Voice recognition versus speech recogniftion
LoopyByteloose
Posts: 12,537
I am just wondering if anyone has done any work with the Propeller as a means to identify people. In a home automation context, the computer might have to listen to several people with different levels of 'rights' to do things.
I do see examples of Propellers learning commands and responding appropriately in a variety of languages, but can it be/has it been used for a secure log in?
I do see examples of Propellers learning commands and responding appropriately in a variety of languages, but can it be/has it been used for a secure log in?
Comments
If it only recognizes my voice but does not understand anything I say then a simple playback of a recording of my voice is all that is required to log in.
LOL. Heater, you're funny. As if solving the big new problem would break the solution for the old simple(r) problem.
But I guess it could happen, just as send robots to Saturn broke our ability to send men to the moon.
When a person enters the room, the Propeller Demo would request an identify. Then that could be associated with other requests about the lighting, temperature, and so forth into a data base that learns individual preferences.
It does mention that recorded voice can cheat just about any system, but for home automation and learning an individuals habits, this is not really an issue. In other words, there are contexts where less than perfectly secure voice recognition provides a valuable and useful tool.
Thanks.
For once I was not intending to be funny:) But perhaps what I was thinking did not come out in what I said.
I'm assuming there are two different problems here:
1) Speech recognition - As in listening to spoken words and understanding what they mean. Could be simple recognition of a few words or numbers interpreted as simple commands all the way up to a full human like understanding of what is said. Where ever the AI takes us.
2) Voice recognition - As in hearing some speach and recognizing who the speaker might me.
Note that 2) does not imply any understanding of actual words or phrases. Just determining who the voice belongs to. Rather like recognizing your favorite singers voice even if you can't make out the words in the song.
I was also assuming that 1) is mostly beyond the powers of the Prop except perhaps in simple cases when trained for a particular speaker. Actually I have no idea about any work anyone has done on that.
That leaves 2) distinguishing speakers by their voices with out understanding the content of the speech. This may be a simpler task with frequency analysis, formant analysis or whatever techniques are available.
In that situation use for secure ID is not such a good idea, as I said, as a simple playback of a recorded voice will fool the system
So, I was not suggesting that "...solving the big new problem would break the solution for the old simple(r) problem." but rather on a small system like the Prop the solution to the old problem (1) does n not exist and the solution to the new problem (2) is not sufficient for security in anyway.
I remember when I worked in in Linux a few years back, 10 or more, an app that did exactly what Loopy mentioned, it was voice Identification as opposed to recognition. Do you remember the name of it? It was around version 5.2 of S.u.S.E., and came with the install disks, wish I still had my disks.
Anyhow, you match/analyze a voice PATTERN with a recorded voice, with the same word/sentence to be matched, as opposed to analyzing speech phenoms. If the voice print matches, then you'd assign privileges accordingly. Or something like that.
EDIT: btw, I think the STOMP project might have some audio code you might look into.