Python vggish_params.SAMPLE_RATE Examples

The following are 13 code examples of vggish_params.SAMPLE_RATE(). You can vote up the ones you like or vote down the ones you don't like, and go to the original project or source file by following the links above each example. You may also want to check out all available functions/classes of the module vggish_params , or try the search function .
Example #1
Source File: vggish_input.py    From edusense with BSD 3-Clause "New" or "Revised" License 5 votes vote down vote up 
def waveform_to_examples(data, sample_rate):
    # Convert to mono.
    if len(data.shape) > 1:
        data = np.mean(data, axis=1)
    # Resample to the rate assumed by VGGish.
    if sample_rate != vggish_params.SAMPLE_RATE:
        data = resampy.resample(data, sample_rate, vggish_params.SAMPLE_RATE)

    # Compute log mel spectrogram features.
    log_mel = mel_features.log_mel_spectrogram(
        data,
        audio_sample_rate=vggish_params.SAMPLE_RATE,
        log_offset=vggish_params.LOG_OFFSET,
        window_length_secs=vggish_params.STFT_WINDOW_LENGTH_SECONDS,
        hop_length_secs=vggish_params.STFT_HOP_LENGTH_SECONDS,
        num_mel_bins=vggish_params.NUM_MEL_BINS,
        lower_edge_hertz=vggish_params.MEL_MIN_HZ,
        upper_edge_hertz=vggish_params.MEL_MAX_HZ)

    # Frame features into examples.
    features_sample_rate = 1.0 / vggish_params.STFT_HOP_LENGTH_SECONDS
    example_window_length = int(round(
        vggish_params.EXAMPLE_WINDOW_SECONDS * features_sample_rate))
    example_hop_length = int(round(
        vggish_params.EXAMPLE_HOP_SECONDS * features_sample_rate))
    log_mel_examples = mel_features.frame(
        log_mel,
        window_length=example_window_length,
        hop_length=example_hop_length)
    return log_mel_examples
Example #2
Source File: vggish_input.py    From edusense with BSD 3-Clause "New" or "Revised" License 5 votes vote down vote up 
def waveform_to_examples_subtract_bg(data, sample_rate, bg):

    # Convert to mono.
    if len(data.shape) > 1:
        data = np.mean(data, axis=1)
    # Resample to the rate assumed by VGGish.
    if sample_rate != vggish_params.SAMPLE_RATE:
        data = resampy.resample(data, sample_rate, vggish_params.SAMPLE_RATE)

    # Compute log mel spectrogram features.
    log_mel = mel_features.log_mel_spectrogram_subtract_bg(
        data,
        bg,
        audio_sample_rate=vggish_params.SAMPLE_RATE,
        log_offset=vggish_params.LOG_OFFSET,
        window_length_secs=vggish_params.STFT_WINDOW_LENGTH_SECONDS,
        hop_length_secs=vggish_params.STFT_HOP_LENGTH_SECONDS,
        num_mel_bins=vggish_params.NUM_MEL_BINS,
        lower_edge_hertz=vggish_params.MEL_MIN_HZ,
        upper_edge_hertz=vggish_params.MEL_MAX_HZ)

    # Frame features into examples.
    features_sample_rate = 1.0 / vggish_params.STFT_HOP_LENGTH_SECONDS
    example_window_length = int(round(
        vggish_params.EXAMPLE_WINDOW_SECONDS * features_sample_rate))
    example_hop_length = int(round(
        vggish_params.EXAMPLE_HOP_SECONDS * features_sample_rate))
    log_mel_examples = mel_features.frame(
        log_mel,
        window_length=example_window_length,
        hop_length=example_hop_length)
    return log_mel_examples
Example #3
Source File: vggish_input.py    From sklearn-audio-transfer-learning with ISC License 4 votes vote down vote up 
def waveform_to_examples(data, sample_rate):
  """Converts audio waveform into an array of examples for VGGish.

  Args:
    data: np.array of either one dimension (mono) or two dimensions
      (multi-channel, with the outer dimension representing channels).
      Each sample is generally expected to lie in the range [-1.0, +1.0],
      although this is not required.
    sample_rate: Sample rate of data.

