def _get_2D_feature(audio, feature_type, audio_feature_dict,
sampling_rate_in_hz):
window_length_in_s = audio_feature_dict['window_length_in_s']
window_shift_in_s = audio_feature_dict['window_shift_in_s']
if 'num_fft_points' in audio_feature_dict:
num_fft_points = audio_feature_dict['num_fft_points']
else:
num_fft_points = get_length_in_samp(window_length_in_s,
sampling_rate_in_hz)
if 'window_type' in audio_feature_dict:
window_type = audio_feature_dict['window_type']
else:
window_type = 'hamming'
if feature_type == 'stft_phase':
return get_phase_stft_magnitude(audio, sampling_rate_in_hz,
window_length_in_s,
window_shift_in_s, num_fft_points,
window_type)
if feature_type == 'stft':
return get_stft_magnitude(audio, sampling_rate_in_hz,
window_length_in_s, window_shift_in_s,
num_fft_points, window_type)
if feature_type == 'group_delay':
return get_group_delay(audio, sampling_rate_in_hz,
window_length_in_s, window_shift_in_s,
num_fft_points, window_type)
def _get_feature_dim(audio_feature_dict, sampling_rate_in_hz):
feature_type = audio_feature_dict[TYPE]
if feature_type == "raw":
feature_dim = 1
elif feature_type == "stft_phase":
feature_dim_symmetric = get_length_in_samp(audio_feature_dict["window_length_in_s"], sampling_rate_in_hz)
feature_dim = 2 * get_non_symmetric_length(feature_dim_symmetric)
elif feature_type in ["stft", "group_delay"]:
feature_dim_symmetric = get_length_in_samp(audio_feature_dict["window_length_in_s"], sampling_rate_in_hz)
feature_dim = get_non_symmetric_length(feature_dim_symmetric)
elif feature_type == "fbank":
feature_dim = audio_feature_dict["num_filter_bands"]
else:
raise ValueError(f"{feature_type} is not recognized.")
return feature_dim
def _get_feature_dim(audio_feature_dict, sampling_rate_in_hz):
feature_type = audio_feature_dict['type']
if feature_type == 'raw':
feature_dim = 1
elif feature_type == 'stft_phase':
feature_dim_symmetric = get_length_in_samp(
audio_feature_dict['window_length_in_s'], sampling_rate_in_hz)
feature_dim = 2 * get_non_symmetric_length(feature_dim_symmetric)
elif feature_type in ['stft', 'group_delay']:
feature_dim_symmetric = get_length_in_samp(
audio_feature_dict['window_length_in_s'], sampling_rate_in_hz)
feature_dim = get_non_symmetric_length(feature_dim_symmetric)
else:
raise ValueError('{} is not recognized.'.format(feature_type))
return feature_dim
def _get_2D_feature(audio, feature_type, audio_feature_dict,
sampling_rate_in_hz):
window_length_in_s = audio_feature_dict["window_length_in_s"]
window_shift_in_s = audio_feature_dict["window_shift_in_s"]
window_length_in_samp = get_length_in_samp(window_length_in_s,
sampling_rate_in_hz)
if "num_fft_points" in audio_feature_dict:
num_fft_points = audio_feature_dict["num_fft_points"]
if num_fft_points < window_length_in_samp:
raise ValueError("num_fft_points: {} < window length in "
"samples: {} (corresponds to window length"
" in s: {}".format(num_fft_points,
window_length_in_s,
window_length_in_samp))
else:
num_fft_points = window_length_in_samp
if "window_type" in audio_feature_dict:
window_type = audio_feature_dict["window_type"]
else:
window_type = "hamming"
if feature_type == "stft_phase":
return get_phase_stft_magnitude(audio, sampling_rate_in_hz,
window_length_in_s,
window_shift_in_s, num_fft_points,
window_type)
if feature_type == "stft":
return get_stft_magnitude(audio, sampling_rate_in_hz,
window_length_in_s, window_shift_in_s,
num_fft_points, window_type)
if feature_type == "group_delay":
return get_group_delay(audio, sampling_rate_in_hz,
window_length_in_s, window_shift_in_s,
