S.A.Kryukov wrote:
Please, make it clear. There are two possible understanding of it:
#1) the program will do recognition and will generate the string "John" as the result of this recognition;
#2) or the program will analyze the voice and find out that it was pronounced by one of the persons, known to the system from previous records: it was pronounced be Pete, not Alice...
Which one did you mean?
Dana Sorani replied:
the second one
That is really hard thing, make take a lifetime of research. The first thing is implemented in Windows, you could use System.Speech assembly (
http://msdn.microsoft.com/en-us/library/system.speech.recognition%28v=vs.110%29.aspx[
^]). However, there are projects for voice-based authentication. But: 1) So far, I never heard anything open-source, except experimental projects; 2) Research study I saw says that, compared to other ways of authentication, speech-based authentication is notoriously unreliable. I concluded that at the present level of science, it's really hard to get any results of any practical significance. Sorry, I don't have any links right now, I just explain my opinion; you can conduct your own search.
I could be mistaken though, but here is one argument: how well can you tell one person from another one by voice? Yes, we easily recognize few relatives and friends who call us regularly on the phone, but this is mostly because we've used to it and expect their calls. In an unusual environment, we can easily make mistakes. A good parody artist can easily mimic the voice of some well-known performers and politicians. And so on…
Even though such technology exists, even if it is mostly experimental, I would advise not to hold your breath.