def extract_spec(filename,
samp_freq,
frame_length_ms=25,
frame_shift_ms=10,
round_to_power_of_two=True,
snip_edges=True):
'''
extract spectrogram using kaldi
args:
filename: wav file path
samp_freq: sample frequence
return:
spectrogram: (frame, fre)
'''
# get rspec and wspec
with open('wav.scp', 'w') as f:
f.write('test1 ' + filename + '\n')
rspec = 'scp,p:' + 'wav.scp'
wspec = 'ark,t:' + 'spec.ark'
# set po
usage = """Extract MFCC features.Usage: example.py [opts...] <rspec> <wspec>"""
po = ParseOptions(usage)
po.register_float("min-duration", 0.0, "minimum segment duration")
opts = po.parse_args()
# set options
spec_opts = SpectrogramOptions()
spec_opts.frame_opts.samp_freq = samp_freq
spec_opts.frame_opts.frame_length_ms = frame_length_ms
spec_opts.frame_opts.frame_shift_ms = frame_shift_ms
spec_opts.frame_opts.round_to_power_of_two = round_to_power_of_two
spec_opts.frame_opts.snip_edges = snip_edges
spec_opts.register(po)
spec = Spectrogram(spec_opts)
sf = spec_opts.frame_opts.samp_freq
with SequentialWaveReader(rspec) as reader, MatrixWriter(wspec) as writer:
for key, wav in reader:
if wav.duration < opts.min_duration:
continue
assert (wav.samp_freq >= sf)
assert (wav.samp_freq % sf == 0)
s = wav.data()
s = s[:, ::int(wav.samp_freq / sf)]
m = SubVector(mean(s, axis=0))
f = spec.compute_features(m, sf, 1.0)
f_array = np.array(f)
writer[key] = f
return f_array
"%(filename)s:%(lineno)s) %(message)s".format(__version__),
level=logging.INFO)
usage = """Use Principal component analysis for dimension reduction.
For the details, Please refer to website:
https://scikit-learn.org/stable/modules/generated/sklearn.decomposition.PCA.html
Usage: pca-vector.py [options] <vector-rspecifier> <vector-wspecifier
e.g.
pca-vector.py scp:data/train/ivector.scp ark:data/train/low_dim_vector.ark
see also: two-dim-vector-visual.py
"""
po = ParseOptions(usage)
po.register_int(
"output-dim", 2, "dimension of the output vectors."
" For visualization, only 2 is allowed in this program. (2 by default)"
)
opts = po.parse_args()
if (po.num_args() != 2):
po.print_usage()
sys.exit()
vector_rspecifier = po.get_arg(1)
vector_wspecifier = po.get_arg(2)
isSuccess = pca_vector(vector_rspecifier,
vector_wspecifier,
output_dim=opts.output_dim)
if not isSuccess:
sys.exit()