kaldifeat.FbankOptions¶
If you want to construct an instance of kaldifeat.Fbank or kaldifeat.OnlineFbank, you have to provide an instance of kaldifeat.FbankOptions.
The following code shows how to construct an instance of kaldifeat.FbankOptions.
$ python3
Python 3.8.0 (default, Oct 28 2019, 16:14:01)
[GCC 8.3.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import kaldifeat
>>> opts = kaldifeat.FbankOptions()
>>> print(opts)
frame_opts:
samp_freq: 16000
frame_shift_ms: 10
frame_length_ms: 25
dither: 1
preemph_coeff: 0.97
remove_dc_offset: 1
window_type: povey
round_to_power_of_two: 1
blackman_coeff: 0.42
snip_edges: 1
max_feature_vectors: -1
mel_opts:
num_bins: 23
low_freq: 20
high_freq: 0
vtln_low: 100
vtln_high: -500
debug_mel: 0
htk_mode: 0
use_energy: 0
energy_floor: 0
raw_energy: 1
htk_compat: 0
use_log_fbank: 1
use_power: 1
device: cpu
>>> print(opts.dither)
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
AttributeError: '_kaldifeat.FbankOptions' object has no attribute 'dither'
>>>
>>> print(opts.frame_opts.dither)
1.0
>>> opts.frame_opts.dither = 0 # disable dither
>>> print(opts.frame_opts.dither)
0.0
>>> import torch
>>> print(opts.device)
cpu
>>> opts.device = 'cuda:0'
>>> print(opts.device)
cuda:0
>>> opts.device = torch.device('cuda', 1)
>>> print(opts.device)
cuda:1
>>> opts.device = 'cpu'
>>> print(opts.device)
cpu
>>> print(opts.mel_opts.num_bins)
23
>>> opts.mel_opts.num_bins = 80
>>> print(opts.mel_opts.num_bins)
80
Note that we reuse the same option name with compute-fbank-feats from Kaldi:
$ compute-fbank-feats --help
compute-fbank-feats
Create Mel-filter bank (FBANK) feature files.
Usage: compute-fbank-feats [options...] <wav-rspecifier> <feats-wspecifier>
Options:
--allow-downsample : If true, allow the input waveform to have a higher frequency than the specified --sample-frequency (and we'll downsample). (bool, default = false)
--allow-upsample : If true, allow the input waveform to have a lower frequency than the specified --sample-frequency (and we'll upsample). (bool, default = false)
--blackman-coeff : Constant coefficient for generalized Blackman window. (float, default = 0.42)
--channel : Channel to extract (-1 -> expect mono, 0 -> left, 1 -> right) (int, default = -1)
--debug-mel : Print out debugging information for mel bin computation (bool, default = false)
--dither : Dithering constant (0.0 means no dither). If you turn this off, you should set the --energy-floor option, e.g. to 1.0 or 0.1 (float, default = 1)
--energy-floor : Floor on energy (absolute, not relative) in FBANK computation. Only makes a difference if --use-energy=true; only necessary if --dither=0.0. Suggested values: 0.1 or 1.0 (float, default = 0)
--frame-length : Frame length in milliseconds (float, default = 25)
--frame-shift : Frame shift in milliseconds (float, default = 10)
--high-freq : High cutoff frequency for mel bins (if <= 0, offset from Nyquist) (float, default = 0)
--htk-compat : If true, put energy last. Warning: not sufficient to get HTK compatible features (need to change other parameters). (bool, default = false)
--low-freq : Low cutoff frequency for mel bins (float, default = 20)
--max-feature-vectors : Memory optimization. If larger than 0, periodically remove feature vectors so that only this number of the latest feature vectors is retained. (int, default = -1)
--min-duration : Minimum duration of segments to process (in seconds). (float, default = 0)
--num-mel-bins : Number of triangular mel-frequency bins (int, default = 23)
--output-format : Format of the output files [kaldi, htk] (string, default = "kaldi")
--preemphasis-coefficient : Coefficient for use in signal preemphasis (float, default = 0.97)
--raw-energy : If true, compute energy before preemphasis and windowing (bool, default = true)
--remove-dc-offset : Subtract mean from waveform on each frame (bool, default = true)
