DiffSinger: Diffusion Acoustic Model for Singing Voice Synthesis

By Jinglin Liu, Chengxi Li, Yi Ren, Feiyang Chen, Peng Liu, Zhou Zhao. 2021

32 citations recorded by the source corpus.

Subfields: Speech & Audio

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Speech & Audio

1. Zhou Zhao
2. Rongjie Huang
3. Yi Ren
4. Jinglin Liu
5. Chenye Cui
6. Songxiang Liu
7. Huadai Liu
8. Feiyang Chen

Citations in the corpus

• On Fast Sampling of Diffusion Probabilistic Models (2021, background): “, 2021), music synthesis (Liu et al., 2021; Mittal et al., 2021), 3-D point cloud generation (Luo & Hu, 2021; Zhou et al.”
• ProDiff: Progressive Fast Diffusion Model for High-Quality Text-to-Speech (2022, methodology): “For sampling, we use 64 steps in Grad-TTS and 128 steps in DiffSpeech, respectively, following [13] and [29]. denoising diffusion probabilistic models.”
• Multi-Singer: Fast Multi-Singer Singing Voice Vocoder With A Large-Scale Corpus (2021, unknown): “Singing voice synthesis (SVS) systems [2, 14, 22] take music score and lyric information as input to generate singing voices, and these systems have been widely deployed in music softwares, music boxes, and so on.”
• SingGAN: Generative Adversarial Network For High-Fidelity Singing Voice Generation (2021, methodology): “As illustrated in Figure 1, similar to text-to-speech (TTS) systems [5, 9, 22, 26, 29], SVS systems [2, 7, 17, 27] generally adopt an acoustic model to convert the musical score 3 into the intermediate features (i.”
• SRDiff: Single Image Super-Resolution with Diffusion Probabilistic Models (2021, unknown): “DiffSinger [22] applied the diffusion model to synthesize high-quality and expressive singing voice to tackle the over-smoothing and unstable training issues in the previous methods.”
• A Survey on Recent Deep Learning-driven Singing Voice Synthesis Systems (2021, methodology): “To further enhance the acoustic model’s prediction accuracy and robustness of Mel-spectrograms, DiffSinger [8] utilizes a generative model paradigm novel at the time of this writing: the denoising diffusion probabilistic model [9].”
• Implicit Diffusion Models for Continuous Super-Resolution (2023, unknown): “Based on these algorithms, [6, 14, 19] further achieve continuous image super-resolution with meta-learning or implicit neural representation.”
• DiffGAN-TTS: High-Fidelity and Efficient Text-to-Speech with Denoising Diffusion GANs (2022, methodology): “Acoustic reconstruction loss is also used as additional loss to train the acoustic generator following FastSpeech2 (Ren et al., 2021a), as: where d , p and e are target duration, pitch and energy, respectively, and ˆd , ˆp and ˆe are their corresponding predicted values. λ d , λ p and λ e are loss w”