num_fft_points, window_type)
if feature_type == "fbank":
num_filter_bands = audio_feature_dict["num_filter_bands"]
return get_fbank(
audio,
sampling_rate_in_hz,
window_length_in_s,
window_shift_in_s,
num_fft_points,
window_type,
num_filter_bands,
)
def _get_2D_feature(audio, feature_type, audio_feature_dict,
sampling_rate_in_hz):
window_length_in_s = audio_feature_dict['window_length_in_s']
window_shift_in_s = audio_feature_dict['window_shift_in_s']
window_length_in_samp = get_length_in_samp(window_length_in_s,
sampling_rate_in_hz)
if 'num_fft_points' in audio_feature_dict:
num_fft_points = audio_feature_dict['num_fft_points']
if num_fft_points < window_length_in_samp:
raise ValueError(
'num_fft_points: {} < window length in '
'samples: {} (corresponds to window length'
' in s: {}'.format(num_fft_points, window_length_in_s,
window_length_in_samp))
else:
num_fft_points = window_length_in_samp
if 'window_type' in audio_feature_dict:
window_type = audio_feature_dict['window_type']
else:
window_type = 'hamming'
if feature_type == 'stft_phase':
return get_phase_stft_magnitude(audio, sampling_rate_in_hz,
window_length_in_s,
window_shift_in_s, num_fft_points,
window_type)
if feature_type == 'stft':
return get_stft_magnitude(audio, sampling_rate_in_hz,
window_length_in_s, window_shift_in_s,
num_fft_points, window_type)
if feature_type == 'group_delay':
return get_group_delay(audio, sampling_rate_in_hz,
window_length_in_s, window_shift_in_s,
num_fft_points, window_type)
if feature_type == 'fbank':
num_filter_bands = audio_feature_dict['num_filter_bands']
return get_fbank(audio, sampling_rate_in_hz,
window_length_in_s, window_shift_in_s,
num_fft_points, window_type, num_filter_bands)
def _get_2D_feature(
audio: torch.Tensor,
feature_type: str,
audio_feature_dict: Dict[str, Union[float, int, str]],
sampling_rate_in_hz: int,
) -> torch.Tensor:
window_length_in_s = audio_feature_dict["window_length_in_s"]
window_shift_in_s = audio_feature_dict["window_shift_in_s"]
assert torch.jit.isinstance(window_length_in_s, float)
assert torch.jit.isinstance(window_shift_in_s, float)
window_length_in_samp = get_length_in_samp(window_length_in_s,
sampling_rate_in_hz)
if "num_fft_points" in audio_feature_dict:
num_fft_points = audio_feature_dict["num_fft_points"]
assert torch.jit.isinstance(num_fft_points, int)
if num_fft_points < window_length_in_samp:
raise ValueError("num_fft_points: {} < window length in "
"samples: {} (corresponds to window length"
" in s: {}".format(num_fft_points,
window_length_in_s,
window_length_in_samp))
else:
num_fft_points = window_length_in_samp
if "window_type" in audio_feature_dict:
window_type = audio_feature_dict["window_type"]
assert torch.jit.isinstance(window_type, str)
else:
window_type = "hamming"
if feature_type == "stft_phase":
return get_phase_stft_magnitude(audio, sampling_rate_in_hz,
window_length_in_s,
window_shift_in_s, num_fft_points,
window_type)
elif feature_type == "stft":
return get_stft_magnitude(audio, sampling_rate_in_hz,
window_length_in_s, window_shift_in_s,
num_fft_points, window_type)
elif feature_type == "group_delay":
return get_group_delay(audio, sampling_rate_in_hz,
window_length_in_s, window_shift_in_s,
num_fft_points, window_type)
elif feature_type == "fbank":
num_filter_bands = audio_feature_dict["num_filter_bands"]
assert torch.jit.isinstance(num_filter_bands, int)
return get_fbank(
audio,
sampling_rate_in_hz,
window_length_in_s,
window_shift_in_s,
num_fft_points,
window_type,
num_filter_bands,
)
else:
raise ValueError(
f'feature_type "{feature_type}" is not recognized.')
