Python librosa.resample方法代碼示例

# 需要導入模塊: import librosa [as 別名]
# 或者: from librosa import resample [as 別名]
def doFileStuff(line,isSlow):
    myobj = gTTS(text=line, lang='en', slow=isSlow) 
    myobj.save("placeholder.mp3")
    
    y, sr = librosa.load("placeholder.mp3")
    data = librosa.resample(y, sr, SAMPLE_RATE)
    librosa.output.write_wav('placeholder.wav', data, SAMPLE_RATE)
    d, sr = sf.read('placeholder.wav')
    sf.write('placeholder.wav', d, sr)

    y, sr = librosa.load("placeholder.mp3")
    lowData = librosa.resample(y, sr, SAMPLE_RATE*LOW_FACTOR)
    librosa.output.write_wav('lowPlaceholder.wav', lowData, SAMPLE_RATE)
    d, sr = sf.read('lowPlaceholder.wav')
    sf.write('lowPlaceholder.wav', d, sr)

    return data 

# 需要導入模塊: import librosa [as 別名]
# 或者: from librosa import resample [as 別名]
def sample_rate(self):
        """
        ``int``
            Sample rate associated with this object. If audio was read from a file, the sample
            rate will be set to the sample rate associated with the file. If this object was
            initialized from an array then the sample rate is set upon init. This property is
            read-only. To change the sample rate, use :func:`resample`.

        Notes:
            This property is read-only and cannot be set directly. To change

        See Also:
            * :func:`resample` to change the sample rate and resample data in :attr:`sample_rate`.

            * :func:`load_audio_from_array` to read audio from an array and set the sample rate.

            * :var:`nussl.constants.DEFAULT_SAMPLE_RATE` the default sample rate for *nussl*
                if not specified
        """
        return self._sample_rate 

# 需要導入模塊: import librosa [as 別名]
# 或者: from librosa import resample [as 別名]
def transform(self, y, sr):
        '''Transform an audio signal

        Parameters
        ----------
        y : np.ndarray
            The audio signal

        sr : number > 0
            The native sampling rate of y

        Returns
        -------
        dict
            Data dictionary containing features extracted from y

        See Also
        --------
        transform_audio
        '''
        if sr != self.sr:
            y = resample(y, sr, self.sr)

        return self.merge([self.transform_audio(y)]) 

# 需要導入模塊: import librosa [as 別名]
# 或者: from librosa import resample [as 別名]
def read_audio(path, target_fs=None):
    """Read 1 dimension audio sequence from given path. 
    
    Args:
      path: string, path of audio. 
      target_fs: int, resampling rate. 
      
    Returns:
      audio: 1 dimension audio sequence. 
      fs: sampling rate of audio. 
    """
    (audio, fs) = soundfile.read(path)
    if audio.ndim > 1:
        audio = np.mean(audio, axis=1)
    if target_fs is not None and fs != target_fs:
        audio = librosa.resample(audio, orig_sr=fs, target_sr=target_fs)
        fs = target_fs
    return audio, fs 

# 需要導入模塊: import librosa [as 別名]
# 或者: from librosa import resample [as 別名]
def readWave(audio_path, start_frame, end_frame, mono=True, sample_rate=None, clip=True):
    snd_file = SoundFile(audio_path, mode='r')
    inf = snd_file._info
    audio_sr = inf.samplerate

    start_read = max(start_frame, 0)
    pad_front = -min(start_frame, 0)
    end_read = min(end_frame, inf.frames)
    pad_back = max(end_frame - inf.frames, 0)

    snd_file.seek(start_read)
    audio = snd_file.read(end_read - start_read, dtype='float32', always_2d=True) # (num_frames, channels)
    snd_file.close()

    # Pad if necessary (start_frame or end_frame out of bounds)
    audio = np.pad(audio, [(pad_front, pad_back), (0, 0)], mode="constant", constant_values=0.0)

    # Convert to mono if desired
    if mono:
        audio = np.mean(audio, axis=1, keepdims=True)

    # Resample if needed
    if sample_rate is not None and sample_rate != audio_sr:
        res_length = int(np.ceil(float(audio.shape[0]) * float(sample_rate) / float(audio_sr)))
        audio = np.pad(audio, [(1, 1), (0,0)], mode="reflect")  # Pad audio first
        audio = librosa.resample(audio.T, audio_sr, sample_rate, res_type="kaiser_fast").T
        skip = (audio.shape[0] - res_length) // 2
        audio = audio[skip:skip+res_length,:]

    # Clip to [-1,1] if desired
    if clip:
        audio = np.minimum(np.maximum(audio, -1.0), 1.0)

    return audio, audio_sr 

# 需要導入模塊: import librosa [as 別名]
# 或者: from librosa import resample [as 別名]
def resample(self, new_sample_rate, **kwargs):
        """
        Resample the data in :attr:`audio_data` to the new sample rate provided by
        :param:`new_sample_rate`. If the :param:`new_sample_rate` is the same as :attr:`sample_rate`
        then nothing happens.

