PRiSM: Benchmarking Phone Realization in Speech Models

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Bharadwaj, Shikhar, Li, Chin-Jou, Kim, Yoonjae, Choi, Kwanghee, Yeo, Eunjung, Shim, Ryan Soh-Eun, Zhou, Hanyu, Boldt, Brendon, Jacome, Karen Rosero, Chang, Kalvin, Agrawal, Darsh, Xu, Keer, Yang, Chao-Han Huck, Zhu, Jian, Watanabe, Shinji, Mortensen, David R.
Formato: Preprint
Publicado: 2026
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866908776447082496
author Bharadwaj, Shikhar
Li, Chin-Jou
Kim, Yoonjae
Choi, Kwanghee
Yeo, Eunjung
Shim, Ryan Soh-Eun
Zhou, Hanyu
Boldt, Brendon
Jacome, Karen Rosero
Chang, Kalvin
Agrawal, Darsh
Xu, Keer
Yang, Chao-Han Huck
Zhu, Jian
Watanabe, Shinji
Mortensen, David R.
author_facet Bharadwaj, Shikhar
Li, Chin-Jou
Kim, Yoonjae
Choi, Kwanghee
Yeo, Eunjung
Shim, Ryan Soh-Eun
Zhou, Hanyu
Boldt, Brendon
Jacome, Karen Rosero
Chang, Kalvin
Agrawal, Darsh
Xu, Keer
Yang, Chao-Han Huck
Zhu, Jian
Watanabe, Shinji
Mortensen, David R.
contents Phone recognition (PR) serves as the atomic interface for language-agnostic modeling for cross-lingual speech processing and phonetic analysis. Despite prolonged efforts in developing PR systems, current evaluations only measure surface-level transcription accuracy. We introduce PRiSM, the first open-source benchmark designed to expose blind spots in phonetic perception through intrinsic and extrinsic evaluation of PR systems. PRiSM standardizes transcription-based evaluation and assesses downstream utility in clinical, educational, and multilingual settings with transcription and representation probes. We find that diverse language exposure during training is key to PR performance, encoder-CTC models are the most stable, and specialized PR models still outperform Large Audio Language Models. PRiSM releases code, recipes, and datasets to move the field toward multilingual speech models with robust phonetic ability: https://github.com/changelinglab/prism.
format Preprint
id arxiv_https___arxiv_org_abs_2601_14046
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle PRiSM: Benchmarking Phone Realization in Speech Models
Bharadwaj, Shikhar
Li, Chin-Jou
Kim, Yoonjae
Choi, Kwanghee
Yeo, Eunjung
Shim, Ryan Soh-Eun
Zhou, Hanyu
Boldt, Brendon
Jacome, Karen Rosero
Chang, Kalvin
Agrawal, Darsh
Xu, Keer
Yang, Chao-Han Huck
Zhu, Jian
Watanabe, Shinji
Mortensen, David R.
Computation and Language
Sound
Phone recognition (PR) serves as the atomic interface for language-agnostic modeling for cross-lingual speech processing and phonetic analysis. Despite prolonged efforts in developing PR systems, current evaluations only measure surface-level transcription accuracy. We introduce PRiSM, the first open-source benchmark designed to expose blind spots in phonetic perception through intrinsic and extrinsic evaluation of PR systems. PRiSM standardizes transcription-based evaluation and assesses downstream utility in clinical, educational, and multilingual settings with transcription and representation probes. We find that diverse language exposure during training is key to PR performance, encoder-CTC models are the most stable, and specialized PR models still outperform Large Audio Language Models. PRiSM releases code, recipes, and datasets to move the field toward multilingual speech models with robust phonetic ability: https://github.com/changelinglab/prism.
title PRiSM: Benchmarking Phone Realization in Speech Models
topic Computation and Language
Sound
url https://arxiv.org/abs/2601.14046