An Empirical Recipe for Universal Phone Recognition
Fuente:
arXiv
Enregistré dans:
| Auteurs principaux: | , , , , , , |
|---|---|
| Format: | Preprint |
| Publié: |
2026
|
| Sujets: | |
| Accès en ligne: | |
| Tags: |
Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
|
| _version_ | 1866911558502711296 |
|---|---|
| author | Bharadwaj, Shikhar Li, Chin-Jou Choi, Kwanghee Yeo, Eunjung Chen, William Watanabe, Shinji Mortensen, David R. |
| author_facet | Bharadwaj, Shikhar Li, Chin-Jou Choi, Kwanghee Yeo, Eunjung Chen, William Watanabe, Shinji Mortensen, David R. |
| contents | Phone recognition (PR) is a key enabler of multilingual and low-resource speech processing tasks, yet robust performance remains elusive. Highly performant English-focused models do not generalize across languages, while multilingual models underutilize pretrained representations. It also remains unclear how data scale, architecture, and training objective contribute to multilingual PR. We present PhoneticXEUS -- trained on large-scale multilingual data and achieving state-of-the-art performance on both multilingual (17.7% PFER) and accented English speech (10.6% PFER). Through controlled ablations with evaluations across 100+ languages under a unified scheme, we empirically establish our training recipe and quantify the impact of SSL representations, data scale, and loss objectives. In addition, we analyze error patterns across language families, accented speech, and articulatory features. All data and code are released openly. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2603_29042 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | An Empirical Recipe for Universal Phone Recognition Bharadwaj, Shikhar Li, Chin-Jou Choi, Kwanghee Yeo, Eunjung Chen, William Watanabe, Shinji Mortensen, David R. Computation and Language Machine Learning Sound Audio and Speech Processing Phone recognition (PR) is a key enabler of multilingual and low-resource speech processing tasks, yet robust performance remains elusive. Highly performant English-focused models do not generalize across languages, while multilingual models underutilize pretrained representations. It also remains unclear how data scale, architecture, and training objective contribute to multilingual PR. We present PhoneticXEUS -- trained on large-scale multilingual data and achieving state-of-the-art performance on both multilingual (17.7% PFER) and accented English speech (10.6% PFER). Through controlled ablations with evaluations across 100+ languages under a unified scheme, we empirically establish our training recipe and quantify the impact of SSL representations, data scale, and loss objectives. In addition, we analyze error patterns across language families, accented speech, and articulatory features. All data and code are released openly. |
| title | An Empirical Recipe for Universal Phone Recognition |
| topic | Computation and Language Machine Learning Sound Audio and Speech Processing |
| url | https://arxiv.org/abs/2603.29042 |