NNSVS
github.com
- Free plan
- Windows
- macOS
- Linux
What NNSVS is
NNSVS is a free, open-source neural singing voice synthesis library intended for research. It lets users build voicebanks from their own datasets and includes reproducible Kaldi- and ESPnet-style recipes for preprocessing, training, and synthesis. Those workflows can be adapted to custom data and models. Packed models accept MusicXML or UST input and produce singing voice, with WAV listed as an export format. Its documented pipeline includes time-lag, duration, acoustic, post-filter, and vocoder modules. Users can create custom PyTorch models through the NNSVS BaseModel interface. Optuna handles hyperparameter optimization, while MLflow displays and tracks search results. The project documents systems made for eight or more languages. NNSVS supports Linux, Mac OS X, and Windows; Linux is recommended for development. Installation needs Python 3.7 or newer and a C/C++ compiler. GPU/CUDA is recommended, while Google Colab is available without a GPU but discouraged for hyperparameter optimization. Acoustic training may take several hours or a full day.
Who it suits
It is aimed at researchers and developers building customized singing synthesis systems. Non-researchers may prefer the more user-friendly tools named by the project.
Good to know, and what to watch
- Open source under the MIT license
- Custom voicebanks from user datasets
- Reproducible training and synthesis recipes
- MusicXML and UST inputs
- Optuna and MLflow integrations
- GPU/CUDA is recommended for performance
- Acoustic training can take hours or a day
- Post-filter is currently not recommended
- Neural-vocoder support is limited
The verdict
NNSVS gives technical users a configurable research workflow for singing voice synthesis. Its compute needs, long training times, and limited neural-vocoder support make it less suited to casual use.
What the maker publishes
- Compute
- GPU/CUDA is strongly recommended for performance, while Google Colab is listed as an option for users without GPUs. source
- Customization
- Users can implement custom PyTorch models by inheriting NNSVS's BaseModel interface. source
- Input formats
- Packed models can generate singing voice from MusicXML or UST files. source
- Inputs and outputs
- With a packed model, users can generate singing voice by inputting MusicXML or UST files. source
- Integrations
- NNSVS integrates Optuna for hyperparameter optimization and uses MLflow to visualize and track search results. source
- Intended users
- The project says it was originally designed for research and points non-researchers and non-developers to more user-friendly tools such as ENUNU and Simple-ENUNU. source
- Intended users
- The project says it was originally designed for research and recommends other tools for users who are neither researchers nor software developers. source
- Languages
- The documentation says NNSVS has been used to create singing voice synthesis systems for 8 or more languages. source
- Languages
- NNSVS has been used to create singing voice synthesis systems for multiple languages, with the documentation saying 8 or more as far as its author knows. source
- Licensing
- The project’s GitHub repository lists the MIT license. source
- Notable limitation
- The documentation says the post-filter is currently not recommended and neural-vocoder support is limited. source
- Open source
- NNSVS is fully open source and lets users create voicebanks with their own datasets. source
- Open source
- NNSVS is fully open-source, and users can create their own voicebanks with their dataset. source
- Optional compute
- Users without a GPU can run NNSVS on Google Colab, though the documentation does not recommend Colab for hyperparameter optimization. source
- Purpose
- NNSVS is a neural network based singing voice synthesis library designed for research. source
- Purpose
- NNSVS is a neural network based singing voice synthesis library designed for research. source
- Requirements
- The guide requires Python 3.7 or later and a C/C++ compiler; GPU/CUDA is recommended for best performance. source
- Requirements
- The guide specifies Python 3.7 or later and says a C/C++ compiler must be installed in advance. source
- Research recipes
- It includes reproducible Kaldi and ESPnet style recipes for creating baseline systems in research. source
- Research workflows
- NNSVS includes reproducible recipes modeled on Kaldi and ESPnet approaches for building baseline systems in research. source
- Supported operating systems
- The installation guide lists Linux, Mac OS X, and Windows as supported platforms and says Linux is recommended for development. source
- Supported platforms
- The installation guide lists Linux, Mac OS X, and Windows as supported and says NNSVS is tested on them with GitHub Actions. source
- Support information
- The opened project pages provide installation guides, tutorials, and API reference documentation; they do not state a dedicated support service. source
- Synthesis pipeline
- Its documented system includes time-lag, duration, acoustic, post-filter, and vocoder modules. source
Checked against the maker's pages on 8 Oct 2026.