Effect of Phonetic Modeling on Manipuri Digit Recognition Systems Using CDHMMs

This paper evaluates the phonetic modeling for continuous Manipuri speech recognition based on primitive speech units: monophones, diphones and triphones. The study is based on experiments conducted for recognition of Manipuri numerals using Manipuri phonetic structure. The results found after a series of experiments show that diphone and triphone units are more effective and less sensitive to the amount of training data than monophone units for speaker dependent continuous speech recognition for Manipuri.