Saved in:
Bibliographic Details
Main Authors: Wagner, Laurin, Thallinger, Bernhard, Zusag, Mario
Format: Preprint
Published: 2024
Subjects:
Online Access:https://arxiv.org/abs/2408.16589
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866914928902799360
author Wagner, Laurin
Thallinger, Bernhard
Zusag, Mario
author_facet Wagner, Laurin
Thallinger, Bernhard
Zusag, Mario
contents We demonstrate that carefully adjusting the tokenizer of the Whisper speech recognition model significantly improves the precision of word-level timestamps when applying dynamic time warping to the decoder's cross-attention scores. We fine-tune the model to produce more verbatim speech transcriptions and employ several techniques to increase robustness against multiple speakers and background noise. These adjustments achieve state-of-the-art performance on benchmarks for verbatim speech transcription, word segmentation, and the timed detection of filler events, and can further mitigate transcription hallucinations. The code is available open https://github.com/nyrahealth/CrisperWhisper.
format Preprint
id arxiv_https___arxiv_org_abs_2408_16589
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle CrisperWhisper: Accurate Timestamps on Verbatim Speech Transcriptions
Wagner, Laurin
Thallinger, Bernhard
Zusag, Mario
Machine Learning
We demonstrate that carefully adjusting the tokenizer of the Whisper speech recognition model significantly improves the precision of word-level timestamps when applying dynamic time warping to the decoder's cross-attention scores. We fine-tune the model to produce more verbatim speech transcriptions and employ several techniques to increase robustness against multiple speakers and background noise. These adjustments achieve state-of-the-art performance on benchmarks for verbatim speech transcription, word segmentation, and the timed detection of filler events, and can further mitigate transcription hallucinations. The code is available open https://github.com/nyrahealth/CrisperWhisper.
title CrisperWhisper: Accurate Timestamps on Verbatim Speech Transcriptions
topic Machine Learning
url https://arxiv.org/abs/2408.16589