Is malcolm really trying to say that intuitive tone control means the same as automatic tone control?
this state off the art machine, that was not available in 2001, can make me sound enough like Mark to fool his mum?
Malcolm, maybe it can make you sound like Willie Nelson
Please god, no more
Nope. It's worse. He's trying to claim that "intuitive tone control" means "automatic pitch shifting to fix vocals by singers who couldn't hit a tone with a shotgun."
Somerled is convinced that fake phone calls were made using voice morphing technology. As proof of this, he has offered the following:
1. The "original karaoke machines" had this pitch shift that automatically shifted the singer's voice to the proper key. Busted, as the original machines from 1971 had tapes and no pitch changin on either playback or vocals. Later machines do have pitch change for the playback, but that's to match the music to singers who know their range.
2. Relatively recent toy that does pitch shifting of voices. All processing done in the (over sized) microphone housing. Busted, 'cause it sounds like $%#@#$%, which pitch shifting tends to do.
3. A pitch shifted video on YouTube of an Avril Lavigne song. Sounds like $#%^$%, 'cause pitch shifting tends to do that. Busted, both technically and 'cause the song sucks.
4. This latest piece of idiocy involving a microphone pre-amp with compressor and selection of common DSP effects, like echo and chorus. Busted, 'cause it don't even do pitch shifting.
Pitch shifting has inherent limits. Those limitations come from the limits of the FFT used to do the shifting.
FFTs are done in blocks. The time length of those blocks depends on the sampling rate and the number of samples you use in the FFT. As an example, CD audio is sampled at 44100 samples per second per channel. The FFT might be done on 1024 samples, giving a time length of 23 milliseconds.
An FFT done in those conditions gives you a pitch shift accuracy of 43 Hertz. That is, the smallest pitch change you can make (without resorting to god awful tricks that fark things up even more) is 43 Hertz. That's not real good when you consider that the difference between an A 440Hz and the B above that (466Hz) is 26 Hertz.
You can improve the frequency resolution by using a longer FFT. 2048 points in the FFT gets you down to 22 Hz resolution, and 4096 would get you down to around 11 Hz resolution.
BUT, you can't just do that. The longer FFT also means a longer time length. At 4096 samples, that'd be 93 milliseconds. That's 1 tenth of a second to you pedestrians, and that's longer than most phonemes. So, you've going to end up squishing together at least two phonemes into one.
More likely, since the FFT won't be aligned to the edge of your spoken (or sung) phonemes, you'll end up with pieces of a couple of phonemes mashed together with a transition where it doesn't belong.
Translation: It'll sound like **** and be hard to understand.
The shorter the FFT the less it farks up the intelligibility, but the less precise you can be in your shifting.
This leaves out entirely all of the fun involved in doing an inverse FFT on an FFT that has been manipulated. Suffice to say that there's problems there that make the situation even worse than I've described.
In case anybody is wondering:
I've written FFTs, and discrete JTF analysis code. I've written filtering software for DSPs, and done FFT based filtering using hand written assembler code. No wimpy DSP C code for me - the DSPs I was using didn't have enough ooomph to handle the overhead and still get the job done.