https://uberduck.ai/ logo
Powered by
# machine-learning
  • c
    You can change wav 0 to 81
  • c
    In the code I assumed people would start from 1
  • h
    ...why does your code assume filenames at all
  • c
    Because if I didn't do that it would translate the numbers in the filenames
  • h
    that's what the splitting step is for
  • h
    it splits at the |
  • h
    so you can then process just the bit after the |
  • c
    Makes sense, but I don't know why it isn't working when there is a 0
  • c
    I may have to check
  • c
    I'm new to python and I'm learning things yet, as I'm learning by myself my code gets a little bit messy, but I'll eventually get better at this
  • h
    fair enough
  • t
    @hecko Got anything for English voice datasts besides MLP? I just realised it may be an issue for me moving forward with my voice cloning work.
  • t
    I figured out the metadata issue--there are files in each speaker's folder that have to be renamed
  • t
    That being said, it needs to be formatted in a specific way and whatever the MLP dataset does is completely unorthodox
  • t
    And I'm not about to rewrite it
  • t
    Unless...let me check again
  • u
    how bad will the voice sound if i can still hear faint instrumentals in the source? lalala.ai couldnt get rid of everything unfortunately
  • u
    I don't think it'll have a significant effect
  • t
    ...yeah, either I need to get some kind of metadata collection for the MLP audio or I need a brand new English dataset
  • t
    Which has been tough for me to add to Google Drive
  • t
    Though I think I remember a collection I can use
  • s
    for the making datasets, I mean it could ruin it
  • s
    but for resource audio
  • s
    not really
  • h
    if the bg audio isnt too loud then itll work
  • h
    just make sure to clean up the audio so that the duration timings can detect stuff
  • u
    gonna be dumb question but what's resource audio? the stuff you can optionally put on talknet?
  • h
    your honor, league of legends
  • h
    also overwatch which is a smaller dataset but pre-transcribed
  • h
    and both of those also have spanish i believe
1...634635636...1068Latest