--round-to-power-of-two : If true, round window size to power of two by zero-padding input to FFT. (bool, default = true)
--sample-frequency : Waveform data sample frequency (must match the waveform file, if specified there) (float, default = 16000)
--snip-edges : If true, end effects will be handled by outputting only frames that completely fit in the file, and the number of frames depends on the frame-length. If false, the number of frames depends only on the frame-shift, and we reflect the data at the ends. (bool, default = true)
--subtract-mean : Subtract mean of each feature file [CMS]; not recommended to do it this way. (bool, default = false)
--use-energy : Add an extra dimension with energy to the FBANK output. (bool, default = false)
--use-log-fbank : If true, produce log-filterbank, else produce linear. (bool, default = true)
--use-power : If true, use power, else use magnitude. (bool, default = true)
--utt2spk : Utterance to speaker-id map (if doing VTLN and you have warps per speaker) (string, default = "")
--vtln-high : High inflection point in piecewise linear VTLN warping function (if negative, offset from high-mel-freq (float, default = -500)
--vtln-low : Low inflection point in piecewise linear VTLN warping function (float, default = 100)
--vtln-map : Map from utterance or speaker-id to vtln warp factor (rspecifier) (string, default = "")
--vtln-warp : Vtln warp factor (only applicable if vtln-map not specified) (float, default = 1)
--window-type : Type of window ("hamming"|"hanning"|"povey"|"rectangular"|"sine"|"blackmann") (string, default = "povey")
--write-utt2dur : Wspecifier to write duration of each utterance in seconds, e.g. 'ark,t:utt2dur'. (string, default = "")
Standard options:
--config : Configuration file to read (this option may be repeated) (string, default = "")
--help : Print out usage message (bool, default = false)
--print-args : Print the command line arguments (to stderr) (bool, default = true)
--verbose : Verbose level (higher->more logging) (int, default = 0)
Please refer to the output of compute-fbank-feats --help for the meaning
of each field of kaldifeat.FbankOptions.
One thing worth noting is that kaldifeat.FbankOptions has a field device,
which is an instance of torch.device. You can assign it either a string, e.g.,
"cpu" or "cuda:0", or an instance of torch.device, e.g., torch.device("cpu") or
torch.device("cuda", 1).
Hint
You can use this field to control whether the feature computer constructed from it performs computation on CPU or CUDA.
Caution
If you use a CUDA device, make sure that you have installed a CUDA version of PyTorch.
Example usage¶
The following code from https://github.com/csukuangfj/kaldifeat/blob/master/kaldifeat/python/tests/test_fbank_options.py demonstrate the usage of kaldifeat.FbankOptions:
#!/usr/bin/env python3
#
# Copyright (c) 2021 Xiaomi Corporation (authors: Fangjun Kuang)
import pickle
import torch
import kaldifeat
def test_default():
opts = kaldifeat.FbankOptions()
print(opts)
assert opts.frame_opts.samp_freq == 16000
assert opts.frame_opts.frame_shift_ms == 10.0
assert opts.frame_opts.frame_length_ms == 25.0
assert opts.frame_opts.dither == 1.0
assert abs(opts.frame_opts.preemph_coeff - 0.97) < 1e-6
assert opts.frame_opts.remove_dc_offset is True
assert opts.frame_opts.window_type == "povey"
assert opts.frame_opts.round_to_power_of_two is True
assert abs(opts.frame_opts.blackman_coeff - 0.42) < 1e-6
assert opts.frame_opts.snip_edges is True
assert opts.mel_opts.num_bins == 23
assert opts.mel_opts.low_freq == 20
assert opts.mel_opts.high_freq == 0
assert opts.mel_opts.vtln_low == 100
assert opts.mel_opts.vtln_high == -500
assert opts.mel_opts.debug_mel is False
assert opts.mel_opts.htk_mode is False
assert opts.use_energy is False
assert opts.energy_floor == 0.0
assert opts.raw_energy is True
assert opts.htk_compat is False
assert opts.use_log_fbank is True