        Args:
            new_sample_rate (int): The new sample rate of :attr:`audio_data`.
            kwargs: Keyword arguments to librosa.resample.

        """

        if new_sample_rate == self.sample_rate:
            warnings.warn('Cannot resample to the same sample rate.')
            return

        resampled_signal = []

        for channel in self.get_channels():
            resampled_channel = librosa.resample(
                channel, self.sample_rate, new_sample_rate, **kwargs)
            resampled_signal.append(resampled_channel)

        self.audio_data = np.array(resampled_signal)
        self.original_signal_length = self.signal_length
        self._sample_rate = new_sample_rate

    ##################################################
    #              Channel Utilities
    ################################################## 

# 需要導入模塊: import librosa [as 別名]
# 或者: from librosa import resample [as 別名]
def read_audio(audio_path, target_fs=None):
    (audio, fs) = soundfile.read(audio_path)

    if audio.ndim > 1:
        audio = np.mean(audio, axis=1)
        
    if target_fs is not None and fs != target_fs:
        audio = librosa.resample(audio, orig_sr=fs, target_sr=target_fs)
        fs = target_fs
        
    return audio, fs 

# 需要導入模塊: import librosa [as 別名]
# 或者: from librosa import resample [as 別名]
def mel(features,path,dataset=None):
    
    """
    This function extracts mel-spectrogram from audio.
    Make sure, you pass a dictionary containing all attributes
    and a path to audio.
    """
    fsx=features['fs'][0]
    n_mels=features['n_mels'][0]
    #print n_mels
    fmin=features['fmin'][0]
    fmax=features['fmax'][0]
    mono=features['mono'][0]
    hamming_window=features['hamming_window'][0]
    noverlap=features['noverlap'][0]
    detrend=features['detrend'][0]
    return_onesided=features['return_onesided'][0]
    mode=features['mode'][0]
    wav, fs = read_audio('librosa',path,dataset)
    #fsx = librosa.resample(wav,fs, 44100)
    #wav, fs = librosa.load(path)
    wav=convert_mono(wav,mono)
    if fs != fsx:
        raise Exception("Assertion Error. Sampling rate Found {} Expected {}".format(fs,fsx))
    ham_win = np.hamming(hamming_window)
    [f, t, X] = signal.spectral.spectrogram(wav,fs, window=ham_win, nperseg=hamming_window, noverlap=noverlap, detrend=detrend, return_onesided=return_onesided, mode=mode )
    X = X.T

    # define global melW, avoid init melW every time, to speed up.
    if globals().get('melW') is None:
        global melW
        melW = librosa.filters.mel( fs, n_fft=hamming_window, n_mels=n_mels, fmin=fmin, fmax=fmax )
        melW /= np.max(melW, axis=-1)[:,None]
    
    X = np.dot( X, melW.T )
    X = X[:, 0:]
    X=feature_normalize(X)
    return X 

# 需要導入模塊: import librosa [as 別名]
# 或者: from librosa import resample [as 別名]
def __call__(self, x, uttid=None, train=True):
        if not train:
            return x

        x = x.astype(numpy.float32)
        if self.accept_uttid:
            ratio = self.utt2ratio[uttid]
        else:
            ratio = self.state.uniform(self.lower, self.upper)

        # Note1: resample requires the sampling-rate of input and output,
        #        but actually only the ratio is used.
        y = librosa.resample(x, ratio, 1, res_type=self.res_type)

        if self.keep_length:
            diff = abs(len(x) - len(y))
            if len(y) > len(x):
                # Truncate noise
                y = y[diff // 2 : -((diff + 1) // 2)]
            elif len(y) < len(x):
                # Assume the time-axis is the first: (Time, Channel)
                pad_width = [(diff // 2, (diff + 1) // 2)] + [
                    (0, 0) for _ in range(y.ndim - 1)
                ]
                y = numpy.pad(
                    y, pad_width=pad_width, constant_values=0, mode="constant"
                )
        return y 

