Improving Zero-shot Voice Style Transfer via Disentangled Representation Learning
arXiv:2103.09420
Abstract
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.
To appear in ICLR 2021
References in corpus (9)
- Instance Normalization: The Missing Ingredient for Fast Stylization
- VoxCeleb: a large-scale speaker identification dataset
- Variational Inference with Normalizing Flows
- Voice Recognition Algorithms using Mel Frequency Cepstral Coefficient (MFCC) and Dynamic Time Warping (DTW) Techniques
- Challenging Common Assumptions in the Unsupervised Learning of Disentangled Representations
- Estimating divergence functionals and the likelihood ratio by convex risk minimization
- Deep Variational Information Bottleneck
- Ensemble estimation of multivariate f-divergence
- Learning Controllable Fair Representations