Uberduck

yo gosmoke can you help me in dm? (nothing tech related i got that)

So in theory, could isolating singular vocals be possible (e.g., separate each singer vocal in a duet)?

If it’s already a thing, has there been any papers about it?

theres a paper on source separation for choirs but i have no clue for duets

Deep Learning Based Source Separation Applied To Choir Ensembles <https://arxiv.org/abs/2008.07645>

Also wanna ask, are there any online courses that can get me up to speed with the required mathematics to fully understand Machine Learning development?

I think I need some touch up on linear algebra myself

google has a course on ml but it specifically targets tensorflow (<https://developers.google.com/machine-learning/crash-course>)

I’m an iOS developer myself, and I’ve played around with CreateML and CoreML a bit

<@954626733598965781>  did a short test of mario and he actually spoke well at 20.

from what I can assume, I think the 2 main issues here were that:

1. some bits of the transcript are a bit off compared to how the lines are delivered (probably doesn't really matter but still).
and 2. I think you may have over-trained him a bit.

https://drive.google.com/file/d/1-B7jF1fkK2G4CcaqTvVua3BUrr3I2zXr/view?usp=sharing

Why is there a comma between Congratulations and you made it? I don't think commas in transcripts for audio files without pauses should be there.

I was trying to make it a bit more accurate to how the line was delivered. Though looking back now it's kinda redundant.

I retrained the model with 20 epoch this time.

So this is Lemon Demon's Cabinet Man (Verse 3 - Part 2) sung by the Jaiden Animation model

And for some reason, the model decided to add some sort of... accent? right after "done"

Bit of context, in the original song, there's supposed to be a rest between the lines "here are done." and "Cause there's a tiny"

But the model just decided to ignore that rest and continue the rest of the bar.

Here's the input text:
```
I only want to have fun.
 But now they're telling me my days here are done.
 Cause there's a tiny little box that they make in Jah pan.
 And pretty soon it's gonna fit in your hand.
```

Main theory is that talknet could be interpreting that breath as that accent?

(ngl though I kind of like the bot's interpretation of the verse better than the original)

Read **#answers**, <@155531453793304576>.

hmm, interesting tts trainer, what program are you using ?