ynop / spych Goto Github PK

Add functions to the dataset class to generate batches. Iteration through utterances randomly and create batches of utterances (one batch --> list of utterances).

Add possibility to add features to the dataset. So an utterance can have different features (same principle as with segmentations). Every type of feature can be stored with a key (e.g. mfcc, fbank, ...).

dataset.features['utt_id']['key'] = feature

Import ctm, audacity label files into a dataset.

# import a ctm file as segmentation (ctm first value of line --> utterance-id)
spych dataset-import [dataset-path] --ctm [path-to-ctm-file]

# import a audacity label file as segmentation for a given utterance
spych dataset-import [dataset-path] --audacity [path-to-ac-file] --utterance-idx [utt-id]

Otherwise also delete the speaker.