Replies: 1 comment 1 reply
-
Thanks for the info! I'll test it when I have time :) As for the hanging words, that's often a side effect of WhisperX. Sometimes there are halizinations. I have a few ideas to "reduce" them. |
Beta Was this translation helpful? Give feedback.
1 reply
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
-
Hi Rakuri,
let me start to give compliments about this brilliant software!
Works really well for songs with clear vocals and we have spent hours singing the Ultrasinger songs!!
When songs have less clear vocals, I started noticing big differences between the current version (Version: 0.0.9 Date: 2024.02.06) and an older version. In various songs (both EN and NL) It seems the transcription goes out of sync and keeps repeating words.
Example of the (sorry, Dutch) song at this url: https://www.youtube.com/watch?v=z9E4Nl2d60U
This is a very difficult song to transcribe because of the heavy processing, so some errors are to be expected.
However, the transcription with the new version of Ultrasinger goes completely of the rails.
Old version of Ultrasinger does a much better job:
Linda, Roos & Jessica - Ademnood.txt
New version keeps "hanging" in the la la la la and ohohohoh:
Linda, Roos & Jessica - Ademnood NEW.txt
Command:
py UltraSinger.py -i https://www.youtube.com/watch?v=z9E4Nl2d60U --language nl --disable_karaoke True
Old version source:
UltraSinger-main.zip
Old version console output:
console_output_old.txt
New version source:
UltraSinger-main (1).zip
New version console output:
console_output_new.txt
What could be happening? (please note I am testing the transcription with the old and new version right after eachther on the same system, with the same command. System should be no problem, with Core i5 13600k, 64 GB RAM and RTX 4070 Ti Super 16 GB.
Thnx,
DJP!
Beta Was this translation helpful? Give feedback.
All reactions