data_utils.py 文件源码

def make_spectrum(self, filename, use_normalize):
        sr, y = wav.read(filename)
        if sr != 16000:
            raise ValueError('Sampling rate is expected to be 16kHz!')
        if y.dtype!='float32':
            y = np.float32(y/32767.)

        D=librosa.stft(y,n_fft=512,hop_length=256,win_length=512,window=scipy.signal.hamming)
        Sxx=np.log10(abs(D)**2) 
        if use_normalize:
            mean = np.mean(Sxx, axis=1).reshape((257,1))
            std = np.std(Sxx, axis=1).reshape((257,1))+1e-12
            Sxx = (Sxx-mean)/std  
        slices = []
        for i in range(0, Sxx.shape[1]-self.FRAMELENGTH, self.OVERLAP):
            slices.append(Sxx[:,i:i+self.FRAMELENGTH])
        return np.array(slices)