Feature request: Romanized lyrics for non-Latin script songs (e.g., Japanese songs displayed in romaji)

[inuryuutarou]

Feature request: When generating lyrics or transcripts for songs in non-Latin scripts (e.g., Chinese, Japanese, Korean), add an option to display the text in romanized form (e.g., Pinyin, Romaji, Romanized Korean). Additionally, if possible allow users to provide their own existing lyrics so the tool has less work to do and can focus on other tasks.

[rzru]

hey @inuryuutarou, thanks! I'll think of it!

[inuryuutarou]

Thanks for the response. I also found some minor problems with songs that have mixed languages — the generated lyrics became only one language, and the timing and placement were also wrong.

[rzru]

@inuryuutarou yeah, unfortunately that's not something I can fix programmatically since that's how models work. I'm planning to introduce a wider variety of models as well as the internal editor to fix these minor issues in the future, so stay tuned!

[inuryuutarou]

Thanks again for the response. I was actively testing it at the time, and those were the results I found from actually trying it out. The time it took to generate timing for a single song was quite long, which is why I suggested the idea of providing the lyrics in advance—so the model would only need to handle the timing.

[rzru]

@inuryuutarou that's actually already how it works, I'm trying to find the lyrics in https://lrclib.net/ b4 generating a transcript, however now it just picks the best compatible lyrics w/o giving you the option to choose from the options or provide your own. that's already in my feature backlog, but it's not the biggest priority. I will definitely ship it tho, so bear with me