with np.warnings.catch_warnings():
np.warnings.filterwarnings('ignore')
for path, name in SAMPLED_WAV_FILE:
feat = recipe.transform(path)
assert feat['bnf'].shape[0] == feat['mspec'].shape[0]
V.plot_multiple_features(feat, title=feat['name'])
V.plot_save(os.path.join(PATH_EXP, 'features_%s.pdf' % args.recipe))
exit()
# ===========================================================================
# Prepare the processor
# ===========================================================================
with np.warnings.catch_warnings():
np.warnings.filterwarnings('ignore')
jobs = list(WAV_FILES.keys())
processor = pp.FeatureProcessor(
jobs=jobs,
path=os.path.join(PATH_ACOUSTIC_FEAT, args.recipe),
extractor=recipe,
n_cache=1200,
ncpu=min(18,
cpu_count() - 2),
override=True,
identifier='name',
log_path=os.path.join(PATH_EXP, 'processor_%s.log' % args.recipe),
stop_on_failure=False)
processor.run()
pp.validate_features(processor,
nb_samples=12,
path=os.path.join(PATH_EXP, args.recipe),
override=True)
pp.base.DeleteFeatures(input_name=('stft', 'spec', 'sad_threshold')
),
pp.speech.AcousticNorm(mean_var_norm=True,
windowed_mean_var_norm=True,
input_name=('mspec', 'mfcc')),
# ====== post processing ====== #
pp.base.AsType(dtype='float16'),
],
debug=False)
with np.warnings.catch_warnings():
np.warnings.filterwarnings('ignore')
processor = pp.FeatureProcessor(jobs=all_files,
path=PATH_ACOUSTIC_FEATURES,
extractor=extractors,
n_cache=120,
ncpu=None,
override=True,
identifier='name',
log_path=os.path.join(
EXP_DIR, 'processor.log'),
stop_on_failure=True)
processor.run()
# pp.validate_features(processor,
# nb_samples=12,
# path=os.path.join(EXP_DIR, 'feature_validation'),
# override=True)
ds = F.Dataset(PATH_ACOUSTIC_FEATURES, read_only=True)
print(ds)
indices = list(ds['indices_%s' % args.feat].items())
print("Utterances length:")
print(" ",
describe([end - start for name, (start, end) in indices], shorten=True))
# ===========================================================================
# ====== basic path ====== #
output_dataset_path = os.path.join(PATH_ACOUSTIC_FEATURES, FEATURE_RECIPE)
processor_log_path = os.path.join(EXP_DIR, 'processor_%s.log' % FEATURE_RECIPE)
if os.path.exists(processor_log_path):
os.remove(processor_log_path)
print("Log path:", ctext(processor_log_path, 'cyan'))
ds_validation_path = os.path.join(EXP_DIR, 'validate_%s.pdf' % FEATURE_RECIPE)
if os.path.exists(ds_validation_path):
os.remove(ds_validation_path)
print("Validation path:", ctext(ds_validation_path, 'cyan'))
# ====== running the processing ====== #
with catch_warnings_ignore(Warning):
processor = pp.FeatureProcessor(jobs=ALL_FILES,
path=output_dataset_path,
extractor=recipe,
n_cache=320,
ncpu=NCPU,
override=True,
identifier='name',
log_path=processor_log_path,
stop_on_failure=False)
processor.run()
# ===========================================================================
# Make some visualization
# ===========================================================================
validate_features_dataset(output_dataset_path, ds_validation_path)
if not os.path.exists(outpath) or args.ds:
extractors = pp.make_pipeline(steps=[
pp.speech.AudioReader(sr=None,
sr_new=8000,
best_resample=True,
remove_dc=True),
pp.base.Converter(
converter=lambda x: os.path.basename(x).split('.')[0],
input_name='path',
output_name='name'),
pp.base.AsType(dtype='float16', input_name='raw')
],
debug=False)
processor = pp.FeatureProcessor(jobs=jobs,
path=outpath,
extractor=extractors,
n_cache=0.08,
ncpu=None,
override=True)
processor.run()
pp.validate_features(processor,
path='/tmp/tidigits',
nb_samples=12,
override=True)
with open(os.path.join(outpath, 'README'), 'w') as f:
f.write(README)
# ====== check the preprocessed dataset ====== #
ds = F.Dataset(outpath, read_only=True)
print(ds)
print(ctext(ds.md5, 'yellow'))
ds.close()
# ====== compress ====== #
continue
# extract acoustic feature from scratch
feat_dir = os.path.join(PATH_ACOUSTIC_FEATURES,
'%s_%s' % (dsname, EXTRACTOR_NAME))
log_path = get_logpath(name='%s_%s.log' % (dsname, EXTRACTOR_NAME),
increasing=True,
odin_base=False,
root=EXP_DIR)
# check if need running the feature extraction
if _check_running_feature_extraction(feat_dir, n_files=len(file_list)):
with np.warnings.catch_warnings():
np.warnings.filterwarnings('ignore')
processor = pp.FeatureProcessor(jobs=file_list,
path=feat_dir,
extractor=extractor,
ncpu=NCPU,
override=True,
identifier='name',
log_path=log_path,
stop_on_failure=False)
processor.run()
# store the extracted dataset
ds = F.Dataset(path=feat_dir, read_only=True)
assert FEATURE_NAME in ds, \
"Cannot find feature with name: %s, from: %s" % (FEATURE_NAME, ds.path)
acoustic_features[dsname] = [
ds[FEATURE_NAME],
dict(ds['indices_%s' % FEATURE_NAME].items()),
dict(ds['spkid'].items()),
dict(ds['path'].items()),
]
# ====== print log ====== #
exit()
else:
V.plot_multiple_features(tmp, title=name)
V.plot_save(os.path.join(PATH_EXP, 'feature_debug.pdf'))
exit()
# ===========================================================================
# Processor
# ===========================================================================
with np.warnings.catch_warnings():
np.warnings.filterwarnings('ignore')
processor = pp.FeatureProcessor(
jobs=all_files,
path=PATH_ACOUSTIC,
extractor=extractors,
n_cache=0.12,
ncpu=min(18,
cpu_count() - 2) if args.ncpu <= 0 else int(args.ncpu),
override=True,
identifier='name',
log_path=os.path.join(PATH_EXP, 'processor.log'),
stop_on_failure=True # small dataset, enable stop on failure
)
with UnitTimer():
processor.run()
n_error = len(processor.error_log)
print(processor)
# ====== copy readme and check the preprocessed dataset ====== #
if n_error == 0:
readme_path = os.path.join(
audio.path, [i for i in os.listdir(audio.path) if 'README' in i][0])
shutil.copy(readme_path, os.path.join(PATH_ACOUSTIC, 'README.md'))