Python librosa.stft方法代码示例

# 需要导入模块: import librosa [as 别名]
# 或者: from librosa import stft [as 别名]
def test_compute(self):
        test_file_path = resources.sample_wav_file('wav_1.wav')
        y, sr = librosa.load(test_file_path, sr=None)
        frames = librosa.util.frame(y, frame_length=2048, hop_length=1024).T

        # EXPECTED
        S = np.abs(librosa.stft(y, center=False, n_fft=2048, hop_length=1024)) ** 2
        S = librosa.feature.melspectrogram(S=S, n_mels=128, sr=sr)
        S = librosa.power_to_db(S)
        onsets = librosa.onset.onset_strength(S=S, center=False)
        exp_tgram = librosa.feature.tempogram(onset_envelope=onsets, sr=sr, win_length=11, center=True).T

        # ACTUAL
        tgram_step = pipeline.Tempogram(win_length=11)
        tgrams = tgram_step.process_frames(frames, sr, last=True)

        assert np.allclose(tgrams, exp_tgram) 

# 需要导入模块: import librosa [as 别名]
# 或者: from librosa import stft [as 别名]
def test_compute_online(self):
        # Data: 41523 samples, 16 kHz
        # yields 40 frames with frame-size 2048 and hop-size 1024
        test_file_path = resources.sample_wav_file('wav_1.wav')
        y, sr = librosa.load(test_file_path, sr=None)

        # EXPECTED
        y_pad = np.pad(y, (0, 1024), mode='constant', constant_values=0)
        S = np.abs(librosa.stft(y_pad, center=False, n_fft=2048, hop_length=1024)) ** 2
        S = librosa.feature.melspectrogram(S=S, n_mels=128, sr=sr)
        S = librosa.power_to_db(S)
        onsets = librosa.onset.onset_strength(S=S, center=False)
        exp_tgram = librosa.feature.tempogram(onset_envelope=onsets, sr=sr, win_length=4, center=True).T

        # ACTUAL
        test_file = tracks.FileTrack('idx', test_file_path)
        tgram_step = pipeline.Tempogram(win_length=4)
        tgram_gen = tgram_step.process_track_online(test_file, 2048, 1024, chunk_size=5)

        chunks = list(tgram_gen)
        tgrams = np.vstack(chunks)

        assert np.allclose(tgrams, exp_tgram) 

# 需要导入模块: import librosa [as 别名]
# 或者: from librosa import stft [as 别名]
def test_compute_cleanup_after_one_utterance(self):
        test_file_path = resources.sample_wav_file('wav_1.wav')
        y, sr = librosa.load(test_file_path, sr=None)
        frames = librosa.util.frame(y, frame_length=2048, hop_length=1024).T

        # EXPECTED
        S = np.abs(librosa.stft(y, center=False, n_fft=2048, hop_length=1024)) ** 2
        S = librosa.feature.melspectrogram(S=S, n_mels=128, sr=sr)
        S = librosa.power_to_db(S)
        onsets = librosa.onset.onset_strength(S=S, center=False)
        exp_tgram = librosa.feature.tempogram(onset_envelope=onsets, sr=sr, win_length=11, center=True).T

        # ACTUAL
        tgram_step = pipeline.Tempogram(win_length=11)

        # FIRST RUN
        tgrams = tgram_step.process_frames(frames, sr, last=True)

        assert np.allclose(tgrams, exp_tgram)

        # SECOND RUN
        tgrams = tgram_step.process_frames(frames, sr, last=True)

        assert np.allclose(tgrams, exp_tgram) 

# 需要导入模块: import librosa [as 别名]
# 或者: from librosa import stft [as 别名]
def test_compute(self):
        test_file_path = resources.sample_wav_file('wav_1.wav')
        y, sr = librosa.load(test_file_path, sr=None)
        frames = librosa.util.frame(y, frame_length=2048, hop_length=1024).T

        # EXPECTED
        S = np.abs(librosa.stft(y, center=False, n_fft=2048, hop_length=1024)) ** 2
        S = librosa.feature.melspectrogram(S=S, n_mels=128, sr=sr)
        S = librosa.power_to_db(S)
        exp_onsets = librosa.onset.onset_strength(S=S, center=False).T
        exp_onsets = exp_onsets.reshape(exp_onsets.shape[0], 1)

