what synthesis technique are you gonna use for the speech synth?
if you want recommendations, the paper linked by pink trombone is a pretty good one. I'm not an aficionado or particularly well versed, but it's the most realistic non-ML method i've heard. it's also a physically informed technique.
there's also Software Automatic Mouth, for Commodore 64, at the other end of realism.
links:
dood.al/pinktrombone (there's also a more phonetics inclined version of this but idk where it webresides)
discordier.github.io/sam
tips for rolling your own:
if you wanna additively synthesize the formants instead of using a filter on the glottal pulse train, one useful method (that SAM uses) is to reset the phase of the formant oscillators each period of f0 (voice pitch). this makes the frequency content distributed across multiples of f0 (and verymuch correlates it with the fundamental) which gloms it on perceptually as part of the same sound instead of an unrelated sound. as for what wave to use, i'd suggest A*sin(ωt)e^-at for reasons. ω controls formant center frequency and a controls broadness, A is just volume. if it's too clicky, multiply in a factor of t.
webaudioapi deprecated main-thread audio processing effects and the MDN docs give the impression you need to fetch a file for the worklet. if you don't likey this, you can create the audioworklett using a text blob turned into a uri. you can make the text from a string or a script element with non-javascript type.
anywyas good luck meows at you meows at you
I should probably explain the steps that led me to creating the project!
I've imagined my character Sauce and his siblings to have a sort of wave-like voice—i.e Sauce would use a sine wave, Granola a square wave—and with that and a combination of a noise channel for voiceless sounds it would allow a speech-like sound
the original plan I had was to take an existing audio file and extract the lowest formants, generating the wave from that. this turns out to actually be EXTREMELY hard and Probably Impossible (at least with the limited knowledge I had)
turning to Vocaloid soundbanks, The Fucking English Language Has So Many Fucking Diphones. and also that kind of software Probably isn't made to do that??? just a hunch??? and any guides aren't helpful with the little patience I have
as such I thought about it some more and turned to just hand-crafting a synthesizer myself. for how I'm marketing it I'm using "voice synthesis" in an extremely broad term as the "voice" I'm going for is highly stylized and might be uninterpretable once I get it working but THE IDEA IS THERE
for the implementation: there are a set of "phonemes" (really it can be anything) which have an enter (transition in), sustain (gemination or vowel lengthening), and release (transition out) form, determining how phones combine together. the actual sounds themselves are just basic waves with interpolation for parameters like pitch, volume, and a few more DAW-like things because I could and I already did Hahahaha
this fortunately means that I get to basically throw all of the necessary research needed out of the window because I actually don't give a crap and I love reinventing the wheel and it actually isn't traditional human voice (though that wave formula does look interesting). this unfortunately means your textwall was for naught........
thank you for the tips though !! and for the luck !!
