Credits & Licenses

Utavie uses the following open source software. Thank you to all the developers and researchers behind it.

AI & Machine Learning

Sony CSL ParisReal-time pitch estimation AI model (130K parameters, 98.3% RPA on MIR-1K)

SpotifyNeural-network-based automatic music transcription (AMT)

Anjok07 / UVR CommunityOn-device vocal separation model (9482 LiteRT as the default, Kim_Vocal_2 ONNX kept for PC compatibility)

LiteRT.jsApache 2.0

GoogleOn-device AI inference runtime shared across PC, Android, and iOS (WebGPU / WebAssembly)

ZFTurbo (model) / bgkb (ONNX)In-browser vocal separation model (WebGPU, 103MB ONNX)

anvuewLead vocal extraction model (converted to ONNX with unmodified weights, WebGPU, 104MB)

MicrosoftIn-browser AI model inference engine (WebAssembly/WebGL)

Hugging FaceIn-browser Whisper speech recognition (automatic lyrics generation)

GoogleIn-browser machine learning runtime (powers Basic Pitch inference)

Peter RudenkoYIN pitch detection algorithm (fallback)

Dexie.jsApache 2.0

David FahlanderIndexedDB wrapper (persists score history, recordings, and training records)

LRCLIB CommunityLyrics database (used for song lookup)

takuyaaJapanese morphological analysis (used for lyrics reading conversion)

IPAdic辞書IPAdic License (BSD系)

Taku Kudo / MeCab ProjectMorphological analysis dictionary for kuromoji.js (commercial use allowed)

Lucide ContributorsIcon library

VercelReact framework (App Router)

MetaUI library

Tailwind LabsUtility-first CSS framework