OmniVoice: Towards Omnilingual Zero-Shot Text-to-Speech with Diffusion Language Models

TL;DR

OmniVoice is a novel zero-shot multilingual TTS model using a diffusion language model architecture, capable of synthesizing speech in over 600 languages with state-of-the-art quality.

Contribution

It introduces a direct text-to-acoustic token mapping with innovative training strategies, significantly expanding language coverage and improving TTS performance.

Findings

Achieves state-of-the-art results in Chinese, English, and multilingual benchmarks.

Supports over 600 languages with high-quality speech synthesis.

Utilizes a large open-source dataset of 581k hours for training.

Abstract

We present OmniVoice, a massively multilingual zero-shot text-to-speech (TTS) model that scales to over 600 languages. At its core is a novel diffusion language model-style discrete non-autoregressive (NAR) architecture. Unlike conventional discrete NAR models that suffer from performance bottlenecks in complex two-stage (text-to-semantic-to-acoustic) pipelines, OmniVoice directly maps text to multi-codebook acoustic tokens. This simplified approach is facilitated by two key technical innovations: (1) a full-codebook random masking strategy for efficient training, and (2) initialization from a pre-trained LLM to ensure superior intelligibility. By leveraging a 581k-hour multilingual dataset curated entirely from open-source data, OmniVoice achieves the broadest language coverage to date and delivers state-of-the-art performance across Chinese, English, and diverse multilingual benchmarks. Our code and pre-trained models are publicly available at https://github.com/k2-fsa/OmniVoice.