  Returns:
    3-D np.array of shape [num_examples, num_frames, num_bands] which represents
    a sequence of examples, each of which contains a patch of log mel
    spectrogram, covering num_frames frames of audio and num_bands mel frequency
    bands, where the frame length is vggish_params.STFT_HOP_LENGTH_SECONDS.
  """
  # Convert to mono.
  if len(data.shape) > 1:
    data = np.mean(data, axis=1)
  # Resample to the rate assumed by VGGish.
  if sample_rate != vggish_params.SAMPLE_RATE:
    data = resampy.resample(data, sample_rate, vggish_params.SAMPLE_RATE)

  # Compute log mel spectrogram features.
  log_mel = mel_features.log_mel_spectrogram(
      data,
      audio_sample_rate=vggish_params.SAMPLE_RATE,
      log_offset=vggish_params.LOG_OFFSET,
      window_length_secs=vggish_params.STFT_WINDOW_LENGTH_SECONDS,
      hop_length_secs=vggish_params.STFT_HOP_LENGTH_SECONDS,
      num_mel_bins=vggish_params.NUM_MEL_BINS,
      lower_edge_hertz=vggish_params.MEL_MIN_HZ,
      upper_edge_hertz=vggish_params.MEL_MAX_HZ)

  # Frame features into examples.
  features_sample_rate = 1.0 / vggish_params.STFT_HOP_LENGTH_SECONDS
  example_window_length = int(round(
      vggish_params.EXAMPLE_WINDOW_SECONDS * features_sample_rate))
  example_hop_length = int(round(
      vggish_params.EXAMPLE_HOP_SECONDS * features_sample_rate))
  log_mel_examples = mel_features.frame(
      log_mel,
      window_length=example_window_length,
      hop_length=example_hop_length)
  return log_mel_examples
Example #4
Source File: vggish_input.py    From Tensorflow-Audio-Classification with Apache License 2.0 4 votes vote down vote up 
def waveform_to_examples(data, sample_rate):
  """Converts audio waveform into an array of examples for VGGish.

  Args:
    data: np.array of either one dimension (mono) or two dimensions
      (multi-channel, with the outer dimension representing channels).
      Each sample is generally expected to lie in the range [-1.0, +1.0],
      although this is not required.
    sample_rate: Sample rate of data.

  Returns:
    3-D np.array of shape [num_examples, num_frames, num_bands] which represents
    a sequence of examples, each of which contains a patch of log mel
    spectrogram, covering num_frames frames of audio and num_bands mel frequency
    bands, where the frame length is vggish_params.STFT_HOP_LENGTH_SECONDS.
  """
  # Convert to mono.
  if len(data.shape) > 1:
    data = np.mean(data, axis=1)
  # Resample to the rate assumed by VGGish.
  if sample_rate != vggish_params.SAMPLE_RATE:
    data = resampy.resample(data, sample_rate, vggish_params.SAMPLE_RATE)

  # Compute log mel spectrogram features.
  log_mel = mel_features.log_mel_spectrogram(
      data,
      audio_sample_rate=vggish_params.SAMPLE_RATE,
      log_offset=vggish_params.LOG_OFFSET,
      window_length_secs=vggish_params.STFT_WINDOW_LENGTH_SECONDS,
      hop_length_secs=vggish_params.STFT_HOP_LENGTH_SECONDS,
      num_mel_bins=vggish_params.NUM_MEL_BINS,
      lower_edge_hertz=vggish_params.MEL_MIN_HZ,
      upper_edge_hertz=vggish_params.MEL_MAX_HZ)

  # Frame features into examples.
  features_sample_rate = 1.0 / vggish_params.STFT_HOP_LENGTH_SECONDS
  example_window_length = int(round(
      vggish_params.EXAMPLE_WINDOW_SECONDS * features_sample_rate))
  example_hop_length = int(round(
      vggish_params.EXAMPLE_HOP_SECONDS * features_sample_rate))
  log_mel_examples = mel_features.frame(
      log_mel,
      window_length=example_window_length,
      hop_length=example_hop_length)
  return log_mel_examples
Example #5
Source File: vggish_input.py    From yolo_v2 with Apache License 2.0 4 votes vote down vote up 
def waveform_to_examples(data, sample_rate):
  """Converts audio waveform into an array of examples for VGGish.

