NNSVS
Open-source PyTorch toolkit for neural-network-based singing voice synthesis research.
About
NNSVS is a research toolkit for building singing voice synthesis systems from scratch, inspired by the open-source pioneer Sinsy but fully reimplemented and extended. Its modular pipeline covers time-lag, duration, acoustic, and vocoder models, with multi-stream architectures, autoregressive fundamental-frequency modeling, and modern neural vocoders like uSFGAN and HiFi-GAN. It ships Kaldi/ESPnet-style recipes and extensive docs for reproducible experiments.
The toolkit (paper arXiv 2210.15987, October 2022, from LINE Corp and Nagoya University researchers) is used by VocalSynth communities to build voicebanks in 8+ languages, and friendlier spin-offs like ENUNU exist for non-researchers. It is MIT licensed, pip-installable, and its best systems reportedly beat Sinsy reproductions in listening tests, though the toolkit is explicitly research-oriented rather than a consumer app.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Author-reported (arXiv 2210.15987): best NNSVS system significantly outperforms reproduced Sinsy baselines in listening tests
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 7/10
Scored 7/10 (metrics tier, metrics-v1). Evidence: peer-reviewed toolkit paper (arXiv 2210.15987, Oct 2022) from LINE Corp / Nagoya University; official repo github.com/nnsvs/nnsvs with MIT license badge and docs at nnsvs.github.io; community voicebanks in 8+ languages confirm real-world reuse. Strengths: fully open, reproducible recipes, strong community adoption in SVS research, permissive license. Docked for: (1) research-oriented and not beginner-friendly (ENUNU exists for that gap); (2) maintenance cadence is slow; (3) quality depends heavily on user-supplied training data. License: MIT.
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 7.0/10
- Maintenance · recency, releases, issue load
- 6.0/10
- Openness · license + docs
- 9.0/10
- Manual review · sanity read + red-flag check
- 7.0/10
- · Paper arXiv 2210.15987 (Oct 2022), LINE Corp / Nagoya University
- · Community voicebanks in 8+ languages; ENUNU friendlier spin-off
- · Multi-stream + autoregressive F0, neural vocoders (uSFGAN, HiFi-GAN)