PAST: Phonetic-Acoustic Speech Tokenizer

TL;DR

PAST is an innovative end-to-end speech tokenizer that integrates phonetic information directly, outperforming existing models in speech representation and reconstruction, and supports real-time applications.

Contribution

It introduces a supervised, joint phonetic-acoustic tokenization framework with a causal variant for real-time use, eliminating reliance on pretrained models.

Findings

Outperforms baseline tokenizers in phonetic and reconstruction metrics.

Effective as a speech representation for language models.

Supports real-time speech processing.

Abstract

We present PAST, a novel end-to-end framework that jointly models phonetic information alongside signal reconstruction, eliminating the need for external pretrained models. Unlike previous approaches that rely on pretrained self-supervised models, PAST employs supervised phonetic data, directly integrating domain knowledge into the tokenization process via auxiliary tasks. Additionally, we introduce a streamable, causal variant of PAST, enabling real-time speech applications. Results demonstrate that PAST surpasses existing evaluated baseline tokenizers across common evaluation metrics, including phonetic representation and speech reconstruction. Notably, PAST also achieves superior performance when serving as a speech representation for speech language models, further highlighting its effectiveness as a foundation for spoken language generation. To foster further research, we release the full implementation. For code, model checkpoints, and samples see: https://pages.cs.huji.ac.il/adiyoss-lab/PAST