def extract(feature1: AcousticFeature, feature2: AcousticFeature, dtype='int64'):
aligner = MelCepstrumAligner(feature1.mc, feature2.mc)
indexes1 = (aligner.normed_path_x * len(feature1.mc)).astype(dtype)
indexes2 = (aligner.normed_path_y * len(feature2.mc)).astype(dtype)
align_indexes = AlignIndexes(
feature1=feature1,
feature2=feature2,
indexes1=indexes1,
indexes2=indexes2,
)
return align_indexes
def generate_feature(path1, path2):
out1 = Path(arguments.output1_directory, path1.stem + '.npy')
out2 = Path(arguments.output2_directory, path2.stem + '.npy')
if out1.exists() and out2.exists() and not arguments.enable_overwrite:
return
# load wave and padding
wave_file_load_process = WaveFileLoadProcess(
sample_rate=arguments.sample_rate,
top_db=arguments.top_db,
pad_second=arguments.pad_second,
)
wave1 = wave_file_load_process(path1, test=True)
wave2 = wave_file_load_process(path2, test=True)
# make acoustic feature
acoustic_feature_process1 = AcousticFeatureProcess(
frame_period=arguments.frame_period,
order=arguments.order,
alpha=arguments.alpha,
f0_estimating_method=arguments.f0_estimating_method,
f0_floor=arguments.f0_floor1,
f0_ceil=arguments.f0_ceil1,
)
acoustic_feature_process2 = AcousticFeatureProcess(
frame_period=arguments.frame_period,
order=arguments.order,
alpha=arguments.alpha,
f0_estimating_method=arguments.f0_estimating_method,
f0_floor=arguments.f0_floor2,
f0_ceil=arguments.f0_ceil2,
)
f1 = acoustic_feature_process1(wave1,
test=True).astype_only_float(numpy.float32)
f2 = acoustic_feature_process2(wave2,
test=True).astype_only_float(numpy.float32)
# pre convert
if pre_convert:
f1_ref = pre_converter1.convert_to_feature(f1)
else:
f1_ref = f1
# alignment
if not arguments.disable_alignment:
aligner = MelCepstrumAligner(f1_ref.mfcc, f2.mfcc)
f0_1, f0_2 = aligner.align(f1.f0, f2.f0)
spectrogram_1, spectrogram_2 = aligner.align(f1.spectrogram,
f2.spectrogram)
aperiodicity_1, aperiodicity_2 = aligner.align(f1.aperiodicity,
f2.aperiodicity)
mfcc_1, mfcc_2 = aligner.align(f1.mfcc, f2.mfcc)
voiced_1, voiced_2 = aligner.align(f1.voiced, f2.voiced)
f1 = AcousticFeature(
f0=f0_1,
spectrogram=spectrogram_1,
aperiodicity=aperiodicity_1,
mfcc=mfcc_1,
voiced=voiced_1,
)
f2 = AcousticFeature(
f0=f0_2,
spectrogram=spectrogram_2,
aperiodicity=aperiodicity_2,
mfcc=mfcc_2,
voiced=voiced_2,
)
f1.validate()
f2.validate()
# save
acoustic_feature_save_process = AcousticFeatureSaveProcess(
validate=True, ignore=arguments.ignore_feature)
acoustic_feature_save_process({'path': out1, 'feature': f1})
print('saved!', out1)
acoustic_feature_save_process({'path': out2, 'feature': f2})
print('saved!', out2)
def generate_feature(path1, path2):
out1 = Path(arguments.output1_directory, path1.stem + '.npy')
out2 = Path(arguments.output2_directory, path2.stem + '.npy')
if out1.exists() and out2.exists() and not arguments.enable_overwrite:
return
# load wave and padding
wave_file_load_process = WaveFileLoadProcess(
sample_rate=arguments.sample_rate,
top_db=arguments.top_db,
pad_second=arguments.pad_second,
)
wave1 = wave_file_load_process(path1, test=True)
wave2 = wave_file_load_process(path2, test=True)
# make acoustic feature
acoustic_feature_process1 = AcousticFeatureProcess(
frame_period=arguments.frame_period,
order=arguments.order,
alpha=arguments.alpha,
f0_estimating_method=arguments.f0_estimating_method,
f0_floor=arguments.f0_floor1,
f0_ceil=arguments.f0_ceil1,
)
acoustic_feature_process2 = AcousticFeatureProcess(
frame_period=arguments.frame_period,
order=arguments.order,
alpha=arguments.alpha,
f0_estimating_method=arguments.f0_estimating_method,
f0_floor=arguments.f0_floor2,
f0_ceil=arguments.f0_ceil2,
)
f1 = acoustic_feature_process1(wave1, test=True).astype_only_float(numpy.float32)
f2 = acoustic_feature_process2(wave2, test=True).astype_only_float(numpy.float32)
# pre convert
if pre_convert:
f1_ref = pre_converter1.convert_to_feature(f1)
else:
f1_ref = f1
# alignment
if not arguments.disable_alignment:
aligner = MelCepstrumAligner(f1_ref.mfcc, f2.mfcc)
f0_1, f0_2 = aligner.align(f1.f0, f2.f0)
spectrogram_1, spectrogram_2 = aligner.align(f1.spectrogram, f2.spectrogram)
aperiodicity_1, aperiodicity_2 = aligner.align(f1.aperiodicity, f2.aperiodicity)
mfcc_1, mfcc_2 = aligner.align(f1.mfcc, f2.mfcc)
voiced_1, voiced_2 = aligner.align(f1.voiced, f2.voiced)
f1 = AcousticFeature(
f0=f0_1,
spectrogram=spectrogram_1,
aperiodicity=aperiodicity_1,
mfcc=mfcc_1,
voiced=voiced_1,
)
f2 = AcousticFeature(
f0=f0_2,
spectrogram=spectrogram_2,
aperiodicity=aperiodicity_2,
mfcc=mfcc_2,
voiced=voiced_2,
)
f1.validate()
f2.validate()
# save
acoustic_feature_save_process = AcousticFeatureSaveProcess(validate=True, ignore=arguments.ignore_feature)
acoustic_feature_save_process({'path': out1, 'feature': f1})
print('saved!', out1)
acoustic_feature_save_process({'path': out2, 'feature': f2})
print('saved!', out2)