Pocketsphinx works great for telephony.. just don't load 10000 word dictionary or grammar :P the pizza demo uses it.. and it works great from every phone I have tested it with... Rome wasn't built in a day and we need more people that have the skills to really build a general purpose acoustical model that works in more situations. /b On Jan 12, 2009, at 11:46 AM, mszlazak@aol.com wrote:
Yup, or just get pocketsphinx "tuned" up for telephony and then no one will have to bother with ASR vendors.
I believe that some speech data from a good size sample for training is needed to make it more "speaker independent" and better suited for use with phone calls. I have a list of things from the Sphinx forums that would be good to have for a telephony ready PocketSphinx. There is a "wsj" database but I don't know if that's would help??
Best. Mark.