objective-c - iOS でオーディオ フィンガープリントを作成するためのフレームワーク FFT の高速化

Question

まず、私の下手な英語で申し訳ありません。数週間前に Objective-C を使い始めたばかりです。2 つの iOS デバイスから録音された 2 つのオーディオ信号を比較する必要があるプロジェクトを行っています。これまでのところ、iPhone 4s と iPhone 4 から 2 つの .aif ファイルを記録することができました。次に、次のアルゴリズム「高度に堅牢なオーディオ フィンガープリンティング システム」を適用してみます。by: Jaap Haitsma" を使用して 2 つのフィンガープリント (バイナリ ビット パターン 101011010) を取得し、それらをビットごとに比較します。しかし、これまでのところ、得られた結果は 45% から 55% の間であり、ほぼランダムな確率です。 0 と 1 の間. だから誰かが私に何かアドバイスを与えることができます. これまでのコードは次のとおりです:

CalculateFingerprint *myCalculateFingerprint = [CalculateFingerprint alloc];    
SInt16 *inputBuffer;    


    path4 = [documentsDirectory stringByAppendingPathComponent:fileName4];


    /////////Calculate for the 4 file
    fileURL = [NSURL fileURLWithPath:path4];   

    status = AudioFileOpenURL((__bridge CFURLRef)fileURL, kAudioFileReadPermission,kAudioFileAIFFType, &myAudioFile);

    status = AudioFileGetPropertyInfo(myAudioFile, 
                                      kAudioFilePropertyAudioDataPacketCount, 
                                      &propertySizeDataPacketCount, 
                                      &writabilityDataPacketCount);    

    status = AudioFileGetProperty(myAudioFile, 
                                  kAudioFilePropertyAudioDataPacketCount, 
                                  &propertySizeDataPacketCount, 
                                  &numberOfPackets);    

    status = AudioFileGetPropertyInfo (myAudioFile, 
                                       kAudioFilePropertyMaximumPacketSize, 
                                       &propertySizeMaxPacketSize, 
                                       &writabilityMaxPacketSize);

    status = AudioFileGetProperty(myAudioFile, 
                                  kAudioFilePropertyMaximumPacketSize, 
                                  &propertySizeMaxPacketSize, 
                                  &maxPacketSize);

    inputBuffer = (SInt16 *)malloc(numberOfPackets * maxPacketSize);

    currentPacket = 0;
    status = AudioFileReadPackets(myAudioFile, 
                                  false, &numberOfBytesRead, 
                                  NULL, 
                                  currentPacket, 
                                  &numberOfPackets, 
                                  inputBuffer); 



    [myCalculateFingerprint calculateFingerprint:inputBuffer sampleCount:numberOfPackets index:indexFile];


    status = AudioFileClose(myAudioFile);

指紋コードの計算は次のとおりです。

-(void)calculateFingerprint :(SInt16*)samples
             sampleCount:(int)sampleCount
            index:(int)indexFile{
//Divide the audio signal into 32 frames
frames myFrames [32];
int stepFrames = sampleCount / 62;

int number = 0;
int index ;
for (int i = 0; i < 32; ++i){
    index = 0;
    myFrames[i].start = number;
    myFrames[i].end = number + (32*stepFrames);
    myFrames[i].dataFrames = (SInt16*)malloc((myFrames[i].end -number+1)*sizeof(SInt16));        
    for (int j = number;j<=myFrames[i].end; ++j){

        myFrames[i].dataFrames[index] = samples[j];
        ++index;
    }
    number = number + stepFrames;
}


//Calculate FFT for each of the audio signal frames.
CalculateFFT *myCalculateFFT = [[CalculateFFT alloc] init];

theFFT myFFTData [32];
for (int i = 0; i <32; ++i){
    myFFTData[i].FFTdata = [myCalculateFFT calculateFFTForData:myFrames[i].dataFrames];
}

//each index represent the frequency as followed:
// index i is frequency i * 44100/1024
//We only need 33 bands from 300 Hz to 2000Hz, so we will get the FFTdata from the index 7 to 40
float energy [33][33];

for (int i =0; i < 33; ++i){
    energy[0][i] = 0;
}
int stepBand;
for (int i = 1; i < 33; ++i){
    for (int j = 0; j < 33; ++j){

        energy[i][j] = myFFTData[i].FFTdata[j+7];

    }

}


//next we calculate the bits for the audio fingerprint
Float32 check  = 0;
int fingerPrint [32][32];
NSMutableString *result = [[NSMutableString alloc]init];
for (int i = 0; i < 32; ++i){
    for (int j = 0; j <32; ++j){
        check = energy[i+1][j] -energy[i+1][j+1] -energy[i][j] +energy[i][j+1];

        if (check > 0){
            fingerPrint[i][j] = 1;//[tempBitFingerPrint addObject:[NSNumber numberWithInt:1]];
        }else {
            fingerPrint[i][j] = 0;//[tempBitFingerPrint addObject:[NSNumber numberWithInt:0]];
        }
[result appendString:[NSString stringWithFormat:@"%d",fingerPrint[i][j]]];
    }

}

最後に FFT 計算コード:

-(void)FFTSetup{
UInt32 maxFrames = 1024;
originalReal = (float*) malloc(maxFrames*sizeof(float));
originalRealTransfer = (float*)malloc(maxFrames*sizeof(float));
obtainedReal = (float*) malloc(maxFrames *sizeof(float));
freqArray = (Float32*) malloc((maxFrames/2) *sizeof(Float32));
fftLog2n = log2f(maxFrames);
fftN = 1 << fftLog2n;
fftNOver2 = maxFrames/2;
fftBufferCapacity = maxFrames;
fftIndex = 0;
fftA.realp = (float*)malloc(fftNOver2*sizeof(float));
fftA.imagp = (float*)malloc(fftNOver2*sizeof(float));
fftSetup = vDSP_create_fftsetup(fftLog2n,FFT_RADIX2);

}

-(Float32*) calculateFFTForData:(SInt16*)sampleData { [self FFTSetup];

int stride = 1;
for (int i = 0; i < fftN; ++i){
    originalReal[i] = (float) sampleData[i];
}

UInt32 maxFrames = 1024;
//Apply Hann window on the data
int windowSize = maxFrames;

float * window = (float*)malloc(sizeof(float)*windowSize);
memset(window, 0, sizeof(float)*windowSize);
vDSP_hann_window(window, windowSize, vDSP_HANN_NORM); 
vDSP_vmul(originalReal,1,window,1,originalRealTransfer,1,windowSize);

vDSP_ctoz((COMPLEX*) originalRealTransfer,2,&fftA,1,fftNOver2);
vDSP_fft_zrip(fftSetup,&fftA, stride,fftLog2n,FFT_FORWARD);

float scale = (float) 1.0 /(2*fftN);
vDSP_vsmul(fftA.realp,1,&scale,fftA.realp,1,fftNOver2);
vDSP_vsmul(fftA.imagp,1,&scale,fftA.imagp,1,fftNOver2);

vDSP_ztoc(&fftA,1,(COMPLEX*)obtainedReal,2,fftNOver2);
int index = 0;
NSMutableString *testResult = [[NSMutableString alloc]init];
for (int i = 0; i < fftN; i=i+2){
    freqArray[index] = (obtainedReal[i]*obtainedReal[i])+(obtainedReal[i+1]*obtainedReal[i+1]);
    [testResult appendString:[NSString stringWithFormat:@"%f ",freqArray[index]]];
    ++index;
}


 return freqArray;

}