From: Jessica P. H. <jph...@ar...> - 2002-11-22 16:34:53
|
On Fri, 22 Nov 2002, Jan Ciger wrote: > I have got the same results even with the large vocabulary, just with a > slightly increased number of errors. (i.e. - I used Rick's acoustic models > without training, turtle LM and full dictionary instead of the small > turtle.dic). So this is not a problem, IMHO. What happens when you use an LM that only recognizes certain words, and you use other words, as in this case? I'll do some tests too, but I'm interested to learn that the results aren't too bad. I would have expected them to be much worse. > Secondly, if you give LM the weight of zero, it is not used, so yes, that's > another possibility for testing it. I had forgotten that this was the way to "turn off" the LM. I just ran sphinx2-simple again with -langwt set to 0 (default is 6.5). The results: really really slow (more than 60 seconds to recognize a phrase) and extremely bad rates: "turn around and go back" 254: CHURN AIR EDGE ETCH DARE > I think, that there is even an option to not use LM at all, but you have to > check sphinx switches to see, how to do that. I'll try to figure this out. Perhaps using the turtle LM with the full dictionary is actually the way to go. j |