Hi people,

I've been looking at paper after paper, trying to get my head around Cepstral processing, Liftering, Quefrencies and what not, but I just can't seem to get past the "first stage"

I understand Spectral processing quite well, and how to theoretically produce a cepstrum, (windowing~ rfft~, cartopol~ log~ rfft~ (roughly) = huzzah, the cepstrum) and return back to a time-domain signal. And yet, I can find nothing about processing audio in meaninful ways using the cepstrum (Formant shifting being one of my goals).

I've tried splitting the real cepstrum into sections with basic array-filtering, shifting and what not, and at this point i'm throwing stuff at the wall to see what sticks.

I'd rather not rely on externals such as TimbreID, while they are handy for analysis, it's hard to use these objects in terms of an actual audio "effect", they're more for indexing (at least, as far as I know)

If anyone has any Ideas, patches or perhaps even more papers for me to explore, I'd greatly appreciate it.

Cheers, Ewan

Just throwing this here, in case it helps, as I haven't done stuff with Cepstral processing. FluCoMa includes a Mel-Frequency Cepstral Coefficients object, but I think that's all. You can have a look though, as it includes a bunch of other stuff for analysis and processing. https://www.flucoma.org/

If I were you I'd peruse (preferably non-copyleft) code on github
https://github.com/r9y9/MelGeneralizedCepstrums.jl
https://github.com/sp-nitech/SPTK
https://github.com/SuperKogito/Voice-based-gender-recognition
https://github.com/suryaturaga3142/dsp-toolkit-v1
https://github.com/talkhouse/noyes

this one says it has a formant detector..
https://github.com/parsa-mz/Pitch-and-Formant-Detector

also this answer seems pretty good: https://dsp.stackexchange.com/questions/37141/preserving-formants-using-cepstrum

and this resource
https://ccrma.stanford.edu/~jos/SpecEnv/Spectral_Envelope_Extraction.html

Thanks for the ideas!

I've done some more work and have made a "real cepstrum", and morphing two signals sounds amazing and crisp, WAY clearer than spectral morphing.

I'm looking at keeping it as vanilla as possible, but I still apprecaite the reccomendations.

I'll keep my eyes on this thread over the weekend, see if anyone else has anything.