  Args:
    data: np.array of either one dimension (mono) or two dimensions
      (multi-channel, with the outer dimension representing channels).
      Each sample is generally expected to lie in the range [-1.0, +1.0],
      although this is not required.
    sample_rate: Sample rate of data.

  Returns:
    3-D np.array of shape [num_examples, num_frames, num_bands] which represents
    a sequence of examples, each of which contains a patch of log mel
    spectrogram, covering num_frames frames of audio and num_bands mel frequency
    bands, where the frame length is vggish_params.STFT_HOP_LENGTH_SECONDS.
  """
  # Convert to mono.
  if len(data.shape) > 1:
    data = np.mean(data, axis=1)
  # Resample to the rate assumed by VGGish.
  if sample_rate != vggish_params.SAMPLE_RATE:
    data = resampy.resample(data, sample_rate, vggish_params.SAMPLE_RATE)

  # Compute log mel spectrogram features.
  log_mel = mel_features.log_mel_spectrogram(
      data,
      audio_sample_rate=vggish_params.SAMPLE_RATE,
      log_offset=vggish_params.LOG_OFFSET,
      window_length_secs=vggish_params.STFT_WINDOW_LENGTH_SECONDS,
      hop_length_secs=vggish_params.STFT_HOP_LENGTH_SECONDS,
      num_mel_bins=vggish_params.NUM_MEL_BINS,
      lower_edge_hertz=vggish_params.MEL_MIN_HZ,
      upper_edge_hertz=vggish_params.MEL_MAX_HZ)

  # Frame features into examples.
  features_sample_rate = 1.0 / vggish_params.STFT_HOP_LENGTH_SECONDS
  example_window_length = int(round(
      vggish_params.EXAMPLE_WINDOW_SECONDS * features_sample_rate))
  example_hop_length = int(round(
      vggish_params.EXAMPLE_HOP_SECONDS * features_sample_rate))
  log_mel_examples = mel_features.frame(
      log_mel,
      window_length=example_window_length,
      hop_length=example_hop_length)
  return log_mel_examples
Example #6
Source File: vggish_input.py    From Gun-Detector with Apache License 2.0 4 votes vote down vote up 
def waveform_to_examples(data, sample_rate):
  """Converts audio waveform into an array of examples for VGGish.

  Args:
    data: np.array of either one dimension (mono) or two dimensions
      (multi-channel, with the outer dimension representing channels).
      Each sample is generally expected to lie in the range [-1.0, +1.0],
      although this is not required.
    sample_rate: Sample rate of data.

  Returns:
    3-D np.array of shape [num_examples, num_frames, num_bands] which represents
    a sequence of examples, each of which contains a patch of log mel
    spectrogram, covering num_frames frames of audio and num_bands mel frequency
    bands, where the frame length is vggish_params.STFT_HOP_LENGTH_SECONDS.
  """
  # Convert to mono.
  if len(data.shape) > 1:
    data = np.mean(data, axis=1)
  # Resample to the rate assumed by VGGish.
  if sample_rate != vggish_params.SAMPLE_RATE:
    data = resampy.resample(data, sample_rate, vggish_params.SAMPLE_RATE)

  # Compute log mel spectrogram features.
  log_mel = mel_features.log_mel_spectrogram(
      data,
      audio_sample_rate=vggish_params.SAMPLE_RATE,
      log_offset=vggish_params.LOG_OFFSET,
      window_length_secs=vggish_params.STFT_WINDOW_LENGTH_SECONDS,
      hop_length_secs=vggish_params.STFT_HOP_LENGTH_SECONDS,
      num_mel_bins=vggish_params.NUM_MEL_BINS,
      lower_edge_hertz=vggish_params.MEL_MIN_HZ,
      upper_edge_hertz=vggish_params.MEL_MAX_HZ)

