Speech Codec Probing from Semantic and Phonetic Perspectives
This addresses a critical bottleneck for connecting speech to LLMs in multimodal systems by revealing a semantic-phonetic mismatch in current tokenizers.
The paper systematically analyzed what information speech tokenizers encode, finding they primarily capture phonetic rather than lexical-semantic structure, which creates a mismatch with text-derived semantics that can degrade multimodal LLM performance.
Speech tokenizers are essential for connecting speech to large language models (LLMs) in multimodal systems. These tokenizers are expected to preserve both semantic and acoustic information for downstream understanding and generation. However, emerging evidence suggests that what is termed "semantic" in speech representations does not align with text-derived semantics: a mismatch that can degrade multimodal LLM performance. In this paper, we systematically analyze the information encoded by several widely used speech tokenizers, disentangling their semantic and phonetic content through word-level probing tasks, layerwise representation analysis, and cross-modal alignment metrics such as CKA. Our results show that current tokenizers primarily capture phonetic rather than lexical-semantic structure, and we derive practical implications for the design of next-generation speech tokenization methods.