DualVC 2: Dynamic Masked Convolution for Unified Streaming and Non-Streaming Voice Conversion

Ziqian Ning^{1, 2}, Yuepeng Jiang¹, Pengcheng Zhu², Shuai Wang³, Jixun Yao¹, Lei Xie¹, Mengxiao Bi² ¹Audio, Speech and Language Processing Group (ASLP@NPU), School of Computer Science,
Northwestern Polytechnical University, Xi'an, China ²Fuxi AI Lab, NetEase Inc., Hangzhou, China ³Shenzhen Research Institute of Big Data,
The Chinese University of Hong Kong, Shenzhen (CUHK-Shenzhen), China

1. Abstract

Voice conversion is becoming increasingly popular, and a growing number of application scenarios require models with streaming inference capabilities. The recently proposed DualVC attempts to achieve this objective through streaming model architecture design and intra-model knowledge distillation along with hybrid predictive coding to compensate for the lack of future information. However, DualVC encounters several problems that limit its performance. First, the autoregressive decoder has error accumulation in its nature and limits the inference speed as well. Second, the causal convolution enables streaming capability but cannot sufficiently use future information within chunks. Third, the model is unable to effectively address the noise in the unvoiced segments, lowering the sound quality. In this paper, we propose DualVC 2 to address these issues. Specifically, the model backbone is migrated to a Conformer-based architecture, empowering parallel inference. Causal convolution is replaced by non-causal convolution with dynamic chunk mask to make better use of within-chunk future information. Also, quiet attention is introduced to enhance the model's noise robustness. Experiments show that DualVC 2 outperforms DualVC and other baseline systems in both subjective and objective metrics, with only 186.4 ms latency.

2. Demo -- Real-time voice conversion in real RTC scenario

3. Demo -- Recordings

DualVC nonstreaming: Non-streaming mode of our previous proposed model [1] .
DualVC streaming: Streaming mode of our previous proposed model [1] .
IBF-VC: Reinplement [2] using the backbone model.
DualVC 2 nonstreaming: Non-streaming mode of DualVC 2.
DualVC 2 streaming: Streaming mode of DualVC 2.

[1] Z. Ning, Y. Jiang, P. Zhu, J. Yao, S. Wang, L. Xie, and M. Bi, “Dualvc: Dual-mode voice conversion using intra-model knowledge distillation and hybrid predictive coding,” CoRR, vol. abs/2305.12425, 2023.
[2] Y. Chen, M. Tu, T. Li, X. Li, Q. Kong, J. Li, Z. Wang, Q. Tian, Y. Wang, and Y. Wang, “Streaming voice conversion via intermediate bottleneck features and non-streaming teacher guidance,” CoRR, vol. abs/2210.15158, 202