# 需要導入模塊: import librosa [as 別名]
# 或者: from librosa import resample [as 別名]
def compute_melspec(filename, indir, outdir):
    wav = np.load(indir + filename + '.npy')
    wav = librosa.resample(wav, 44100, 22050)
    melspec = librosa.feature.melspectrogram(wav,
                                             sr=22050,
                                             n_fft=1764,
                                             hop_length=220,
                                             n_mels=64)
    logmel = librosa.core.power_to_db(melspec)
    np.save(outdir + filename + '.npy', logmel) 

# 需要導入模塊: import librosa [as 別名]
# 或者: from librosa import resample [as 別名]
def _load_audio(self, path, center_timestamp, nearest_resample=False):
        audio = np.zeros(self.audLen, dtype=np.float32)

        # silent
        if path.endswith('silent'):
            return audio

        # load audio
        audio_raw, rate = self._load_audio_file(path)

        # repeat if audio is too short
        if audio_raw.shape[0] < rate * self.audSec:
            n = int(rate * self.audSec / audio_raw.shape[0]) + 1
            audio_raw = np.tile(audio_raw, n)

        # resample
        if rate > self.audRate:
            # print('resmaple {}->{}'.format(rate, self.audRate))
            if nearest_resample:
                audio_raw = audio_raw[::rate//self.audRate]
            else:
                audio_raw = librosa.resample(audio_raw, rate, self.audRate)

        # crop N seconds
        len_raw = audio_raw.shape[0]
        center = int(center_timestamp * self.audRate)
        start = max(0, center - self.audLen // 2)
        end = min(len_raw, center + self.audLen // 2)

        audio[self.audLen//2-(center-start): self.audLen//2+(end-center)] = \
            audio_raw[start:end]

        # randomize volume
        if self.split == 'train':
            scale = random.random() + 0.5     # 0.5-1.5
            audio *= scale
        audio[audio > 1.] = 1.
        audio[audio < -1.] = -1.

        return audio 

# 需要導入模塊: import librosa [as 別名]
# 或者: from librosa import resample [as 別名]
def resample(audio, orig_sr, new_sr):
    return librosa.resample(audio.T, orig_sr, new_sr).T 

# 需要導入模塊: import librosa [as 別名]
# 或者: from librosa import resample [as 別名]
def resample(self, target_fs, scale=True, res_type='kaiser_best'):
        """Resample audio data.

        Parameters
        ----------
        target_fs : int
            Target sampling rate

        scale : bool
            Scale the resampled signal to have approximately equal total energy (see `librosa.core.resample`).
            Default value True

        res_type : str
            Resample type (see `librosa.core.resample`)
            Default value 'kaiser_best'

        Returns
        -------
        self

        """

        if target_fs != self.fs:
            self._data = numpy.asfortranarray(self._data)
            self._data = librosa.resample(
                y=self._data,
                orig_sr=self.fs,
                target_sr=target_fs,
                scale=scale,
                res_type=res_type
            )
            self.fs = target_fs

        return self 

# 需要導入模塊: import librosa [as 別名]
# 或者: from librosa import resample [as 別名]
def read_audio(path, target_fs=None):

    (audio, fs) = soundfile.read(path)

    if audio.ndim > 1:
        audio = np.mean(audio, axis=1)

    if target_fs is not None and fs != target_fs:
        audio = librosa.resample(audio, orig_sr=fs, target_sr=target_fs)
        fs = target_fs

    return audio, fs 

# 需要導入模塊: import librosa [as 別名]
# 或者: from librosa import resample [as 別名]
def preprocess_wav(fpath_or_wav: Union[str, Path, np.ndarray], source_sr: Optional[int]=None):
    """
    Applies preprocessing operations to a waveform either on disk or in memory such that  
    The waveform will be resampled to match the data hyperparameters.

    :param fpath_or_wav: either a filepath to an audio file (many extensions are supported, not 
    just .wav), either the waveform as a numpy array of floats.
    :param source_sr: if passing an audio waveform, the sampling rate of the waveform before 
    preprocessing. After preprocessing, the waveform'speaker sampling rate will match the data 
    hyperparameters. If passing a filepath, the sampling rate will be automatically detected and 
    this argument will be ignored.
    """
    # Load the wav from disk if needed
    if isinstance(fpath_or_wav, str) or isinstance(fpath_or_wav, Path):
        wav, source_sr = librosa.load(str(fpath_or_wav), sr=None)
    else:
        wav = fpath_or_wav
    
    # Resample the wav
    if source_sr is not None:
        wav = librosa.resample(wav, source_sr, sampling_rate)
        
    # Apply the preprocessing: normalize volume and shorten long silences 
    wav = normalize_volume(wav, audio_norm_target_dBFS, increase_only=True)
    wav = trim_long_silences(wav)
    
    return wav