For feature extraction i suggest MFCC,
MFCC is currently the most popular feature coefficient used in the speech
recognition, and it can obtain the more accurate results of speech
recognition under a non-noise condition. In the MFCC algorithm, first use
the FFT to calculate the signal frequency spectrum, and then use DCT to
further reduce the speech signal"s redundant information, and reach the aim
of regulating the speech signal into feature coefficients with small
For pattern matching i suggest Dynamic Time Warping (DTW)
This is Automatic speech recognition algorithm it may help :
1) Capturing the signal from beginning to the end.
3) Feature extraction.
4) Use Dynamic time wrapping algorithm to find the best match from a stored
5) Perform the action specified to each command.
Hope this helps.
On Mon, May 9, 2011 at 9:37 PM, wrote:
> Good day to everyone. I am new to speech processing
and recognition. So the
> question may look elementary to many. I am trying to develop a small system
> that would require the system just to recognize 6 words. What kind of
> feature extraction and pattern matching would be optimum for me?
> Anwer, Dhaka, Bangladesh