pnpl.tasks.libribrain.SpeechDetection

class pnpl.tasks.libribrain.SpeechDetection(tmin=0.0, tmax=0.5, stride=None, oversample_silence_jitter=0, _classes=<factory>)

Binary speech vs silence classification task.

This task slides a time window across continuous MEG data and labels each window based on whether it contains speech or silence.

Parameters:

• tmin (float) – Start time of sample window relative to sliding position (seconds)

• tmax (float) – End time of sample window relative to sliding position (seconds)

• stride (int | None) – Step size for sliding window (samples). If None, uses window size.

• oversample_silence_jitter (int) – If > 0, oversample silence segments with this stride

• _classes (list)

__init__(tmin=0.0, tmax=0.5, stride=None, oversample_silence_jitter=0, _classes=<factory>)

Parameters:

• tmin (float)

• tmax (float)

• stride (int | None)

• oversample_silence_jitter (int)

• _classes (list)

Return type:

None

Methods

__init__([tmin, tmax, stride, ...])
collect_samples(dataset)	Collect speech/silence samples from all runs.
get_label(sample)	Extract label array from sample.

Attributes

label_info	Label metadata for binary classification.
oversample_silence_jitter
stride
tmax
tmin