• 需求說明：錄音后，需要有變聲選項谊却，選擇變聲模式試聽，滿意后可保存哑芹。保存的變聲文件格式為AMR

• 解決方案：錄音問題很好解決炎辨，每次錄音后保存一個文件，格式為WAV绩衷。變聲功能采用網上比較常見的fmod框架蹦魔〖ぢ剩可試聽可保存文件咳燕，文件格式為WAV。保存后進行文件轉格式,由WAV轉為amr乒躺。方案是采用Android自帶的AmrInputStream 完成招盲。

遇到的問題有：

• 變聲后文件如何保存。
• 變聲后文件轉格式后嘉冒，聲音被拉長兩倍曹货。

下面是具體的解決方法。

變聲采用fmod框架來實現(xiàn)讳推，具體方案可以在網上搜索顶籽，這里貼上cpp代碼：

#include "inc/fmod.hpp"
#include <stdlib.h>
#include <unistd.h>
#include  "com_kidosc_voicechange_jni_VoiceFixer.h"

#include <android/log.h>

#define LOGI(FORMAT, ...) __android_log_print(ANDROID_LOG_INFO,"zph",FORMAT,##__VA_ARGS__);
#define LOGE(FORMAT, ...) __android_log_print(ANDROID_LOG_ERROR,"zph",FORMAT,##__VA_ARGS__);

#define MODE_NORMAL 0
#define MODE_FUNNY 1
#define MODE_UNCLE 2
#define MODE_LOLITA 3
#define MODE_ROB0T 4
#define MODE_ETHEREAL 5
#define MODE_CHORUS 6
#define MODE_HORROR 7
using namespace FMOD;


Sound *sound;
DSP *dsp;
Channel *channel;
bool playing;
float frequency;
System *mSystem;
JNIEnv *mEnv;
void stopPlaying();

JNIEXPORT void JNICALL Java_com_kidosc_voicechange_jni_VoiceFixer_init(JNIEnv *env,
                                                                       jclass jcls){

}

JNIEXPORT void JNICALL Java_com_kidosc_voicechange_jni_VoiceFixer_fix(JNIEnv *env,
                                                                      jclass jcls, jstring path_jstr,
                                                                      jint type,jint save) {


    playing = true;
    frequency = 0;
    System_Create(&mSystem);
    mSystem->setSoftwareFormat(8000,FMOD_SPEAKERMODE_MONO,0); //設置采樣率為8000，channel為1
    mEnv=env;
    const char *path_cstr = mEnv->GetStringUTFChars(path_jstr, NULL);
    if(save==1){
        char cDest[200] = "sdcard/xxx.wav";
        mSystem->setOutput(FMOD_OUTPUTTYPE_WAVWRITER); //保存文件格式為WAV
        mSystem->init(32, FMOD_INIT_NORMAL | FMOD_INIT_PROFILE_ENABLE,cDest);
    } else{
        mSystem->init(32, FMOD_INIT_NORMAL,NULL);
    }
    try {
        //創(chuàng)建聲音
        mSystem->createSound(path_cstr, FMOD_DEFAULT, NULL, &sound);
        mSystem->playSound(sound, 0, false, &channel);
        switch (type) {
            case MODE_NORMAL:
                LOGI("%s", path_cstr);
                LOGI("%s", "fix normal");
                break;
            case MODE_FUNNY:
                mSystem->createDSPByType(FMOD_DSP_TYPE_NORMALIZE, &dsp);
                channel->getFrequency(&frequency);
                frequency = frequency * 1.6;
                channel->setFrequency(frequency);
                break;
            case MODE_UNCLE:
                mSystem->createDSPByType(FMOD_DSP_TYPE_PITCHSHIFT, &dsp);
                dsp->setParameterFloat(FMOD_DSP_PITCHSHIFT_PITCH, 0.8);
                channel->addDSP(0, dsp);
                break;
            case MODE_LOLITA:
                mSystem->createDSPByType(FMOD_DSP_TYPE_PITCHSHIFT, &dsp);
                dsp->setParameterFloat(FMOD_DSP_PITCHSHIFT_PITCH,
                                       1.8);
                channel->addDSP(0, dsp);
                break;
            case MODE_ROB0T:
                mSystem->createDSPByType(FMOD_DSP_TYPE_ECHO, &dsp);
                dsp->setParameterFloat(FMOD_DSP_ECHO_DELAY, 50);
                dsp->setParameterFloat(FMOD_DSP_ECHO_FEEDBACK, 60);
                channel->addDSP(0, dsp);
                break;
            case MODE_ETHEREAL:
                mSystem->createDSPByType(FMOD_DSP_TYPE_ECHO, &dsp);
                dsp->setParameterFloat(FMOD_DSP_ECHO_DELAY, 300);
                dsp->setParameterFloat(FMOD_DSP_ECHO_FEEDBACK, 20);
                channel->addDSP(0, dsp);
                break;
            case MODE_CHORUS:
                mSystem->createDSPByType(FMOD_DSP_TYPE_ECHO, &dsp);
                dsp->setParameterFloat(FMOD_DSP_ECHO_DELAY, 100);
                dsp->setParameterFloat(FMOD_DSP_ECHO_FEEDBACK, 50);
                channel->addDSP(0, dsp);
                break;
            case MODE_HORROR:
                mSystem->createDSPByType(FMOD_DSP_TYPE_TREMOLO, &dsp);
                dsp->setParameterFloat(FMOD_DSP_TREMOLO_SKEW, 0.8);
                channel->addDSP(0, dsp);
                break;
            default:
                break;
        }

