Is it during pocketsphinx batch the cmn value is updated only once?
While in continuous the cmn is updated many times because the audio is read sequentially ?
Does changing the cmn value improve the recognition accuracy?
The default value for cminit is 40,3,-1
So if i try to decode 2 audio say
Audio. 1 cmn value. 57 -11(and the reamainin 11 values
Auduo 2 cmn value 49 -9 (and the remaining 11 values)
So I need need to choose a cminit of 53,10( and the remaining 11 values averages)
So that the accuracy will improve?( Please correct me if I'm wrong)
Thanks in advance
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
When i try to decode the audio file the cmn value updates many times?
how to update the cmn value only once?
how to choose the cminit values?
Yes.
Modify the source code.
Select average value from the test database.
Is it during pocketsphinx batch the cmn value is updated only once?
While in continuous the cmn is updated many times because the audio is read sequentially ?
Does changing the cmn value improve the recognition accuracy?
The default value for cminit is 40,3,-1
So if i try to decode 2 audio say
Audio. 1 cmn value. 57 -11(and the reamainin 11 values
Auduo 2 cmn value 49 -9 (and the remaining 11 values)
So I need need to choose a cminit of 53,10( and the remaining 11 values averages)
So that the accuracy will improve?( Please correct me if I'm wrong)
Thanks in advance
No, it is recomputed for every utterance.
Yes, it is updated once in 10 seconds or on silence.
Yes if audio is not very diverse.
The default value for cminit is 40,3,-1
2 utterances are not sufficient to estimate the stats, you need 20 at least. But the idea is correct. Accuracy will improve.