Learning Noise-independent Speech Representation for High-quality Voice Conversion for Noisy Target Speakers

TL;DR

This paper introduces a noise-independent speech representation learning method using Glow-WaveGAN for high-quality voice conversion from noisy target speakers, effectively handling contaminated speech data.

Contribution

It proposes a novel noise-invariant latent feature learning approach with a noise-controllable WaveGAN and flow-based model, improving voice conversion quality with noisy inputs.

Findings

Achieves high speech quality in noisy voice conversion

Maintains speaker similarity despite noise interference

Demonstrates robustness with contaminated speech data

Abstract

Building a voice conversion system for noisy target speakers, such as users providing noisy samples or Internet found data, is a challenging task since the use of contaminated speech in model training will apparently degrade the conversion performance. In this paper, we leverage the advances of our recently proposed Glow-WaveGAN and propose a noise-independent speech representation learning approach for high-quality voice conversion for noisy target speakers. Specifically, we learn a latent feature space where we ensure that the target distribution modeled by the conversion model is exactly from the modeled distribution of the waveform generator. With this premise, we further manage to make the latent feature to be noise-invariant. Specifically, we introduce a noise-controllable WaveGAN, which directly learns the noise-independent acoustic representation from waveform by the encoder and conducts noise control in the hidden space through a FiLM module in the decoder. As for the conversion model, importantly, we use a flow-based model to learn the distribution of noise-independent but speaker-related latent features from phoneme posteriorgrams. Experimental results demonstrate that the proposed model achieves high speech quality and speaker similarity in the voice conversion for noisy target speakers.