assert opts.use_power is True
assert opts.device.type == "cpu"
def test_set_get():
opts = kaldifeat.FbankOptions()
opts.use_energy = True
assert opts.use_energy is True
opts.energy_floor = 1
assert opts.energy_floor == 1
opts.raw_energy = False
assert opts.raw_energy is False
opts.htk_compat = True
assert opts.htk_compat is True
opts.use_log_fbank = False
assert opts.use_log_fbank is False
opts.use_power = False
assert opts.use_power is False
opts.device = torch.device("cuda", 1)
assert opts.device.type == "cuda"
assert opts.device.index == 1
def test_set_get_frame_opts():
opts = kaldifeat.FbankOptions()
opts.frame_opts.samp_freq = 44100
assert opts.frame_opts.samp_freq == 44100
opts.frame_opts.frame_shift_ms = 20.5
assert opts.frame_opts.frame_shift_ms == 20.5
opts.frame_opts.frame_length_ms = 1
assert opts.frame_opts.frame_length_ms == 1
opts.frame_opts.dither = 0.5
assert opts.frame_opts.dither == 0.5
opts.frame_opts.preemph_coeff = 0.25
assert opts.frame_opts.preemph_coeff == 0.25
opts.frame_opts.remove_dc_offset = False
assert opts.frame_opts.remove_dc_offset is False
opts.frame_opts.window_type = "hanning"
assert opts.frame_opts.window_type == "hanning"
opts.frame_opts.round_to_power_of_two = False
assert opts.frame_opts.round_to_power_of_two is False
opts.frame_opts.blackman_coeff = 0.25
assert opts.frame_opts.blackman_coeff == 0.25
opts.frame_opts.snip_edges = False
assert opts.frame_opts.snip_edges is False
def test_set_get_mel_opts():
opts = kaldifeat.FbankOptions()
opts.mel_opts.num_bins = 100
assert opts.mel_opts.num_bins == 100
opts.mel_opts.low_freq = 22
assert opts.mel_opts.low_freq == 22
opts.mel_opts.high_freq = 1
assert opts.mel_opts.high_freq == 1
opts.mel_opts.vtln_low = 101
assert opts.mel_opts.vtln_low == 101
opts.mel_opts.vtln_high = -100
assert opts.mel_opts.vtln_high == -100
opts.mel_opts.debug_mel = True
assert opts.mel_opts.debug_mel is True
opts.mel_opts.htk_mode = True
assert opts.mel_opts.htk_mode is True
def test_from_empty_dict():
opts = kaldifeat.FbankOptions.from_dict({})
opts2 = kaldifeat.FbankOptions()
assert str(opts) == str(opts2)
def test_from_dict_partial():
d = {
"energy_floor": 10.5,
"htk_compat": True,
"mel_opts": {"num_bins": 80, "vtln_low": 1},
"frame_opts": {"window_type": "hanning"},
}
opts = kaldifeat.FbankOptions.from_dict(d)
assert opts.energy_floor == 10.5
assert opts.htk_compat is True
assert opts.mel_opts.num_bins == 80
assert opts.mel_opts.vtln_low == 1
assert opts.frame_opts.window_type == "hanning"
mel_opts = kaldifeat.MelBanksOptions.from_dict(d["mel_opts"])
assert str(opts.mel_opts) == str(mel_opts)
def test_from_dict_full_and_as_dict():
opts = kaldifeat.FbankOptions()
opts.htk_compat = True
opts.mel_opts.num_bins = 80
opts.frame_opts.samp_freq = 10
d = opts.as_dict()
assert d["htk_compat"] is True
assert d["mel_opts"]["num_bins"] == 80
assert d["frame_opts"]["samp_freq"] == 10
mel_opts = kaldifeat.MelBanksOptions()
mel_opts.num_bins = 80
assert d["mel_opts"] == mel_opts.as_dict()
frame_opts = kaldifeat.FrameExtractionOptions()
frame_opts.samp_freq = 10
assert d["frame_opts"] == frame_opts.as_dict()
opts2 = kaldifeat.FbankOptions.from_dict(d)
assert str(opts2) == str(opts)
d["htk_compat"] = False
d["device"] = torch.device("cuda", 2)
opts3 = kaldifeat.FbankOptions.from_dict(d)
assert opts3.htk_compat is False
assert opts3.device == torch.device("cuda", 2)
def test_pickle():
opts = kaldifeat.FbankOptions()
opts.use_energy = True
opts.use_power = False
opts.device = torch.device("cuda", 1)
opts.frame_opts.samp_freq = 44100
opts.mel_opts.num_bins = 100
data = pickle.dumps(opts)
opts2 = pickle.loads(data)
assert str(opts) == str(opts2)
def main():
test_default()
test_set_get()
test_set_get_frame_opts()
test_set_get_mel_opts()
test_from_empty_dict()
test_from_dict_partial()
test_from_dict_full_and_as_dict()
test_pickle()
if __name__ == "__main__":
main()