.lrc filetypes for download

[mknepper]

The ability to download lyrics from the database in .lrc would be pretty neat. It'd make things easier when using stuff like Audacious or other music programs that scrape lyrics.

[snshn]

It could be problematic, since this database was initially meant to be used by humans, that is why the plaintext format with a little bit of metadata at the end was chosen, and I still like that format myself.

How hard do you think it would be to have a piece of software automatically time and match lines from plaintext files with when it's in the song, and attaching the timestamp to it? That's the only easy way to generate .lrc out of these files that comes to my mind.

[C0rn3j]

Hard, but not infeasible.

---

VLC is doing almost the same, including the transcription, with models from whisper. That's aimed at shows/movies/videos though

https://www.omgubuntu.co.uk/2025/01/vlc-demo-ai-real-time-subtitles-for-any-video

Songs however have a lot of continuous audio and a lot of noise, reverb, layered sounds needing an arbiter to decide what to keep, etc.

So whatever VLC will end up releasing may not be the best tool for the job, time will tell when they release.

You can try making raw whisper do it and fixing the rest I suppose - openai/whisper#98

There's also Gentle which seems to be also within the LLM approach - https://github.com/itsvaibhav01/Gentle

Subaligner seems actually maintained - https://github.com/baxtree/subaligner

---

There seem to be (mostly dead) old-time non-LLM projects that target to do this exact thing, like https://pypi.org/project/py3-aeneas/, but I have not had the time to check them out yet.

Would require the song source + plaintext lyrics, so the end result would best be stored in the repo, we can talk formats when we figure out how to achieve it in the first place.

It does look like an imperfect solution for music, unfortunately - https://news.ycombinator.com/item?id=13908579

---

Karaoke-level lyrics look like this in LRC (I am sure the good japanese karaoke ones look even crazier since I've seen people do every damn syllable in .ASS subtitles):

[00:53.60]On a dark [00:54.85]desert highway[00:56.26] so that does not look the best for human readability, but hey, the website can simply show the parsed plaintext of this.

---

It is a question of whether we want to settle on LRC though, as something like ASS(I did not pick the acronym) might be more appropriate as it has styling abilities and definitely a good karaoke support.