Tags: bnosac/audio.whisper
Tags
Allow to transcribe only sections of the audio (#57) * add argument sections to predict.whisper * code cleanup * adding example of sections * add extra data check on subset.wav such that offsets/durations are not outside of audio range * VAD on mono * Align the timestamps based on the removed audio segments again * play around with 1 millisecond * more examples on multiple-offsets * add NEWS item, bump version * unit test for multiple sections return the part where it has a voice instead of the non-voice and put thses skipped parts at the top * CI * put alignment timestamps of transcriptions with skipped segments in separate function * add utils * docs * CI * docs * dev code for token_timestamps * CI * docs * CI * use format instead of timezone * parse the timestamps with milliseconds * make sure to parse out the milliseconds + test with token_timestamp * rename sentences to voiced for later readability * bump dependency version of data.table to when data.table contains nafill * there is no guarantee on short segments that whisper.cpp adheres to the requested timeframe - it provides the whole segment - disable that unit test * docs * CI * add segment_offset to keep track of within which offset the segment was part of * add offset/duration as arguments in predict.whisper, include new column called segment_offset in output by default, make sure examples on stereo are run in language es * increase duration to make sure we have a segment * logs * more liberal unit test * more liberal unit test
PreviousNext