        # ACTUAL
        onset = pipeline.OnsetStrength()
        onsets = onset.process_frames(frames, sr, last=True)

        assert np.allclose(onsets, exp_onsets) 

# 需要导入模块: import librosa [as 别名]
# 或者: from librosa import stft [as 别名]
def test_compute_online(self):
        test_file_path = resources.sample_wav_file('wav_1.wav')
        y, sr = librosa.load(test_file_path, sr=None)

        # EXPECTED
        y_pad = np.pad(y, (0, 1024), mode='constant', constant_values=0)
        S = np.abs(librosa.stft(y_pad, center=False, n_fft=2048, hop_length=1024)) ** 2
        S = librosa.feature.melspectrogram(S=S, n_mels=128, sr=sr)
        S = librosa.power_to_db(S)
        exp_onsets = librosa.onset.onset_strength(S=S, center=False).T
        exp_onsets = exp_onsets.reshape(exp_onsets.shape[0], 1)

        # ACTUAL
        test_file = tracks.FileTrack('idx', test_file_path)
        onset = pipeline.OnsetStrength()
        onset_gen = onset.process_track_online(test_file, 2048, 1024, chunk_size=5)

        chunks = list(onset_gen)
        onsets = np.vstack(chunks)

        print(onsets.shape, exp_onsets.shape)

        assert np.allclose(onsets, exp_onsets) 

# 需要导入模块: import librosa [as 别名]
# 或者: from librosa import stft [as 别名]
def wav_to_spec_batch(wavs, n_fft, hop_length = None):

    # Short-time Fourier transform (STFT) for wav matrix in batch
    # n_fft : int > 0 [scalar] FFT window size.
    # hop_length : int > 0 [scalar] number audio of frames between STFT columns. If unspecified, defaults win_length / 4.

    assert (wavs.ndim == 2), 'Single wav uses librosa.stft() directly'

    stft_matrices = list()

    for wav in wavs:
        stft_matrix = librosa.stft(wav, n_fft = n_fft, hop_length = hop_length)
        stft_matrices.append(stft_matrix)

    stft_matrices = np.array(stft_matrices)

    return stft_matrices 

# 需要导入模块: import librosa [as 别名]
# 或者: from librosa import stft [as 别名]
def get_spectrogram(y,
                    n_fft=1024,
                    win_length_samples=0.04,
                    hop_length_samples=0.02,
                    window=scipy.signal.hamming(1024, sym=False),
                    center=True,
                    spectrogram_type='magnitude',
                    params_extract=None):

    if spectrogram_type == 'power':
        return np.abs(librosa.stft(y + params_extract.get('eps'),
                                   n_fft=n_fft,
                                   win_length=win_length_samples,
                                   hop_length=hop_length_samples,
                                   center=center,
                                   window=window)) ** 2 

# 需要导入模块: import librosa [as 别名]
# 或者: from librosa import stft [as 别名]
def get_stft(y,sr,window_size=None, window_shift=None):
    if window_size is None:
        n_fft = int(0.025*sr)
    else:
        n_fft = int(window_size*0.001*sr)
    if window_shift is None:
        hop_length = int(0.010*sr)
    else:
        hop_length = int(window_shift*0.001*sr)
    stft = np.abs(librosa.stft(y,n_fft=n_fft,hop_length=hop_length)) #comes in complex numbers.. have to take absolute value
    stft = np.transpose(stft)
    stft -= (np.mean(stft, axis=0) + 1e-8)
    
    return stft

#super experimental. I wanted fundamental frequency but this was easier 

# 需要导入模块: import librosa [as 别名]
# 或者: from librosa import stft [as 别名]
def apply_noise(y,sr,wavefile):
    #at random apply varying amounts of environment noise
    rand_scale = random.choice([0.0,0.25,0.5,0.75])
    #rand_scale = 0.75
    if rand_scale > 0.0:
        total_length = len(y)/sr
        y_noise,sr = librosa.load(wavefile,sr=16000)
        envnoise_normalized = prep_data_vad_noise.normalize(y_noise)
        envnoise_scaled = prep_data_vad_noise.scale_noise(envnoise_normalized,rand_scale)
        envnoise_matched = prep_data_vad_noise.match_length(envnoise_scaled,sr,total_length)
        if len(envnoise_matched) != len(y):
            diff = int(len(y) - len(envnoise_matched))
            if diff < 0:
                envnoise_matched = envnoise_matched[:diff]
            else:
                envnoise_matched = np.append(envnoise_matched,np.zeros(diff,))
        y += envnoise_matched

