Nightingale

Point Nightingale at your music and it builds the karaoke track itself. The app scans a folder, a Plex Media Server, Jellyfin, Navidrome or a self hosted web library, separates lead vocals from the instrumental, and transcribes lyrics with word level timestamps.

It ships as a single binary. Python, ffmpeg and the machine learning models are downloaded and bootstrapped on first launch, so nothing needs installing by hand before you sing.

During playback the app does several things at once.

• Pitch scoring: live microphone input, pitch detection, star ratings and per song scoreboards.
• Synchronised lyrics: word by word highlighting, with romanised readings for Japanese, Chinese, Cantonese and Korean.
• Key and tempo shifts: adjust a track after analysis, with cached variants for quick retries.
• Audio reactive backgrounds: ten GPU shaders that respond to the microphone, plus video loops.
• Video files: drop in a video and its own picture plays behind the separated vocals.

Lyrics can also come from LRCLIB or from an LRC file you paste in, and a built in editor fixes timings. The engines are pluggable, with Whisper or an experimental Parakeet model for transcription and either the default aligner or GPU forced alignment for timestamps.

Desktop builds cover Linux, macOS and Windows, and a self hosted web mode runs it on a home server so phones, tablets, laptops and TVs join over the local network. Docker images exist for CPU and CUDA machines.