Frank -- Here's an existence proof:
https://aubio.org -- free for research, but not free for commercial use. but this is not an easy problem, particularly for instruments without fixed pitch (voice) or polyphonic (piano) or both. Zero crossings are inexpensive to compute, and not very valuable (there's a lemma in here about the cost of getting information). FFT can certainly be done in realtime, even on a phone, but may require using GPU code or carefully tuned CPU code, depending on the platform. If you can make it work with ZCs, you will have people approaching you with business propositions. Cell phones are quite capable of doing local speech recognition, but it's much cheaper to use customer paid bandwith to centralize large vocabulary ASR on servers, and to use edge devices for very limited cases, such as wake up phrases (hey siri, alexa). -- Sean On Tue, Oct 23, 2018 at 4:10 PM <[email protected]> wrote: > In reply to Frank Znidarsic's message of Tue, 23 Oct 2018 00:57:26 +0000 > (UTC): > Hi Frank, > > I don't think cell phones recognize human speech at all (unless you have > one > that takes voice commands). AFAIK, they just sample it very fast, and > digitize > the samples for transmission. IOW they use an ADC. > However there is real speech recognition software, which is what I was > talking > about. (E.g. built into Windows.) > > > >Thank you Mixnet. I only want to get the frequency not to recognize > speech. Its amazing that a cell phone can understand speech. I makes me > appreciate how behind I really am. If they can do that I should be able to > produce a simple algorithm that extracts a frequency from an array of > amplitude points. > >I am taking a break now. > > > > > >Frank Znidarsic > >ps My bottle of black iron and helium sucked in after a few months. I am > not sure what happened. I detected no anomalous energy. > > > > > Regards, > > > Robin van Spaandonk > > local asymmetry = temporary success > >
