Re: Lower WAV pitch problem
- Posted by Dan Moyer <DANIELMOYER at prodigy.net> Jun 03, 2003
- 438 views
Jeeze!!!! It's even got stuff about neural nets! Thanks, I'll see if I can understand any of it :) Dan Moyer ----- Original Message ----- From: <1evan at sbcglobal.net> To: "EUforum" <EUforum at topica.com> Subject: Re: Lower WAV pitch problem > > > Check out www.dspguide.com. > > Dan Moyer wrote: > > > > >(looking very stupid) when I suggested just tacking copy of wave onto end > >of wave to double duration without pitch shift, I was just considering my > >simple wave, not the general & more realistic case, sigh. <head hits > >keyboard: tyughjvbn> > > > >Does fourier analysis yield a set of sine waves which when re-combined yield > >the original waveform? > > > >How do you do fourier analysis? :)) > > > >How do you discern noise, just high frequency? > > > >I'm interested in changing human voice samples by lowering pitch, but > >thinking to only lower vowel portions of the sample, not sibilance & > >explosives (or whatever they're called, like "PoP", "Boom", "Take", etc), so > >I'd need to be able to discern them. I'm getting the impression it's more > >like juggling chainsaws than knives :) > > > >Dan Moyer > > > >----- Original Message ----- > >From: "David Cuny" <dcuny at LANSET.COM> > >To: "EUforum" <EUforum at topica.com> > >Sent: Tuesday, June 03, 2003 12:18 AM > >Subject: Re: Lower WAV pitch problem > > > > > >>Daryl Van Den Brink wrote: > >> > >> > >>>I don't know what you could be doing to double the duration > >>>and make it's pitch stay the same, but I'd love to find out. > >>> > >>> > >>This turns out to be solvable, but non-trivial. Sound basically decomposes > >>into two types: sound carrying harmonic content, and noise. > >> > >>First, you chop up the sound into sufficiently small chunks. If you make > >> > >> > >the > > > > > >>chunks to small, you don't capture enough harmonic information. If you > >> > >> > >make > > > > > >>them too large, you end up getting "pre echo" because you're including > >>information that doesn't belong in that timeframe. > >> > >>To derive the harmonic content, you do a fourier analysis on each chunk. > >> > >>To derive noise content, once you decide that a chunk contains noise, you > >> > >> > >do > > > > > >>bark banding on it. Noise doesn't have to be pitch shifted in the > >>reconstruction. > >> > >>You also need to look at the volume, so you can build a volume envelope > >> > >> > >when > > > > > >>you rebuild the sound. > >> > >>Now you've got enough information to reconstruct the sound. Take the > >> > >> > >chunks > > > > > >>that have harmonic content, and rebuild their harmonics to the new pitch - > >>just reverse the fourier process. The noise chunks are rebuilt out of the > >>bark bands. Join all the chunks together and recreate the volume envelope > >> > >> > >to > > > > > >>match the original sound. > >> > >>Easy, huh? > >> > >>Sorry, I don't have the references available. That's the point I threw up > >> > >> > >my > > > > > >>hands and decided to try something easier, like herding cats or juggling > <snip> > > > > -- > |\ _,,,---,,_ > /,`.-'`' -. ;-;;,_ > |,4- ) )-,_..;\ ( `'-' > '---''(_/--' `-'\_)`-'\_) > > > > TOPICA - Start your own email discussion group. FREE! > >