  # Frame features into examples.
  features_sample_rate = 1.0 / vggish_params.STFT_HOP_LENGTH_SECONDS
  example_window_length = int(round(
      vggish_params.EXAMPLE_WINDOW_SECONDS * features_sample_rate))
  example_hop_length = int(round(
      vggish_params.EXAMPLE_HOP_SECONDS * features_sample_rate))
  log_mel_examples = mel_features.frame(
      log_mel,
      window_length=example_window_length,
      hop_length=example_hop_length)
  return log_mel_examples
Example #7
Source File: vggish_input.py    From object_detection_kitti with Apache License 2.0 4 votes vote down vote up 
def waveform_to_examples(data, sample_rate):
  """Converts audio waveform into an array of examples for VGGish.

  Args:
    data: np.array of either one dimension (mono) or two dimensions
      (multi-channel, with the outer dimension representing channels).
      Each sample is generally expected to lie in the range [-1.0, +1.0],
      although this is not required.
    sample_rate: Sample rate of data.

  Returns:
    3-D np.array of shape [num_examples, num_frames, num_bands] which represents
    a sequence of examples, each of which contains a patch of log mel
    spectrogram, covering num_frames frames of audio and num_bands mel frequency
    bands, where the frame length is vggish_params.STFT_HOP_LENGTH_SECONDS.
  """
  # Convert to mono.
  if len(data.shape) > 1:
    data = np.mean(data, axis=1)
  # Resample to the rate assumed by VGGish.
  if sample_rate != vggish_params.SAMPLE_RATE:
    data = resampy.resample(data, sample_rate, vggish_params.SAMPLE_RATE)

  # Compute log mel spectrogram features.
  log_mel = mel_features.log_mel_spectrogram(
      data,
      audio_sample_rate=vggish_params.SAMPLE_RATE,
      log_offset=vggish_params.LOG_OFFSET,
      window_length_secs=vggish_params.STFT_WINDOW_LENGTH_SECONDS,
      hop_length_secs=vggish_params.STFT_HOP_LENGTH_SECONDS,
      num_mel_bins=vggish_params.NUM_MEL_BINS,
      lower_edge_hertz=vggish_params.MEL_MIN_HZ,
      upper_edge_hertz=vggish_params.MEL_MAX_HZ)

  # Frame features into examples.
  features_sample_rate = 1.0 / vggish_params.STFT_HOP_LENGTH_SECONDS
  example_window_length = int(round(
      vggish_params.EXAMPLE_WINDOW_SECONDS * features_sample_rate))
  example_hop_length = int(round(
      vggish_params.EXAMPLE_HOP_SECONDS * features_sample_rate))
  log_mel_examples = mel_features.frame(
      log_mel,
      window_length=example_window_length,
      hop_length=example_hop_length)
  return log_mel_examples
Example #8
Source File: vggish_input.py    From object_detection_with_tensorflow with MIT License 4 votes vote down vote up 
def waveform_to_examples(data, sample_rate):
  """Converts audio waveform into an array of examples for VGGish.

  Args:
    data: np.array of either one dimension (mono) or two dimensions
      (multi-channel, with the outer dimension representing channels).
      Each sample is generally expected to lie in the range [-1.0, +1.0],
      although this is not required.
    sample_rate: Sample rate of data.

  Returns:
    3-D np.array of shape [num_examples, num_frames, num_bands] which represents
    a sequence of examples, each of which contains a patch of log mel
    spectrogram, covering num_frames frames of audio and num_bands mel frequency
    bands, where the frame length is vggish_params.STFT_HOP_LENGTH_SECONDS.
  """
  # Convert to mono.
  if len(data.shape) > 1:
    data = np.mean(data, axis=1)
  # Resample to the rate assumed by VGGish.
  if sample_rate != vggish_params.SAMPLE_RATE:
    data = resampy.resample(data, sample_rate, vggish_params.SAMPLE_RATE)

