User:
Nizam
Date: 1/30/2015 3:15 am
Views: 2128
Rating: 2
Hello everyone..
I have few question to ask.. Whether this is possible to be done or not by using HTK. Below is the case:
The recognizer must be able to differentiate this sound (Isolated word recognition):
Word given: Rabbit (User only can speak this word)
Speaker 1 speak: "Rabbit" ----> Recognizer result: CORRECT
Speaker 2 speak: "Wabbit" ----> Recognizer result: WRONG. Replace /r/ with /w/
So basically what happen here is the recognizer is able to recognize and give the result of what consonant the speaker "miss" or "replace" in the word given it they speak wrong. If it speak correct, then the output is CORRECT.
my question is:
Is this is phoneme based recognition?? How to achieve this result? what we need to do? e.g:
- Modified the language model / acoustic model / pronunciation model of ASR component.
- Decelop "special" algorithm for word matching in decoding phase?
Is there any system, or research related to this case so far?? maybe you guys can share your findings.
Thank you
--- (Edited on 1/30/2015 3:15 am [GMT-0600] by Visitor) ---