    } catch (...) {
        LOGE("%s", "發(fā)生異常");
        goto end;
    }
    mSystem->update();
    while (playing) {
        usleep(1000);
        channel->isPlaying(&playing);
    }
    goto end;
    end:
    mEnv->ReleaseStringUTFChars(path_jstr, path_cstr);
    sound->release();
    mSystem->close();
    mSystem->release();

    jclass clazz = mEnv -> FindClass("com/kidosc/voicechange/jni/VoiceFixer");
    if(clazz == NULL){
        printf("not found com/kidosc/voicechange/jni/VoiceFixer class");
        return;
    }
    jmethodID  id = mEnv->GetStaticMethodID(clazz,"setPlayState","()V");
    if (id==NULL){
        printf("method not found");
        return;
    }
    mEnv->CallStaticVoidMethod(clazz,id);
    printf("env->CallStaticVoidMethod(clazz,id);");
}


JNIEXPORT void JNICALL Java_com_kidosc_voicechange_jni_VoiceFixer_stopPlay(JNIEnv *env,
                                                                           jclass jcls){
    stopPlaying();
}

void stopPlaying(){
    channel->stop();
    printf("stopplaying");
}

上面代碼與網上的區(qū)別是银觅，增加了變聲文件保存的功能礼饱，增加了停止變聲播放的功能，改變了默認輸出文件的采樣率和channel究驴。其中最后一條尤為重要setSoftwareFormat方法文檔

變聲臨時文件保存在sdcard/xxx.wav镊绪，可在保存完成后，再進行重新轉格式并保存在自己想要的位置洒忧。

轉格式使用的是Android自帶的AmrInputStream.java處理蝴韭。這個文件被隱藏了，想要直接使用的話熙侍，需要將文件拷貝到自己項目下榄鉴。具體如何使用可自行搜索。AmrInputStream轉格式蛉抓，默認的是8000khz,16bit,單channel庆尘。這些參數(shù)很重要，如果WAV格式與這個不匹配芝雪，轉格式后會出現(xiàn)聲音被拉長等一系列問題减余。所以需要將fmod變聲框架的輸出文件格式改成8000khz,16bit,單channel 。這才是最頭痛的地方惩系！網上根本找不到相關的內容位岔，沒辦法只能自己慢慢摸索了如筛。最開始的思路是將fmod輸出的格式改為8000，成功了之后發(fā)現(xiàn)聲音還是被拉長了兩倍抒抬。懷疑是16bit的問題杨刨，又不知道fmod輸出的是多少bit。所以嘗試更改AmrInputStream中擦剑，將其默認輸入的音頻數(shù)據(jù)更改一下妖胀。更改完之后，聲音都變質了惠勒。所以最終方案還是考慮更改fmod輸出音頻赚抡。找到一個setSoftwareFormat方法，發(fā)現(xiàn)可以更改輸出采樣率纠屋。那么現(xiàn)在的問題就是涂臣，為什么還是會被拉長兩倍。

    /**
     * @param inPath 源文件
     * @param outPath 目標文件
     */
    public void systemWav2Amr(String inPath,String outPath){
        try {
            FileOutputStream fileoutputStream = new FileOutputStream(outPath);
            InputStream inputStream =new FileInputStream(inPath);
            AmrInputStream amrInputStream =new AmrInputStream(inputStream);

            fileoutputStream.write(header);
            byte[] buf = new byte[1024];
            int len = 0;
            while ((len = amrInputStream.read(buf)) >0){
                fileoutputStream.write(buf,0,len);
            }
            fileoutputStream.close();
            amrInputStream.close();
        } catch (Exception e) {
            e.printStackTrace();
        }
    }

