Work

Using the Mimi codec for metalinguistic representations

Year: 2026

Type: preprint

Abstract: In this paper, we focus on the dictionary of 2048 tokens used in Mimi semantic token codebook, the neural codec of the Moshi language model. We show that the ABX experiment carried out with Mimi fails to capture the mapping of the semantic tokens to phone realisations. By realigning Mimi representations to the TIMIT corpus transcriptions, we show that the 2048 tokens IDs of the semantic codebook map to quadphone, triphone, biphone, phone and subphone realisations.

Source: arXiv (Cornell University)

DOI: 10.48550/arxiv.2608.15799

Authors: Saloev, Artem, Pacquetet, Erin, Ballier, Nicolas

Institutions: Université Paris Cité, Approches Linguistiques Théoriques, Appliquées et Expérimentales : langues et cultures connectées

Language: English

Topic: Speech Recognition and Synthesis

Subfield: Artificial Intelligence

Field: Computer Science

Domain: Physical Sciences

Open Access status: green

Funders: Agence Nationale de la Recherche, Université de Paris

Awards: G1963894250, G4656594441, G4707887143, Université de Paris

OpenAlex ID: W7203727782

Cites:

Cited by:

Related to: 10