Easily train a good VC model with voice data <= 10 mins!
-
Updated
Apr 18, 2026 - Python
Easily train a good VC model with voice data <= 10 mins!
Singing voice conversion built on LeapSinger's harmonic excitation and rectified flow: content + F0 + loudness -> mel -> NHVSing. Quality is read as a gap from the vocoder ceiling (Japanese, incl. unseen speakers). No weights distributed - training-data licensing.
Singing voice conversion baseline combining SoftVC-style acoustic modeling, RMVPE pitch conditioning, and HiFi-GAN synthesis.
Singing voice conversion powered by HiFiSinger and RefineGAN.
To associate your repository with the rmvpe topic, visit your repo's landing page and select "manage topics."