Papers
arxiv:2601.14046

PRiSM: Benchmarking Phone Realization in Speech Models

Published on Jan 20
· Submitted by
Shikhar Bharadwaj
on Jan 21
Authors:
,
,
,
,
,
,
,
,
,

Abstract

PRiSM benchmark evaluates phonetic perception in speech models through standardized transcription-based metrics and downstream applications across clinical, educational, and multilingual domains.

AI-generated summary

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.

Community

Paper author Paper submitter
This comment has been hidden (marked as Resolved)
Paper author Paper submitter
edited about 16 hours ago

Main take-aways

  1. PRiSM is the first fully-open benchmark that evaluates Phone-Recognition systems on both intrinsic (phone-transcription) and extrinsic (down-stream) tasks across 12 datasets covering clinical, L2-learning and multilingual settings. We find that Large Audio-Language Models still lag behind specialized PR models on such tasks.

  2. Since intrinsic phone recognition capability is not fully indicative of performance in extrinsic settings, we design transcript and representation based probes that allow an exhaustive analysis, interpretability, and fair comparison.

  3. Language exposure > data size: multilingual training with broad, diverse data matters more for cross lingual generalization.

Code, prompts and data are released under permissive licences.

Sign up or log in to comment

Models citing this paper 2

Datasets citing this paper 0

No dataset linking this paper

Cite arxiv.org/abs/2601.14046 in a dataset README.md to link it from this page.

Spaces citing this paper 0

No Space linking this paper

Cite arxiv.org/abs/2601.14046 in a Space README.md to link it from this page.

Collections including this paper 1