OWSM v4 models, trained on a cleaned 166k-hour multilingual YODAS subset, beat prior open OWSM models and are competitive with Whisper and MMS on several benchmarks.
We reveal that large-scale web-crawled data contains incorrect lan- guage labels and audio-text misalignments
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
OWSM v4: Improving Open Whisper-Style Speech Models via Data Scaling and Cleaning
OWSM v4 models, trained on a cleaned 166k-hour multilingual YODAS subset, beat prior open OWSM models and are competitive with Whisper and MMS on several benchmarks.