查找資料發(fā)現(xiàn)售担，采樣速率跟三個值有關赁遗，一個是channel，一個是采樣位數(shù)族铆，一個是采樣率岩四。既然被拉長兩倍，說明采樣速率變?yōu)橹暗囊话敫缛痢Ｕf明fmod輸出的音頻數(shù)據(jù)中剖煌，channel或者采樣位數(shù)中的一個是AmrInputStream的兩倍。最開始猜測是采樣位數(shù)為32bit導致的献丑。更改了AmrInputStream的輸入格式為32bit末捣。發(fā)現(xiàn)并沒有用，時間的確已經保持一致了创橄。說明思路是對的箩做，那么問題就是如何更改fmod輸出的音頻channel了。AmrInputStream中的channel為1妥畏，說明fmod中輸出的為2channel邦邦。fmod中在fmod_codec.h中找到channel相關屬性如下：

struct FMOD_CODEC_WAVEFORMAT
{
    char               name[256];     /* [w] Name of sound.  Optional. */
    FMOD_SOUND_FORMAT  format;        /* [w] Format for (decompressed) codec output, ie FMOD_SOUND_FORMAT_PCM8, FMOD_SOUND_FORMAT_PCM16.  Mandantory - Must be supplied. */
    int                channels;      /* [w] Number of channels used by codec, ie mono = 1, stereo = 2.  Mandantory - Must be supplied.  */
    int                frequency;     /* [w] Default frequency in hz of the codec, ie 44100.  Mandantory - Must be supplied.  */
    unsigned int       lengthbytes;   /* [w] Length in bytes of the source data.  Used for FMOD_TIMEUNIT_RAWBYTES.  Optional. Default = 0. */
    unsigned int       lengthpcm;     /* [w] Length in decompressed, PCM samples of the file, ie length in seconds * frequency.  Used for Sound::getLength and for memory allocation of static decompressed sample data.  Mandantory - Must be supplied. */
    unsigned int       pcmblocksize;  /* [w] Minimum, optimal number of decompressed PCM samples codec can handle.  0 or 1 = no buffering.  Anything higher means FMOD will allocate a PCM buffer of this size to read in chunks.  The codec read callback will be called in multiples of this value.  Optional.  */
    int                loopstart;     /* [w] Loopstart in decompressed, PCM samples of file. Optional. Default = 0. */
    int                loopend;       /* [w] Loopend in decompressed, PCM samples of file. Optional. Default = 0. */
    FMOD_MODE          mode;          /* [w] Mode to determine whether the sound should by default load as looping, non looping, 2d or 3d.  Optional. Default = FMOD_DEFAULT. */
    FMOD_CHANNELMASK   channelmask;   /* [w] Defined channel bitmask to describe which speakers the channels in the codec map to, in order of channel count.  See fmod_common.h.  Optional. Leave at 0 to map to the speaker layout_save defined in FMOD_SPEAKER. */
    FMOD_CHANNELORDER  channelorder;  /* [w] Defined channel order type, to describe where each sound channel should pan for the number of channels specified.  See fmod_common.h.  Optional.  Leave at 0 to play in default speaker order. */
    float              peakvolume;    /* [w] Peak volume of sound. Optional. Default = 0 if not used. */
};

在setSoftwareFormat方法中發(fā)現(xiàn)第二個參數(shù)可以設置為MONO，當其為FMOD_SPEAKERMODE_MONO時醉蚁，channels為1.

typedef enum
{
    FMOD_SPEAKERMODE_DEFAULT,          /* Default speaker mode based on operating system/output mode.  Windows = control panel setting, Xbox = 5.1, PS3 = 7.1 etc. */
    FMOD_SPEAKERMODE_RAW,              /* There is no specific speakermode.  Sound channels are mapped in order of input to output.  Use System::setSoftwareFormat to specify speaker count. See remarks for more information. */
    FMOD_SPEAKERMODE_MONO,             /* The speakers are monaural. */
    FMOD_SPEAKERMODE_STEREO,           /* The speakers are stereo. */
    FMOD_SPEAKERMODE_QUAD,             /* 4 speaker setup.    This includes front left, front right, surround left, surround right.  */
    FMOD_SPEAKERMODE_SURROUND,         /* 5 speaker setup.    This includes front left, front right, center, surround left, surround right. */
    FMOD_SPEAKERMODE_5POINT1,          /* 5.1 speaker setup.  This includes front left, front right, center, surround left, surround right and an LFE speaker. */
    FMOD_SPEAKERMODE_7POINT1,          /* 7.1 speaker setup.  This includes front left, front right, center, surround left, surround right, back left, back right and an LFE speaker. */
    
    FMOD_SPEAKERMODE_MAX,              /* Maximum number of speaker modes supported. */
    FMOD_SPEAKERMODE_FORCEINT = 65536  /* Makes sure this enum is signed 32bit. */
} FMOD_SPEAKERMODE;

如此設置之后燃辖，就可以完美的進行轉格式分享了。