The following forum message was posted by mokhov at http://sourceforge.net/projects/marf/forums/forum/213052/topic/3782565:
Hi ShanTrip, The type of application you describe should be quite easy to do, you just have to change categories from speakers to words and do some experiments to select which algorithms do it best for this category. It is bound to be different than the best algos for speakers themselves. For example, I\'ve done experiments on changing categories to spoken accent and gender instead of identity in here: http://portal.acm.org/citation.cfm?id=1370256.1370262 The words principle is the same. The SpeakerIdentApp\'s .txt db would have to become like a dictionary of word and the associated \"good\" pronunciations files for training and testing. It may also be wise to experiment to split each pronounced word into three subcategories, one per age group and gender to increase accuracy. In this case, a match would be searched in the related categories and them mapped to the same spelled word when the smallest distance is found or the largest similarity is found between the spoken and trained utterances. Else the application is quite feasible and easy to alter from SpeakerIdentApp to the task. Let me know if you still have any more questions. -s ... and it\'s not long at all ;-) -s ------------------------------------------------------------------------------ The Palm PDK Hot Apps Program offers developers who use the Plug-In Development Kit to bring their C/C++ apps to Palm for a share of $1 Million in cash or HP Products. Visit us here for more details: http://ad.doubleclick.net/clk;226879339;13503038;l? http://clk.atdmt.com/CRS/go/247765532/direct/01/ _______________________________________________ marf-devel mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/marf-devel
