Improving Zero-shot Voice Style Transfer via Disentangled Representation Learning
This addresses the challenge of generating realistic voice conversions for new speakers without paired training data, which is incremental over prior work on known speakers.
The paper tackles the problem of zero-shot voice style transfer, which modifies a speaker's voice to sound like an unseen target speaker using non-parallel data, and achieves state-of-the-art results in transfer accuracy and voice naturalness on the VCTK dataset.
Voice style transfer, also called voice conversion, seeks to modify one speaker's voice to generate speech as if it came from another (target) speaker. Previous works have made progress on voice conversion with parallel training data and pre-known speakers. However, zero-shot voice style transfer, which learns from non-parallel data and generates voices for previously unseen speakers, remains a challenging problem. We propose a novel zero-shot voice transfer method via disentangled representation learning. The proposed method first encodes speaker-related style and voice content of each input voice into separated low-dimensional embedding spaces, and then transfers to a new voice by combining the source content embedding and target style embedding through a decoder. With information-theoretic guidance, the style and content embedding spaces are representative and (ideally) independent of each other. On real-world VCTK datasets, our method outperforms other baselines and obtains state-of-the-art results in terms of transfer accuracy and voice naturalness for voice style transfer experiments under both many-to-many and zero-shot setups.