The THU-HCSI Multi-Speaker Multi-Lingual Few-Shot Voice Cloning System for LIMMITS'24 Challenge
Yixuan Zhou, Shuoyi Zhou, Shun Lei, Zhiyong Wu, Menglin Wu
TL;DR
The paper tackles multi-speaker multilingual few-shot voice cloning for LIMMITS'24 by extending YourTTS with VITS2-inspired enhancements including a speaker-aware text encoder, a flow-based decoder with Transformer blocks, and a noise-injected MAS. The approach combines extensive data preprocessing, denoising of few-shot data, and a training regime that mixes pretraining data with few-shot samples alongside a speaker-balanced fine-tuning strategy, enabling effective adaptation to 9 target speakers. In track 1 evaluations, the system achieves a speaker similarity MOS of 4.25 and naturalness MOS of 3.97, ranking first in speaker similarity and demonstrating strong performance in cross-lingual and mono-lingual synthesis. The work advances practical few-shot voice cloning for low-resource, multilingual settings by integrating architectural enhancements, data processing, and balanced fine-tuning to deliver high speaker similarity and competitive naturalness.
Abstract
This paper presents the multi-speaker multi-lingual few-shot voice cloning system developed by THU-HCSI team for LIMMITS'24 Challenge. To achieve high speaker similarity and naturalness in both mono-lingual and cross-lingual scenarios, we build the system upon YourTTS and add several enhancements. For further improving speaker similarity and speech quality, we introduce speaker-aware text encoder and flow-based decoder with Transformer blocks. In addition, we denoise the few-shot data, mix up them with pre-training data, and adopt a speaker-balanced sampling strategy to guarantee effective fine-tuning for target speakers. The official evaluations in track 1 show that our system achieves the best speaker similarity MOS of 4.25 and obtains considerable naturalness MOS of 3.97.