Neural Codec Language Models are Zero-Shot Text to Speech Synthesizers

TLDR

The authors introduce a language‑modeling approach for text‑to‑speech synthesis. They train a neural codec language model, Vall‑E, on 60 k hours of English speech, treating TTS as a conditional language‑modeling task using discrete codec codes. Vall‑E demonstrates strong in‑context learning, producing high‑quality personalized speech from a 3‑second prompt, and outperforms the current state‑of‑the‑art zero‑shot TTS in naturalness and speaker similarity while preserving the prompt’s emotion and acoustic environment. See https://aka.ms/valle for demos.

Abstract

We introduce a language modeling approach for text to speech synthesis (TTS). Specifically, we train a neural codec language model (called Vall-E) using discrete codes derived from an off-the-shelf neural audio codec model, and regard TTS as a conditional language modeling task rather than continuous signal regression as in previous work. During the pre-training stage, we scale up the TTS training data to 60K hours of English speech which is hundreds of times larger than existing systems. Vall-E emerges in-context learning capabilities and can be used to synthesize high-quality personalized speech with only a 3-second enrolled recording of an unseen speaker as an acoustic prompt. Experiment results show that Vall-E significantly outperforms the state-of-the-art zero-shot TTS system in terms of speech naturalness and speaker similarity. In addition, we find Vall-E could preserve the speaker's emotion and acoustic environment of the acoustic prompt in synthesis. See https://aka.ms/valle for demos of our work.