Textless Streaming Speech-to-Speech Translation using Semantic Speech Tokens

TL;DR

This paper introduces a low-latency, streaming speech translation model that directly produces speech tokens, avoiding text generation and reducing error accumulation and latency in speech-to-speech translation.

Contribution

A novel transducer-based model that outputs discrete speech tokens directly for streaming translation, outperforming existing methods in latency and translation quality.

Findings

Outperforms existing approaches in BLEU scores

Achieves state-of-the-art latency and quality metrics

Effective across multiple language pairs

Abstract

Cascaded speech-to-speech translation systems often suffer from the error accumulation problem and high latency, which is a result of cascaded modules whose inference delays accumulate. In this paper, we propose a transducer-based speech translation model that outputs discrete speech tokens in a low-latency streaming fashion. This approach eliminates the need for generating text output first, followed by machine translation (MT) and text-to-speech (TTS) systems. The produced speech tokens can be directly used to generate a speech signal with low latency by utilizing an acoustic language model (LM) to obtain acoustic tokens and an audio codec model to retrieve the waveform. Experimental results show that the proposed method outperforms other existing approaches and achieves state-of-the-art results for streaming translation in terms of BLEU, average latency, and BLASER 2.0 scores for multiple language pairs using the CVSS-C dataset as a benchmark.