Abstract:Automatic speech recognition (ASR) performs well for high-resource languages with abundant paired audio-transcript data, but its accuracy degrades sharply for most languages due to limited publicly available aligned data. To this end, we introduce WorldSpeech, a 24 kHz multilingual speech corpus comprising 65k hours of aligned audio-transcript data across 76 languages, collected from diverse public sources including parliamentary proceedings, international broadcasts, and public-domain audiobooks. For 37 languages, WorldSpeech provides more than 200 hours of aligned speech, with 28 exceeding 500 hours and 24 surpassing 1k hours. Fine-tuning existing ASR models on WorldSpeech results in an average relative Word-Error-Rate reduction of 63.5% across 11 typologically diverse languages.
| Subjects: | Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Machine Learning (cs.LG) |
| Cite as: | arXiv:2605.09167 [cs.CL] |
| (or arXiv:2605.09167v1 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2605.09167 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Luca Lanzendörfer [view email]
[v1]
Sat, 9 May 2026 21:00:01 UTC (184 KB)