Parallel WaveGAN: A fast waveform generation model based on generative adversarial networks with multi-resolution spectrogram

TL;DR

Parallel WaveGAN introduces a fast, non-autoregressive speech waveform generation model that achieves high fidelity and speed without the need for distillation, making it suitable for real-time applications.

Contribution

It presents a novel, distillation-free GAN-based approach for waveform generation that is compact, efficient, and maintains high speech quality.

Findings

Generates 24 kHz speech 28.68 times faster than real-time.

Achieves a MOS of 4.16 in TTS framework, comparable to distillation-based systems.

Uses only 1.44 million parameters, demonstrating efficiency.

Abstract

We propose Parallel WaveGAN, a distillation-free, fast, and small-footprint waveform generation method using a generative adversarial network. In the proposed method, a non-autoregressive WaveNet is trained by jointly optimizing multi-resolution spectrogram and adversarial loss functions, which can effectively capture the time-frequency distribution of the realistic speech waveform. As our method does not require density distillation used in the conventional teacher-student framework, the entire model can be easily trained. Furthermore, our model is able to generate high-fidelity speech even with its compact architecture. In particular, the proposed Parallel WaveGAN has only 1.44 M parameters and can generate 24 kHz speech waveform 28.68 times faster than real-time on a single GPU environment. Perceptual listening test results verify that our proposed method achieves 4.16 mean opinion score within a Transformer-based text-to-speech framework, which is comparative to the best distillation-based Parallel WaveNet system.