def test_read_wav_internal(self):
fs, wav = wavfile.read(self.mono_wav, False)
wave_data = feat.read_wav_kaldi_internal(wav, fs)
# Correct number of channels
self.assertEqual(wave_data.data().num_rows, 1)
# Correct number of samples
self.assertEqual(wav.shape[0], wave_data.data().num_cols)
def get_ppg(wav_path, deps):
"""
Input:
wav_path: *.wav path
deps = DependenciesPPG()
Library:
from scipy.io import wavfile
from common import feat
"""
fs, wav = wavfile.read(wav_path)
wave_data = feat.read_wav_kaldi_internal(wav, fs)
seq = ppg.compute_full_ppg_wrapper(wave_data, deps.nnet, deps.lda, 10)
return seq
def get_monophone_ppg(self) -> ndarray:
"""A wrapper function to initialize the monophone ppg of this utterance.
Requires non-empty waveform, fs, and kaldi_shift.
Returns:
The monophone ppgs in numpy ndarray format.
"""
if self.kaldi_shift < 1: # ms
raise ValueError('Invalid frame kaldi frame shift parameter %d.',
self.kaldi_shift)
if self.wav.size == 0 or self.fs < 0:
raise ValueError('To perform alignment, the object must contain '
'valid speech data and sampling frequency.')
wav_kaldi = read_wav_kaldi_internal(self.wav, self.fs)
ppg_deps = ppg.DependenciesPPG()
self.monophone_ppg = ppg.compute_monophone_ppg(wav_kaldi, ppg_deps.nnet,
ppg_deps.lda,
ppg_deps.monophone_trans,
self.kaldi_shift)
return self.monophone_ppg
def get_ppg(wav_path, deps):
fs, wav = wavfile.read(wav_path)
wave_data = feat.read_wav_kaldi_internal(wav, fs)
seq = ppg.compute_full_ppg_wrapper(wave_data, deps.nnet, deps.lda, 10)
return seq
def get_ppg(wav, fs):
wave_data = feat.read_wav_kaldi_internal(wav, fs)
seq = compute_full_ppg_wrapper(wave_data, ppg_deps.nnet, ppg_deps.lda, mel_window_step)
return seq