def _preprocess_audio(audio_file_path, audio_featurizer, normalize):
"""Load the audio file and compute spectrogram feature."""
data, _ = soundfile.read(audio_file_path)
feature = featurizer.compute_spectrogram_feature(
data, audio_featurizer.sample_rate, audio_featurizer.stride_ms,
audio_featurizer.window_ms)
# Feature normalization
if normalize:
feature = _normalize_audio_feature(feature)
# Adding Channel dimension for conv2D input.
feature = np.expand_dims(feature, axis=2)
return feature
def _preprocess_audio(audio_file_path, audio_featurizer, normalize):
"""Load the audio file and compute spectrogram feature."""
data, _ = soundfile.read(audio_file_path)
feature = featurizer.compute_spectrogram_feature(
data, audio_featurizer.sample_rate, audio_featurizer.stride_ms,
audio_featurizer.window_ms)
# Feature normalization
if normalize:
feature = _normalize_audio_feature(feature)
# Adding Channel dimension for conv2D input.
feature = np.expand_dims(feature, axis=2)
return feature
def _preprocess_audio(audio_file_path, audio_sample_rate, audio_featurizer,
normalize):
"""Load the audio file in memory and compute spectrogram feature."""
tf.logging.info(
"Extracting spectrogram feature for {}".format(audio_file_path))
sample_rate, data = wavfile.read(audio_file_path)
assert sample_rate == audio_sample_rate
if data.dtype not in [np.float32, np.float64]:
data = data.astype(np.float32) / np.iinfo(data.dtype).max
feature = featurizer.compute_spectrogram_feature(
data, audio_featurizer.frame_length, audio_featurizer.frame_step,
audio_featurizer.fft_length)
if normalize:
feature = _normalize_audio_feature(feature)
return feature