pnpl.tasks.gwilliams2022.PhonemeClassification

pnpl.tasks.gwilliams2022.PhonemeClassification#

class pnpl.tasks.gwilliams2022.PhonemeClassification(tmin=-0.2, tmax=0.6, label_type='phoneme', exclude_phonemes=<factory>, require_pronounced=True, _phonemes_sorted=<factory>, _phoneme_to_id=<factory>)[source]#

Multi-class phoneme classification on MEG-MASC.

Each sample is aligned to a phoneme onset extracted from the events file. Labels are ARPABET phonemes (after mapping from TIMIT). Sample tuples follow the continuous-data convention used elsewhere in pnpl: (subject, session, task, run, onset, label_str).

Parameters:

  • tmin (float) – Start time relative to phoneme onset (seconds).

  • tmax (float) – End time relative to phoneme onset (seconds).

  • label_type (str) – "phoneme" for multi-class, "voicing" for binary voiced/unvoiced.

  • exclude_phonemes (list) – ARPABET symbols to drop from samples and the label vocabulary.

  • require_pronounced (bool) – If True, drop events with pronounced set to a falsy value (e.g. silently-read trials).

  • _phonemes_sorted (list)

  • _phoneme_to_id (dict)

__init__(tmin=-0.2, tmax=0.6, label_type='phoneme', exclude_phonemes=<factory>, require_pronounced=True, _phonemes_sorted=<factory>, _phoneme_to_id=<factory>)#
Parameters:

  • tmin (float)

  • tmax (float)

  • label_type (str)

  • exclude_phonemes (list)

  • require_pronounced (bool)

  • _phonemes_sorted (list)

  • _phoneme_to_id (dict)

Return type:

None

Methods

__init__([tmin, tmax, label_type, ...])

collect_samples(dataset)

get_label(sample)

Attributes

label_info

label_type

require_pronounced

tmax

tmin

exclude_phonemes