  # Compute log mel spectrogram features.
  log_mel = mel_features.log_mel_spectrogram(
      data,
      audio_sample_rate=vggish_params.SAMPLE_RATE,
      log_offset=vggish_params.LOG_OFFSET,
      window_length_secs=vggish_params.STFT_WINDOW_LENGTH_SECONDS,
      hop_length_secs=vggish_params.STFT_HOP_LENGTH_SECONDS,
      num_mel_bins=vggish_params.NUM_MEL_BINS,
      lower_edge_hertz=vggish_params.MEL_MIN_HZ,
      upper_edge_hertz=vggish_params.MEL_MAX_HZ)

  # Frame features into examples.
  features_sample_rate = 1.0 / vggish_params.STFT_HOP_LENGTH_SECONDS
  example_window_length = int(round(
      vggish_params.EXAMPLE_WINDOW_SECONDS * features_sample_rate))
  example_hop_length = int(round(
      vggish_params.EXAMPLE_HOP_SECONDS * features_sample_rate))
  log_mel_examples = mel_features.frame(
      log_mel,
      window_length=example_window_length,
      hop_length=example_hop_length)
  return log_mel_examples
Example #9
Source File: vggish_input.py    From audioset_classification with MIT License 4 votes vote down vote up 
def waveform_to_examples(data, sample_rate):
  """Converts audio waveform into an array of examples for VGGish.

  Args:
    data: np.array of either one dimension (mono) or two dimensions
      (multi-channel, with the outer dimension representing channels).
      Each sample is generally expected to lie in the range [-1.0, +1.0],
      although this is not required.
    sample_rate: Sample rate of data.

  Returns:
    3-D np.array of shape [num_examples, num_frames, num_bands] which represents
    a sequence of examples, each of which contains a patch of log mel
    spectrogram, covering num_frames frames of audio and num_bands mel frequency
    bands, where the frame length is vggish_params.STFT_HOP_LENGTH_SECONDS.
  """
  # Convert to mono.
  if len(data.shape) > 1:
    data = np.mean(data, axis=1)
  # Resample to the rate assumed by VGGish.
  if sample_rate != vggish_params.SAMPLE_RATE:
    data = resampy.resample(data, sample_rate, vggish_params.SAMPLE_RATE)

  # Compute log mel spectrogram features.
  log_mel = mel_features.log_mel_spectrogram(
      data,
      audio_sample_rate=vggish_params.SAMPLE_RATE,
      log_offset=vggish_params.LOG_OFFSET,
      window_length_secs=vggish_params.STFT_WINDOW_LENGTH_SECONDS,
      hop_length_secs=vggish_params.STFT_HOP_LENGTH_SECONDS,
      num_mel_bins=vggish_params.NUM_MEL_BINS,
      lower_edge_hertz=vggish_params.MEL_MIN_HZ,
      upper_edge_hertz=vggish_params.MEL_MAX_HZ)

  # Frame features into examples.
  features_sample_rate = 1.0 / vggish_params.STFT_HOP_LENGTH_SECONDS
  example_window_length = int(round(
      vggish_params.EXAMPLE_WINDOW_SECONDS * features_sample_rate))
  example_hop_length = int(round(
      vggish_params.EXAMPLE_HOP_SECONDS * features_sample_rate))
  log_mel_examples = mel_features.frame(
      log_mel,
      window_length=example_window_length,
      hop_length=example_hop_length)
  return log_mel_examples
Example #10
Source File: extract_audioset_embedding.py    From audioset_classification with MIT License 4 votes vote down vote up 
def extract_audioset_embedding():
    """Extract log mel spectrogram features. 
    """
    
    # Arguments & parameters
    mel_bins = vggish_params.NUM_BANDS
    sample_rate = vggish_params.SAMPLE_RATE
    input_len = vggish_params.NUM_FRAMES
    embedding_size = vggish_params.EMBEDDING_SIZE
    
    '''You may modify the EXAMPLE_HOP_SECONDS in vggish_params.py to change the 
    hop size. '''

    # Paths
    audio_path = 'appendixes/01.wav'
    checkpoint_path = os.path.join('vggish_model.ckpt')
    pcm_params_path = os.path.join('vggish_pca_params.npz')
    
    if not os.path.isfile(checkpoint_path):
        raise Exception('Please download vggish_model.ckpt from '
            'https://storage.googleapis.com/audioset/vggish_model.ckpt '
            'and put it in the root of this codebase. ')
        
    if not os.path.isfile(pcm_params_path):
        raise Exception('Please download pcm_params_path from '
        'https://storage.googleapis.com/audioset/vggish_pca_params.npz '
        'and put it in the root of this codebase. ')
    
