Hi this is an audio synth I’ve been working on, I’m not too familiar with audio processing so let me know if you know of any improvements I can make. I think the audio warbling (if thats the correct term) is because I don’t have proper control of the phases of each frequency.
Demo hooked up to TTS
https://www.youtube.com/watch?v=JZw574T0VEQ
I’ve been messing with a similar sound synthesis from spectrograms, except with MIDIs instead
i imagine the 2 methods to fix the warble are adding phase, or increasing the precision of the spectrogram you generate. (I assume you use a spectrogram)
to introduce phase, you might be able to play with the TimePosition property
for precision, you can increase the bin size, or some other methods like peak detection to reduce spectral leakage. since this is for speech, you can optimize for it well.
Another suggestion i have is tweening the volume of each frequency, to prevent popping
If you have any questions, i’d be happy to help 
Is this going to be open-sourced as a Roblox model along with node/python for the server? I’ve tried to figure out how to do this before but Im not much of a audio guy so I couldn’t figure it out. I would appreciate if you explained how you did this and loaded custom audio real-time (Roblox side not server side) .
how did you upload the files in real time to the roblox servers? because in the video you said the files are generated on an external server…
They aren’t uploaded real time, they are played back in real time.
I deconstruct the sound on the server, extract out the important frequencies, and then use additive synthesis on the Roblox side to reconstruct the sound
I’m not sure if I plan on releasing this because I fear it could be abused, but after I work out some of the issues I can send you a copy if you would like.