sphinx to recognize alphabet accuracy is very low

问题:

i'm using CMU sphinx to recognize alphabet letters, but i'm noticing very low accuracy.( <=20%). for example : when I spelling letters: A-P-P-L-E, it come out A B B L E. the accuracy is too low to be useful.

I hope don't have to implement it like some posts mentioned, using "alpha""beta" etc. for improving the recognition rates.

the dict file and lm file in generate in online lmtools BTW: the accuracy rate is above 80% when i limit the dict and speak to microphone with words . so does anyone solve the problem before ? or any idea is appreciate. thx .

回答1:

Yes, accuracy is going to be low because letter names are confusable. The set to recognize E,D,P,B,C,Z is well known to be one of the hardest recognition tasks. Exactly for that reason others use alpha, bravo and so on.

The better solution would be to design your application so it will not require spelling. You can just input words, it's reliable and accurate.

You can always improve accuracy by training your own model for the vocabulary you have or by adapting existing model to your voice.