    # Load model
    sess = tf.Session()
    
    vggish_slim.define_vggish_slim(training=False)
    vggish_slim.load_vggish_slim_checkpoint(sess, checkpoint_path)
    features_tensor = sess.graph.get_tensor_by_name(vggish_params.INPUT_TENSOR_NAME)
    embedding_tensor = sess.graph.get_tensor_by_name(vggish_params.OUTPUT_TENSOR_NAME)
    
    pproc = vggish_postprocess.Postprocessor(pcm_params_path)

    # Read audio
    (audio, _) = read_audio(audio_path, target_fs=sample_rate)
    
    # Extract log mel feature
    logmel = vggish_input.waveform_to_examples(audio, sample_rate)

    # Extract embedding feature
    [embedding_batch] = sess.run([embedding_tensor], feed_dict={features_tensor: logmel})
    
    # PCA
    postprocessed_batch = pproc.postprocess(embedding_batch)
    
    print('Audio length: {}'.format(len(audio)))
    print('Log mel shape: {}'.format(logmel.shape))
    print('Embedding feature shape: {}'.format(postprocessed_batch.shape))
Example #11
Source File: vggish_input.py    From g-tensorflow-models with Apache License 2.0 4 votes vote down vote up 
def waveform_to_examples(data, sample_rate):
  """Converts audio waveform into an array of examples for VGGish.

  Args:
    data: np.array of either one dimension (mono) or two dimensions
      (multi-channel, with the outer dimension representing channels).
      Each sample is generally expected to lie in the range [-1.0, +1.0],
      although this is not required.
    sample_rate: Sample rate of data.

  Returns:
    3-D np.array of shape [num_examples, num_frames, num_bands] which represents
    a sequence of examples, each of which contains a patch of log mel
    spectrogram, covering num_frames frames of audio and num_bands mel frequency
    bands, where the frame length is vggish_params.STFT_HOP_LENGTH_SECONDS.
  """
  # Convert to mono.
  if len(data.shape) > 1:
    data = np.mean(data, axis=1)
  # Resample to the rate assumed by VGGish.
  if sample_rate != vggish_params.SAMPLE_RATE:
    data = resampy.resample(data, sample_rate, vggish_params.SAMPLE_RATE)

  # Compute log mel spectrogram features.
  log_mel = mel_features.log_mel_spectrogram(
      data,
      audio_sample_rate=vggish_params.SAMPLE_RATE,
      log_offset=vggish_params.LOG_OFFSET,
      window_length_secs=vggish_params.STFT_WINDOW_LENGTH_SECONDS,
      hop_length_secs=vggish_params.STFT_HOP_LENGTH_SECONDS,
      num_mel_bins=vggish_params.NUM_MEL_BINS,
      lower_edge_hertz=vggish_params.MEL_MIN_HZ,
      upper_edge_hertz=vggish_params.MEL_MAX_HZ)

  # Frame features into examples.
  features_sample_rate = 1.0 / vggish_params.STFT_HOP_LENGTH_SECONDS
  example_window_length = int(round(
      vggish_params.EXAMPLE_WINDOW_SECONDS * features_sample_rate))
  example_hop_length = int(round(
      vggish_params.EXAMPLE_HOP_SECONDS * features_sample_rate))
  log_mel_examples = mel_features.frame(
      log_mel,
      window_length=example_window_length,
      hop_length=example_hop_length)
  return log_mel_examples
Example #12
Source File: vggish_input.py    From models with Apache License 2.0 4 votes vote down vote up 
def waveform_to_examples(data, sample_rate):
  """Converts audio waveform into an array of examples for VGGish.

  Args:
    data: np.array of either one dimension (mono) or two dimensions
      (multi-channel, with the outer dimension representing channels).
      Each sample is generally expected to lie in the range [-1.0, +1.0],
      although this is not required.
    sample_rate: Sample rate of data.

