A Spectral Energy Distance for Parallel Speech Synthesis

TL;DR

This paper introduces a spectral energy distance method for training parallel speech synthesis models, achieving high-quality audio generation without requiring likelihood functions or adversarial training, and demonstrating state-of-the-art results.

Contribution

The paper proposes a novel spectral energy distance for training implicit speech models, enabling stable, parallel, and likelihood-free learning with statistical guarantees.

Findings

Achieves state-of-the-art quality among implicit models using cFDSD metric.

Improves GAN-TTS scores when combined with adversarial techniques.

Provides a stable, bias-free training method without adversarial learning.

Abstract

Speech synthesis is an important practical generative modeling problem that has seen great progress over the last few years, with likelihood-based autoregressive neural models now outperforming traditional concatenative systems. A downside of such autoregressive models is that they require executing tens of thousands of sequential operations per second of generated audio, making them ill-suited for deployment on specialized deep learning hardware. Here, we propose a new learning method that allows us to train highly parallel models of speech, without requiring access to an analytical likelihood function. Our approach is based on a generalized energy distance between the distributions of the generated and real audio. This spectral energy distance is a proper scoring rule with respect to the distribution over magnitude-spectrograms of the generated waveform audio and offers statistical consistency guarantees. The distance can be calculated from minibatches without bias, and does not involve adversarial learning, yielding a stable and consistent method for training implicit generative models. Empirically, we achieve state-of-the-art generation quality among implicit generative models, as judged by the recently-proposed cFDSD metric. When combining our method with adversarial techniques, we also improve upon the recently-proposed GAN-TTS model in terms of Mean Opinion Score as judged by trained human evaluators.