WaveGrad 2: Iterative Refinement for Text-to-Speech Synthesis

TL;DR

WaveGrad 2 introduces an iterative, non-autoregressive approach to text-to-speech synthesis that generates high-quality audio by refining noise through multiple steps, offering a flexible trade-off between speed and quality.

Contribution

It presents WaveGrad 2, a novel iterative refinement model for TTS that improves upon previous vocoders by enabling flexible inference and high-fidelity audio generation.

Findings

High fidelity audio approaching state-of-the-art quality

Flexible inference speed via adjustable refinement steps

Effective ablation studies on model configurations

Abstract

This paper introduces WaveGrad 2, a non-autoregressive generative model for text-to-speech synthesis. WaveGrad 2 is trained to estimate the gradient of the log conditional density of the waveform given a phoneme sequence. The model takes an input phoneme sequence, and through an iterative refinement process, generates an audio waveform. This contrasts to the original WaveGrad vocoder which conditions on mel-spectrogram features, generated by a separate model. The iterative refinement process starts from Gaussian noise, and through a series of refinement steps (e.g., 50 steps), progressively recovers the audio sequence. WaveGrad 2 offers a natural way to trade-off between inference speed and sample quality, through adjusting the number of refinement steps. Experiments show that the model can generate high fidelity audio, approaching the performance of a state-of-the-art neural TTS system. We also report various ablation studies over different model configurations. Audio samples are available at https://wavegrad.github.io/v2.