Voice Recognition

Status
Not open for further replies.

nadushka

Solid State Member
Messages
18
Hi everybody,
I wonder if there is a hardware that can function as a frequency analyzer in order to recognize voices. I need this hardware to be connected to a computer and it should somehow report to the computer if there is a match between the input voice and a voice already stored.

Please help!!
Thanks,
Nadushka
 
You think a frequency analyser would be sufficient to "match" two voice signals? I haven't done any work on speech recognition - but that sounds highly doubtful. I would imagine that u need extract much more specific features from the signals than the basic frequency content itself.

In anycase, that being said, u want the frequency analysis to be done in the h/w itself? As opposed to just having a microphone and running the analysis s/w on the computer?
 
So you think that the frequency analyzer is not going to be enougth in comparing two voices? If so, what other features will be enougth for making the comparaison?
Yes I prefer the analyzer to be a hardware, but a software will be ok if a hardware is not applicable.

Thanks a lot,
Nadushka
 
Yeah. For one thing, I would imagine that the frequency content depends on what the person is actually saying. I guess in a "long a run conversation" things would average out. But, the basic Fourier frequency spectrum has no notion of any time for example. It's just the frequncy content of the whole conversation in one big lump.

It'll be nice if u can explore teh nuances of the time-series itself. I don't know much details into the current state of Speech recognition or the properties of the speech itself - but u should read some research papers. Try the ieee if u have access or do a site:.edu google search and see what u get. But, I'll be highly doubtful if "frequency" is sufficient. It'll be essential, but not sufficient.

Think about it, if u simply do a frequency (magnitude) analysis, u lose all notions of time (atleast the magnitude plot anyways - since the full magnitude/phase transform is invertible) - unless it's something like a Wavelet analysis. And with a signal that varies so much with the actual content, I'd be surprised if it's enough.
 
Status
Not open for further replies.
Back
Top Bottom