    return y

#collects the actual features, according to the settings assigned
#such as with noise, voice activity detection/beginning silence removal, etc.
#mfcc, fbank, stft, delta, dom_freq 

# 需要导入模块: import librosa [as 别名]
# 或者: from librosa import stft [as 别名]
def griffin_lim(spectrogram):
    '''Applies Griffin-Lim's raw.'''
    X_best = copy.deepcopy(spectrogram)
    for i in range(hp.n_iter):
        X_t = invert_spectrogram(X_best)
        est = librosa.stft(X_t, hp.n_fft, hp.hop_length, win_length=hp.win_length)
        phase = est / np.maximum(1e-8, np.abs(est))
        X_best = spectrogram * phase
    X_t = invert_spectrogram(X_best)
    y = np.real(X_t)

    return y 

# 需要导入模块: import librosa [as 别名]
# 或者: from librosa import stft [as 别名]
def _stft(y):
    return librosa.stft(y=y, n_fft=hparams.fft_size, hop_length=get_hop_size()) 

# 需要导入模块: import librosa [as 别名]
# 或者: from librosa import stft [as 别名]
def plot_spectrum(sound_files, sound_names):
    """plot log power spectrum"""
    i = 1
    fig = plt.figure(figsize=(20, 64))
    for f, n in zip(sound_files, sound_names):
        y, sr = librosa.load(os.path.join('./data/esc10/audio/', f))
        plt.subplot(10, 1, i)
        D = librosa.logamplitude(np.abs(librosa.stft(y)) ** 2, ref_power=np.max)
        librosa.display.specshow(D, sr=sr, y_axis='log')
        plt.title(n + ' - ' + 'Spectrum')

        i += 1

    plt.tight_layout(pad=10)
    plt.show() 

# 需要导入模块: import librosa [as 别名]
# 或者: from librosa import stft [as 别名]
def griffin_lim(spectrogram):
    '''Applies Griffin-Lim's raw.
    '''
    X_best = copy.deepcopy(spectrogram)
    for i in range(hp.n_iter):
        X_t = invert_spectrogram(X_best)
        est = librosa.stft(X_t, hp.n_fft, hp.hop_length, win_length=hp.win_length)
        phase = est / np.maximum(1e-8, np.abs(est))
        X_best = spectrogram * phase
    X_t = invert_spectrogram(X_best)
    y = np.real(X_t)

    return y 

# 需要导入模块: import librosa [as 别名]
# 或者: from librosa import stft [as 别名]
def wavs_to_specs(wavs, n_fft = 1024, hop_length = None):

    stfts = list()
    for wav in wavs:
        stft = librosa.stft(wav, n_fft = n_fft, hop_length = hop_length)
        stfts.append(stft)

    return stfts 

# 需要导入模块: import librosa [as 别名]
# 或者: from librosa import stft [as 别名]
def compute_spec(audio_file,spectro_file):
	# Get actual audio
	audio, sr = librosa.load(audio_file, sr=config['resample_sr'])
	# Compute spectrogram
	if config['spectrogram_type']=='cqt':
		spec = librosa.cqt(audio, sr=sr, hop_length=config['hop'], n_bins=config['cqt_bins'])
	elif config['spectrogram_type']=='mel':
		spec = librosa.feature.melspectrogram(y=audio, sr=sr, hop_length=config['hop'],n_fft=config['n_fft'],n_mels=config['n_mels'])
	elif config['spectrogram_type']=='stft':
		spec = librosa.stft(y=audio,n_fft=config['n_fft'])
	# Write results:
	with open(spectro_file, "w") as f:
		pickle.dump(spec, f, protocol=-1) # spec shape: MxN.