banjuman/vilm-lyrics-aligner — explained in plain English
Analysis updated 2026-05-18
Generate accurately timed SRT subtitles for a song using your own lyrics text.
Handle mixed Korean and English lyric timing that standard dictation tools get wrong.
Create karaoke-style or lyric video subtitles directly inside DaVinci Resolve Studio.
Fine-tune subtitle cue timing manually for live or expressive vocal performances.
| banjuman/vilm-lyrics-aligner | 0xallam/my-recipe | 0xhassaan/nn-from-scratch | |
|---|---|---|---|
| Stars | 0 | — | 0 |
| Language | Python | Python | Python |
| Last pushed | — | 2022-11-22 | — |
| Maintenance | — | Dormant | — |
| Setup difficulty | moderate | moderate | moderate |
| Complexity | 3/5 | 2/5 | 4/5 |
| Audience | general | general | developer |
Figures from each repo's GitHub metadata at analysis time.
First setup downloads several gigabytes of AI models and requires an internet connection.
Vilm Lyrics Aligner creates precisely timed subtitle files for songs, matching original lyrics text to audio or video. It is built for tricky cases that ordinary speech-to-text often gets wrong: live singing, mixed Korean and English lyrics, and vocals where automatic dictation tends to misjudge timing. The lyrics you type or paste in are always used as the exact subtitle text, and the audio recognition is only used to figure out when each line should appear, never to rewrite your words. The tool works by first separating vocals from the instrumental accompaniment, then using a multilingual version of the Whisper speech recognition model as timing evidence, especially useful when lyrics switch between languages mid line. You can process a whole song or select just a section of the waveform, and it preserves the original timing offset for that section. There are two modes: an automatic mode that uses line breaks and pauses to guess good subtitle boundaries, recommended as the starting point, and a manual mode for advanced users that treats every lyric line as an intentional cue with finer control over character count, duration, and how long a line lingers before disappearing. It ships as a standalone desktop application for Windows and Apple silicon Macs, with an optional plugin panel for DaVinci Resolve Studio, a professional video editing program. Other video editors can still use it by importing the generated subtitle file. The installer sets up its own private Python and AI runtime separate from anything else on your machine, which can take several gigabytes of disk space and an internet connection on first setup. It uses GPU acceleration where available, either NVIDIA CUDA on Windows or Apple's Metal on Mac, falling back to the CPU otherwise. The README is upfront that this is a singing-focused alignment tool rather than a general transcription replacement, and that unusual vocal techniques like humming or drawn out notes may still need a manual timing correction afterward. The source code is released under the MIT License, though the Vilm name and visual branding are covered separately under a trademark policy.
A desktop tool that generates precisely timed subtitle files for songs by matching your own lyrics to the audio's real timing.
Mainly Python. The stack also includes Python, Whisper, PyTorch.
Source code is MIT licensed, free to use and modify, but the Vilm name and branding are covered by a separate trademark policy.
Setup difficulty is rated moderate, with roughly 30min to a first successful run.
Mainly general.
This repo across BitVibe Labs
Don't trust strangers blindly. Verify against the repo.