Click for new scientific resources and news about Corona[COVID-19]

Paper Information

Journal:   AMIRKABIR   Fall 2003 , Volume 14 , Number 56-A; Page(s) 1045 To 1066.
 
Paper: 

RECOGNITION OF FARSI NUMBER OVER TELEPHONE: A COMPARISON OF STATISTICAL NEURAL AND ‎HYBRID APPROACHES

 
 
Author(s):  HOMAYOUNPOUR MOHAMMAD MAHDI, KABUDIAN J., BASHIRI H., AHMADPOUR Z.
 
* 
 
Abstract: 

This paper presents results of research on isolated connected and continuous Farsi number recognition over telephone. Effect of selecting basic recognition unit (phoneme or word) is investigated and statistical neural and hybrid approaches are compared. In Farsi digit and number recognition continuous density hidden markov models with word and phoneme models embedded reestimation after baum-welch NN and hybrid methods were used Experimental results show that in noisy and telephony conditions the system employing cepstral vectors derived from Mel frequency filter-bank analysis outperforms the system employing cepstral vectors derived from linear prediction ana11sis. also it was observed that performance of HMM system is more than that of both the NN and hybrid systems. The best performance of digit recognition in isolated and connected Farsi digit recognition (with word models) over Farsdigits1 telephony database was %99.1 and %83.7 respectively and the best performance of word recognition in unconstrained Farsi number recognition (with phoneme models) over Farsnumbersl telephony database was %91.1‏.‏

 
Keyword(s): SPEECH RECOGNITION, NUMBER RECOGNITION , ISOLATED DIGIT RECOGNITION , CONNECTED ‎DIGIT RECOGNITION , HIDDEN MARKOV MODEL MULTI-LAYER PERCEPTRON , HYBRID ‎MLP/HMM SYSTEM
 
References: 
  • ندارد
 
  Persian Abstract Yearly Visit 72
 
Latest on Blog
Enter SID Blog