Post by Rick Price on Dec 13, 2005 16:02:35 GMT -5
Question: We can't get the software to listen for digits and words at the same time. We thought that using FnxCoreRunNodes would work, passing a Grammar node and a Digit node. That seems to cause the results returned from the GetFirstResult and GetNextResult methods to be consistently '1', then '4', regardless of the words spoken. Again, I suspect we are doing something wrong, and would appreciate any help you can give us?
Answer: The problem with the word/digit combination is that the different neural nets don't play well together. To recognize words and digits, the easiest way is to use a grammar node with a general purpose neural net instead of a digit node (but using your digit grammar). The more difficult but probably more accurate way is to try a double-pass recognition. You essentially recognize on the first node, get your results, call FnxCoreGetUnrecognizedRawData to get the rest of the wave, set the second node to use the raw data with FnxCoreSetNodeRawAudioInput, and run that node to get results.
Answer: The problem with the word/digit combination is that the different neural nets don't play well together. To recognize words and digits, the easiest way is to use a grammar node with a general purpose neural net instead of a digit node (but using your digit grammar). The more difficult but probably more accurate way is to try a double-pass recognition. You essentially recognize on the first node, get your results, call FnxCoreGetUnrecognizedRawData to get the rest of the wave, set the second node to use the raw data with FnxCoreSetNodeRawAudioInput, and run that node to get results.