FeaturesExtractor

class features_extractor.FeaturesExtractor(audio_filename_structure=None, feature_filename_structure=None, sampling_frequency=None, lower_frequency=None, higher_frequency=None, filter_bank=None, filter_bank_size=None, window_size=None, shift=None, ceps_number=None, vad=None, snr=None, pre_emphasis=None, save_param=None, keep_all_features=None, feature_type=None, rasta_plp=None, compressed='percentile')[source]

A FeaturesExtractor process an audio file in SPHERE, WAVE or RAW PCM format and extract filter-banks, cepstral coefficients, bottle-neck features (in the future), log-energy and perform a speech activity detection.

extract(show, channel, input_audio_filename=None, output_feature_filename=None, backing_store=False, noise_file_name=None, snr=10, reverb_file_name=None, reverb_level=- 26.0)[source]

Compute the acoustic parameters (filter banks, cepstral coefficients, log-energy and bottleneck features for a single channel from a given audio file.

Parameters

  • show – ID if the show

  • channel – channel number (0 if mono file)

  • input_audio_filename – name of the input audio file to consider if the name of the audio file is independent from the ID of the show

  • output_feature_filename – name of the output feature file to consider if the name of the feature file is independent from the ID of the show

  • backing_store – boolean, if False, nothing is writen to disk, if True, the file is writen to disk when closed

  • rasta – boolean, only for PLP parameters, if True, perform RASTA filtering

Returns

an hdf5 file handler

extract_from_signal(signal, sample_rate, noise_file_name=None, snr=10, reverb_file_name=None, reverb_level=- 26.0)[source]

Compute the acoustic parameters (filter banks, cepstral coefficients, log-energy and bottleneck features for a single channel from a given audio file.

Parameters

  • show – ID if the show

  • channel – channel number (0 if mono file)

  • input_audio_filename – name of the input audio file to consider if the name of the audio file is independent from the ID of the show

  • output_feature_filename – name of the output feature file to consider if the name of the feature file is independent from the ID of the show

  • backing_store – boolean, if False, nothing is writen to disk, if True, the file is writen to disk when closed

  • rasta – boolean, only for PLP parameters, if True, perform RASTA filtering

Returns

an hdf5 file handler

save(show, channel=0, input_audio_filename=None, output_feature_filename=None, noise_file_name=None, snr=10, reverb_file_name=None, reverb_level=- 26.0)[source]

Compute the acoustic parameters (filter banks, cepstral coefficients, log-energy and bottleneck features for a single channel from a given audio file and save them to disk in a HDF5 format

Parameters

  • show –

  • channel –

  • input_audio_filename –

  • output_feature_filename –

Returns

save_list(**kwargs)
Parameters

  • args –

  • kwargs –

Returns

save_multispeakers(idmap, channel=0, input_audio_filename=None, output_feature_filename=None, keep_all=True, skip_existing_file=False, compressed='percentile')[source]
Parameters

  • idmap –

  • channel –

  • input_audio_filename –

  • output_feature_filename –

  • keep_all –

  • skip_existing_file –

Returns