Neural network based singing voice synthesis library


  • Open-source: NNSVS is fully open-source. You can create your own voicebanks with your dataset.

  • Multiple languages: NNSVS has been used for creating singing voice synthesis (SVS) systems for multiple languages by VocalSynth comminities (8+ as far as I know).

  • Research friendly: NNSVS comes with reproducible Kaldi/ESPnet-style recipes. You can use NNSVS to create baseline systems for your research.


Note that NNSVS was originally designed for research purposes. Please check out more user-friendly tools below if you are neither a researcher nor a software developer.

You can find a practical guide for NNSVS/ENUNU at https://nnsvs.carrd.co/ (by xuu). A detailed tutorial for for making voice banks can be found at NNSVS Database Making Tutorial (by PixProcuer).

Selected samples

Diffusion + SiFiGAN for Ritsu (2023/03/12)

Diffusion + SiFiGAN for Cipher (2023/03/14)

Autoregressive model + uSFGAN (2022/09/24)

You can find more from the NNSVS/ENUNU community: YouTube, NicoNico



Meta information

Indices and tables