speech_synthesis.py 文件源码

def ideal_binary_mask(noisy_speech, clean_speech, snr):
    noisy_spect = stft_extractor(noisy_speech, 320, 160, 'hanning')
    clean_spect = stft_extractor(clean_speech, 320, 160, 'hanning')
    spect_snr = np.power(np.abs(clean_spect), 2.0) / np.power(np.abs(noisy_spect - clean_spect), 2.0)
    ibm = np.where(spect_snr > 10**(0.1*snr), 1, 0)

    return ibm