How to use voice recognition on arduino?

[imnimn]

Hello,
I want to ask you if I could recognize special audio signal using arduino and Electret Microphone Board ?
If that is possible please inform me?
Thanks,

 

[doraemon]

Hello!

I think you should forget that.
Voice recognition is quite complex and (I think) impossible to achieve on Arduino.
It depends on the microphone board. If this board has a powerful DSP together with the
microphone, then that may be an option. Otherwise, with a program strictly limited to Arduino,
you will never make it.

Dora.

 

[imnimn]

Thanks Miss Dora , but could you please, tell me if I should connect another hardware device to it , and what it is?
Electret Microphone Board does not contain DSP together , it just could read the voltages sensed from voice .
Thanks again

 

[doraemon]

Hello!

I don't know of any such shield. And I would say there is some kind of contradiction
in doing such a shield. If you develop a voice recognition engine, it means you'r using
some kind of powerful ARM or DSP. Then if you need some arduino-like functionality
(IO pins, LEDs, etc...), then you can program it with the voice recognition board, so
there is no point in having arduino anymore.

By the way, what do you mean by voice recognition?
1. "Listen" to some voice and guess who it is?
2. Recognize the meaning of some words? (in which case it's called "speech" recognition,
not voice).

Beside this, there are many levels in speech recognition:

1. Recognize a few words (e.g "left", "right", "forwards" "backwards", etc..., to control some
kind of robot, with a unique speaker
2. The same recognition, but multi-speaker
3. Recognize the whole dictionary, unique speaker
4. Same recognition, multi-speaker.

You can understand that the constraints are quite different. The first one can be done
(parobably) with a light Cortex M3 (depends on the number of words), but for the 4th,
you need something quite powerful. I'm not sure a plain PC can do it. I have used
dictation softwares, it simply doesn't work.

Dora.

 

[imnimn]

Thanks a lot Miss Dora for your attention.
What I intend to do is to know if a child is crying , by comparing the inserted sound entered to the system by the microphone with saved one , the problem is I don't know How should Ii o that , especially that I have tried to compare several plots resulted of the same sound , and only 3 of 8 plots have some similarities ,(I got the plots by plotting the voltages sensed ) , I know that I should convert them to frequency , but actually I don't know how the algorithm should be "Actually I am not good at signal processing". Thanks a lot

 

[doraemon]

Hello Imnimn!

You can find C code for FFT on the net.
What you are describing should be easier than speech recognition.
And you don't need it to be fast. Even if it takes a few second to decide whether
a baby cry is detected or not, it shouldn't be a problem.

Dora.

 

[imnimn]

Thanks Miss Dora , could please give URL for that
thanks,