Research map / Image generation / Autoregressive and latent generation
M4Singer: A Multi-Style, Multi-Singer and Musical Score Provided Mandarin Singing Corpus
Lichao Zhang, Ruiqi Li, Shoutong Wang, Liqun Deng, Jinglin Liu, Yi Ren, Jinzheng He, Rongjie Huang, Jieming Zhu, Xiao Chen, Zhou Zhao
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use ground-truth F0 as input (ablation) | FFT-Singer · M4Singer SVS | F0-MAE: 8.1011 -> 1.9321 | Improved |
| use ground-truth F0 as input (ablation) | GAN-Singer · M4Singer SVS | F0-MAE: 8.5263 -> 2.1545 | Improved |
| use ground-truth F0 as input (ablation) | Diffsinger · M4Singer SVS | F0-MAE: 8.9990 -> 1.7061 | Improved |
| use ground-truth F0 as input (ablation) | Diffsinger · M4Singer SVS | MOS-P: 3.72 ± 0.15 -> 4.02 ± 0.11 | Improved |
| use ground-truth F0 as input (ablation) | Diffsinger · M4Singer SVS | MOS-Q: 3.92 ± 0.13 -> 3.97 ± 0.10 | No clear effect |
| use ground-truth F0 as input (ablation) | FFT-Singer · M4Singer SVS | MOS-P: 3.57 ± 0.13 -> 3.75 ± 0.12 | Improved |
| replace FFT acoustic model with generative models (GAN, diffusion) | FFT-Singer · M4Singer SVS | MOS-Q: 3.52 ± 0.14 -> 3.92 ± 0.13 | Improved |
| replace FFT acoustic model with diffusion | FFT-Singer · M4Singer SVS | F0-MAE: 8.1011 -> 8.9990 | Hurt |
| shift pitch through score instead of F0 | controllable singing voice · M4Singer random pitch shift | MOS-P: 3.51 ± 0.12 -> 3.63 ± 0.15 | Improved |
| shift pitch through score instead of F0 | controllable singing voice · M4Singer random pitch shift | MOS-S: 3.58 ± 0.11 -> 3.62 ± 0.13 | No clear effect |
| shift across more voice types (bass to soprano) | controllable singing voice · M4Singer voice type shift | CMOS: -0.03 -> -1.24 | Hurt |
| add pitch shift to score (ablation) | Diffsinger SVC, female target, male score source · M4Singer SVC | MOS-P: 3.12 ± 0.16 -> 3.67 ± 0.12 | Improved |
Closest papers by meaning
- GTSinger: A Global Multi-Technique Singing Corpus with Realistic Music Scores for All Singing Tasks (NeurIPS 2024 · similarity 0.89)
- FT-GAN: Fine-Grained Tune Modeling for Chinese Opera Synthesis (AAAI 2024 · similarity 0.86)
- SongSong: A Time Phonograph for Chinese SongCi Music from Thousand of Years Away (AAAI 2025 · similarity 0.86)
- CSSinger: End-to-End Chunkwise Streaming Singing Voice Synthesis System Based on Conditional Variational Autoencoder (AAAI 2025 · similarity 0.85)
- SongGen: A Single Stage Auto-regressive Transformer for Text-to-Song Generation (ICML 2025 · similarity 0.85)
- TechSinger: Technique Controllable Multilingual Singing Voice Synthesis via Flow Matching (AAAI 2025 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
