how do i train pocketsphinx to accurately recognize spoken letters and numbers with near 100% accuracy? What model should i adapt to recognize similar sounding letters like 'b' and 'd'?

2 Answers 2


It is quite hard, unless you have a lot of data.

You'd better use radio alphabet "alpha", "bravo", "charlie".

  • Thanks for the reply. Unfortunately, we cant use radio alphabet in our project. would it be possible if we train the default acoustic model of pocketsphinx with a lot of data? we only need to recognize atleast 2 speakers in our project. thanks! Sep 11, 2016 at 4:54
  • Ok, sure, training tutorial is here: cmusphinx.sourceforge.net/wiki/tutorialam. Data requirements are also listed there, you need 5 hours of recordings at least. Sep 11, 2016 at 10:43

Shrink your language model to just include those keywords you want. The hugely increases accuracy

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.