  Returns:
    3-D np.array of shape [num_examples, num_frames, num_bands] which represents
    a sequence of examples, each of which contains a patch of log mel
    spectrogram, covering num_frames frames of audio and num_bands mel frequency
    bands, where the frame length is vggish_params.STFT_HOP_LENGTH_SECONDS.
  """
  # Convert to mono.
  if len(data.shape) > 1:
    data = np.mean(data, axis=1)
  # Resample to the rate assumed by VGGish.
  if sample_rate != vggish_params.SAMPLE_RATE:
    data = resampy.resample(data, sample_rate, vggish_params.SAMPLE_RATE)

  # Compute log mel spectrogram features.
  log_mel = mel_features.log_mel_spectrogram(
      data,
      audio_sample_rate=vggish_params.SAMPLE_RATE,
      log_offset=vggish_params.LOG_OFFSET,
      window_length_secs=vggish_params.STFT_WINDOW_LENGTH_SECONDS,
      hop_length_secs=vggish_params.STFT_HOP_LENGTH_SECONDS,
      num_mel_bins=vggish_params.NUM_MEL_BINS,
      lower_edge_hertz=vggish_params.MEL_MIN_HZ,
      upper_edge_hertz=vggish_params.MEL_MAX_HZ)

  # Frame features into examples.
  features_sample_rate = 1.0 / vggish_params.STFT_HOP_LENGTH_SECONDS
  example_window_length = int(round(
      vggish_params.EXAMPLE_WINDOW_SECONDS * features_sample_rate))
  example_hop_length = int(round(
      vggish_params.EXAMPLE_HOP_SECONDS * features_sample_rate))
  log_mel_examples = mel_features.frame(
      log_mel,
      window_length=example_window_length,
      hop_length=example_hop_length)
  return log_mel_examples
Example #13
Source File: vggish_input.py    From multilabel-image-classification-tensorflow with MIT License 4 votes vote down vote up 
def waveform_to_examples(data, sample_rate):
  """Converts audio waveform into an array of examples for VGGish.

  Args:
    data: np.array of either one dimension (mono) or two dimensions
      (multi-channel, with the outer dimension representing channels).
      Each sample is generally expected to lie in the range [-1.0, +1.0],
      although this is not required.
    sample_rate: Sample rate of data.

  Returns:
    3-D np.array of shape [num_examples, num_frames, num_bands] which represents
    a sequence of examples, each of which contains a patch of log mel
    spectrogram, covering num_frames frames of audio and num_bands mel frequency
    bands, where the frame length is vggish_params.STFT_HOP_LENGTH_SECONDS.
  """
  # Convert to mono.
  if len(data.shape) > 1:
    data = np.mean(data, axis=1)
  # Resample to the rate assumed by VGGish.
  if sample_rate != vggish_params.SAMPLE_RATE:
    data = resampy.resample(data, sample_rate, vggish_params.SAMPLE_RATE)

  # Compute log mel spectrogram features.
  log_mel = mel_features.log_mel_spectrogram(
      data,
      audio_sample_rate=vggish_params.SAMPLE_RATE,
      log_offset=vggish_params.LOG_OFFSET,
      window_length_secs=vggish_params.STFT_WINDOW_LENGTH_SECONDS,
      hop_length_secs=vggish_params.STFT_HOP_LENGTH_SECONDS,
      num_mel_bins=vggish_params.NUM_MEL_BINS,
      lower_edge_hertz=vggish_params.MEL_MIN_HZ,
      upper_edge_hertz=vggish_params.MEL_MAX_HZ)

  # Frame features into examples.
  features_sample_rate = 1.0 / vggish_params.STFT_HOP_LENGTH_SECONDS
  example_window_length = int(round(
      vggish_params.EXAMPLE_WINDOW_SECONDS * features_sample_rate))
  example_hop_length = int(round(
      vggish_params.EXAMPLE_HOP_SECONDS * features_sample_rate))
  log_mel_examples = mel_features.frame(
      log_mel,
      window_length=example_window_length,
      hop_length=example_hop_length)
  return log_mel_examples