Autoregressive Speech Synthesis without Vector Quantization

TL;DR

MELLE introduces a continuous-valued token approach for text-to-speech synthesis that improves fidelity, robustness, and diversity by avoiding vector quantization and employing novel regression and variational inference techniques.

Contribution

The paper proposes MELLE, a new autoregressive TTS model that generates continuous spectrograms directly, bypassing vector quantization, and incorporates a spectrogram flux loss and variational inference for better performance.

Findings

MELLE outperforms VALL-E in multiple metrics.

It mitigates robustness issues associated with vector-quantized codes.

The model offers a more streamlined TTS paradigm.

Abstract

We present MELLE, a novel continuous-valued token based language modeling approach for text-to-speech synthesis (TTS). MELLE autoregressively generates continuous mel-spectrogram frames directly from text condition, bypassing the need for vector quantization, which is typically designed for audio compression and sacrifices fidelity compared to continuous representations. Specifically, (i) instead of cross-entropy loss, we apply regression loss with a proposed spectrogram flux loss function to model the probability distribution of the continuous-valued tokens; (ii) we have incorporated variational inference into MELLE to facilitate sampling mechanisms, thereby enhancing the output diversity and model robustness. Experiments demonstrate that, compared to the two-stage codec language model VALL-E and its variants, the single-stage MELLE mitigates robustness issues by avoiding the inherent flaws of sampling vector-quantized codes, achieves superior performance across multiple metrics, and, most importantly, offers a more streamlined paradigm. The demos of our work are provided at https://aka.ms/melle.