";  
const int SAMPLE_RATE = 44100;
const int CHANNELS = 2;
const int BITRATE = 128;
const int BUF_LEN = 1024;

void encode(const char* srcfile, const char* outfile) {
    char buf[BUF_LEN] = {0};
    sprintf(buf, "%s -ar %d -ac %d -f s16le -i %s -ar %d -ac %d -b:a %dK -y %s", FFMPEGEXE, SAMPLE_RATE, CHANNELS, srcfile, SAMPLE_RATE, CHANNELS, BITRATE, outfile);
    system(buf);
}

int main(int argc, const char *argv[])
{
    const char filepath[] = "test2.mp3";    
    change_pcm_pitch(filepath);  // xxx.xx.pcm create
    encode("test2.mp3.pcm", "out.mp3");

    return 0;
}

soundtouch,一个开源的音效处理项目(c++代码),可以用更改音频的音调、播放速率、节拍等特征。

上面的demo直接使用了soundtouch的源码来实现音效变化。

soundtouch初始化:
soundtouch初始化

soundtouch作用于pcm:
soundtouch处理pcm

需要注意,soundtouch并没有解码功能,它假设调用层已经有pcm数据。

对于复杂的音效处理,必定会更耗时,对于在线实时的播放,音效的性能是要关注的因素,避免音效太耗时而导致播放卡顿。

小白:我听了处理后的“李香兰”,表示很满意,记得来听我的演唱会。

花满楼:那你要关注我们,并及时在群里面分布信息了!