SEF-VC: Speaker Embedding Free Zero-Shot Voice Conversion with Cross Attention
This work addresses voice conversion for arbitrary unseen speakers, offering an incremental improvement by removing speaker embeddings and enhancing similarity.
The paper tackles the problem of zero-shot voice conversion by proposing SEF-VC, a model that eliminates the need for speaker embeddings and uses cross-attention to incorporate speaker timbre from reference speech, resulting in better speaker similarity than strong baselines, even with short references.
Zero-shot voice conversion (VC) aims to transfer the source speaker timbre to arbitrary unseen target speaker timbre, while keeping the linguistic content unchanged. Although the voice of generated speech can be controlled by providing the speaker embedding of the target speaker, the speaker similarity still lags behind the ground truth recordings. In this paper, we propose SEF-VC, a speaker embedding free voice conversion model, which is designed to learn and incorporate speaker timbre from reference speech via a powerful position-agnostic cross-attention mechanism, and then reconstruct waveform from HuBERT semantic tokens in a non-autoregressive manner. The concise design of SEF-VC enhances its training stability and voice conversion performance. Objective and subjective evaluations demonstrate the superiority of SEF-VC to generate high-quality speech with better similarity to target reference than strong zero-shot VC baselines, even for very short reference speeches.