From: Sean W. <li...@su...> - 2003-10-22 01:21:11
|
Mike Warner wrote: >Sean, > >I just installed Sphinx 2 and am planning on using your scripts as >examples, but, when testing Sphinx (with sphinx2-simple), it didn't do >that great at recognizing words like 'kitchen' or even 'Goodbye', whereas >it got all the numbers. Did you have to train your install? > >Thanks, >Mike > > Nope. It doesn't need training, but it does need a valid language model. When you run sphin2-simple it is configured, by default, to use the "turtle" language model. This includes a few words like "left" and "right" as well as a bunch of numbers. So, it worked according to the language model you "provided" to it. There is a training utility, but that, well, is, um, dramatic overkill. That is something used for training for a different language, I think. To generate a language model, you need a tool to convert your text into the proper data. Check out SimpleLM.pl in either the example stuff or my heavily modified one that I use for sphinx2-mh. I put up a newer version on my website, well, it is transferring as I type this. -- Sean Walker http://sean.surfmy.net/ |