ENUNU

ENUNU is not a new model. It is the packaging layer that turned NNSVS (Nagoya University's open neural singing synthesis toolkit, MIT) into something hobbyists could actually use. Originally built by oatsu-gh as a UTAU plugin powered by NNSVS, ENUNU became the standard way to run NNSVS voicebanks inside OpenUtau: you launch a small server (the official ENUNU-0.4.0 build or community forks like SimpleEnunu, which adds WORLD-diffusion and mel-diffusion vocoders) and render AI singing directly from the piano roll, with automatic timing correction and NNSVS-generated pitch. OpenUtau's own docs call ENUNU the most-used distro of NNSVS and treat the two names as near-synonyms. Because it inherits NNSVS's architecture, ENUNU supports multi-stream acoustic models, autoregressive F0 modeling and neural vocoders, and users can train their own voicebanks from their own singing. It is MIT-licensed and free.

• Inherits NNSVS's published results: best NNSVS system reached 3.86 MOS in the paper's listening tests, significantly over the reproduced Sinsy baseline (author-reported)
• No ENUNU-specific benchmarks published; quality depends on the voicebank used