Sign In:     

Forum: VirtualDJ Technical Support

Topic: Lyrics issues with stems VDJ 2026
Since my post was hijacked and lost with other stuff i am reposing here

when i use this option on this menu in the image no lyrics get generated can we get that changed I only like to use stems this way thanks
 
Posted Mon 08 Dec 25 @ 11:32 pm
I don't mean to highjack this thread, but it has the most descriptive name already for general issues with the AI lyrics functionality.


  1. The lyrics scanning seems to happen after stems analysis
    I'm guessing it depends on the audio signature logic which depends on stems, but that adds to analysis time
  2. Lyrics scanning seems to be serial
    If you have multiple tracks on deck, the analysis is done in FIFO order
    Not generally a problem if the analysis is fast, but that isn't true in all cases
  3. Analysis seems to take a very long time/not work for some files
    For example, I have the following song as a video file and an audio file, and in both cases, it never seems to complete (I can provide the files if needed):
    Busta Rhymes - Pass the Courvasier Pt II
  4. If a file never completed lyric analysis, and you somehow stop the analysis (including a restart), subsequent loads of that track to a deck does not trigger lyrics analysis

    This was unexpected, as a quick trip to the Lyrics editor shows it as not containing any info, so I'm guess the analysis may include checking audio signature generation or some other flag that gets set (if it is, it doesn't rescan for lyrics, even if there aren't actual lyrics present)


If you combine 2 and 3, over time this means that the feature appears to stop working to a user.

4 seems like a bug.

I have getLyrics set to always, and I'm on Mac (if that matters).

Minor suggestion - include a context menu option for lyrics analysis just under the regular BPM analysis (and probably a lyrics analysis verb too) - having to go to the Lyrics Editor to trigger the analysis is an extra step and doesn't allow for batch processing.
 
So I tried again today and the same song I identified worked almost immediately, so I assume it was a central server congestion issue (I still had to manually trigger lyrics re-analyze).

Still doubling up on the local run if possible suggestion, although at the expense of a larger download/more space usage, it will always be more reliable.

I also noticed that the lyricsCensoredWords field does apply, but only in direct matches on word boundaries. Plural cases or substring matches don't work, so restatement of possible occurences of the same word has to be done.

Would it possible to make that field accept regex strings for matching to try to condense that field?
 
Yeah I noticed that with motherf****r - you can't just enter the f word and have it leave mother. You have to either enter the whole mf as one word, or I guess put a space between m and f in the lyrics.
 
I assume that's an easy fix for them. I'm more concerned with occassional lyrics analysis not working (or seeming to not work).
It's possible to pre-plan sets in advance to ensure lyrics are present, but if on any given day it just stalls, then reliability is in trouble.
 
I guess it will always be "use at your own risk" because you can enter a whole bunch of words to censor, but if you play something that has another word that you didn't think of, it won't censor.

Another scenario is phrases with words that on their own are innocent, but together would be rude. Suck is one example. Maybe fine in "that would suck", but not in "suck my d***".
 
It's all about context though. Some songs are suggestive like Lick it by 20 fingers but not explicit as such.
 
Tbh the algorithm has all that context - it has all the lyrics around every word (and that's the power LLMs bring), so it should be able to differentiate by using a window of words, not just words as they occur.
 
thanks for fixing with latest build :-)
 
For a minute there I thought the "reading" of the lyrics had improved (with maybe internet look up rather than 100% AI) but then I tried it with Louie Louie (The Kingsmen)....