I'm trying to run recognition on an existing WAV file.
Even though I couldn't find any example showing how to do it with the PCSDK, I managed to write something that seems to work.
The problem is, that as far as I can tell, when a WAV file has background noise or its volume is low, the recognition process isn't consistent.
When running on the same file over and over, the results change from one execution to another.
To eliminate my code from the equation, I did the following "trick".
- Downloaded and installed the VB-Audio virtual cable from http://vb-audio.pagesperso-orange.fr/Cable/index.htm
- Set the "VB-Audio Virtual Cable" as default playback and default recording device.
- I ran the "voice_recognition.exe" that came with the PCSDK.
- I launched Window Media Player and played a WAV file containing a noisy recording of the text "Climate Rear Defrost".
- Every time I played it I received a different recognition result as can be seen in the following JPG:
So two things:
- Can anyone explain this behavior and how to improve it so at least it will behave in a deterministic way when presented with the exact same input?
- Does anyone have an example of how to VR on an existing wav file using UtilMCaptureFile?
If the answer for "2" is - no then if I post the code I've created will anyone be able to advise as to its correctness?