Speaking Clearly: A Simplified Whisper-Based Codec for Low-Bitrate Speech Coding

TL;DR

This paper introduces SimWhisper-Codec, a low-bitrate speech codec that uses a simplified, semantic-first approach based on Whisper, achieving better semantic and acoustic quality without external supervision.

Contribution

It presents a novel semantic-first approach that simplifies Whisper for high-fidelity acoustic reconstruction, outperforming semantically-supervised codecs.

Findings

SimWhisper-Codec outperforms Mimi Codec and SpeechTokenizer in both semantic and acoustic metrics.

Targeted architectural simplification enhances Whisper's acoustic modeling capabilities.

The approach does not require external supervision, simplifying the training process.

Abstract

Speech codecs serve as bridges between continuous speech signals and large language models, yet face an inherent conflict between acoustic fidelity and semantic preservation. To mitigate this conflict, prevailing methods augment acoustic codecs with complex semantic supervision. We explore the opposite direction: a semantic-first approach that starts from a semantically-capable model and adapts it for high-fidelity acoustic reconstruction. Through empirical analysis, we discover that targeted architectural simplification can unlock the acoustic modeling potential of Whisper, a text-aligned Automatic Speech Recognition (ASR) model. Based on this finding, we propose SimWhisper-Codec, a novel codec that balances the semantic and acoustic preservation by leveraging a frozen, simplified Whisper encoder without requiring external supervision. Experimental results demonstrate that SimWhisper-Codec achieves superior performance in both semantic preservation and acoustic quality compared to semantically-supervised codecs such as Mimi Codec and SpeechTokenizer at similar bitrates, validating the effectiveness of our semantic-first approach. Code is available at https://github.com/ZhangXinWhut/SimWhisper-Codec.