<!DOCTYPE html PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN">
<html>
<head>
<meta content="text/html;charset=ISO-8859-1" http-equiv="Content-Type">
<title></title>
</head>
<body bgcolor="#ffffff" text="#000000">
Shawn Rutledge wrote:
<blockquote
cite="mid:d9a6586e0803261551m39edc58ah71c7b6628d63a649@mail.gmail.com"
type="cite">
<pre wrap="">On Wed, Mar 26, 2008 at 3:18 PM, saurabh gupta
<a class="moz-txt-link-rfc2396E" href="mailto:saurabhgupta1403@gmail.com"><saurabhgupta1403@gmail.com></a> wrote:
</pre>
<blockquote type="cite">
<pre wrap="">1)Speech Recognition Method:
i have already developed a speech recognition method on 8-bit
microcontroller. i used the digital filters(chebyshev II) and correlation
method to match the spoken word with that already stored in memory. i am
further planning to improve it using Hidden markov model(HMM) through Linear
Predictive Coding (LPC) and fast fourier transform (FFT). A success rate of
90% can be achieved using this method.
</pre>
</blockquote>
<pre wrap=""><!---->
I'm impressed you managed to do that on an 8-bit microcontroller!
Free software for speech recognition has been slow to develop in the
past. Now there are a couple:
<a class="moz-txt-link-freetext" href="http://freshmeat.net/search/?q=speech+recognition§ion=projects&Go.x=0&Go.y=0">http://freshmeat.net/search/?q=speech+recognition&section=projects&Go.x=0&Go.y=0</a>
E.g. there is Sphinx and Julius, and there is also an HMM toolkit
which might be useful (or not).
Do you expect to achieve better performance than those?
</pre>
</blockquote>
I'm assuming you used dynamic time warping with your correlation
method. I can see not too much processing power being needed for
correlation with a limited vocabulary.<br>
<br>
I would be excited to see existing free voice software fully integrated
into Openmoko with a custom language model for command & control.
PocketSphinx (<a class="moz-txt-link-freetext" href="http://www.speech.cs.cmu.edu/pocketsphinx/">http://www.speech.cs.cmu.edu/pocketsphinx/</a>) may be a good
starting point, since it's supposed to be optimized for embedded ARM
applications. <br>
</body>
</html>