The article presents research on a neural vocoder based on a denoising diffusion probabilistic model (DDPM), which incorporates explicit periodic signals as auxiliary conditioning signals. The proposed model can generate high-quality waveforms and offers improved pitch control. This development is particularly significant for applications such as speech and singing voice synthesis. The study suggests that this model outperforms conventional DDPM-based neural vocoders in terms of sound quality and pitch control.
Publication date: 23 Feb 2024
Project Page: Not provided
Paper: https://arxiv.org/pdf/2402.14692
