Design of speaker independent turkish speech control system in noisy vehicle environment


Patlar Akbulut F., Akbulut A.

IASTED International Conference on Signal and Image Processing and Applications, SIPA 2011, Crete, Yunanistan, 22 - 24 Haziran 2011, ss.93-97 identifier

  • Yayın Türü: Bildiri / Tam Metin Bildiri
  • Doi Numarası: 10.2316/p.2011.738-072
  • Basıldığı Şehir: Crete
  • Basıldığı Ülke: Yunanistan
  • Sayfa Sayıları: ss.93-97
  • Anahtar Kelimeler: Hidden Markov Model (HMM), MFCC vectors, Noisy utterance, Remote control, Speaker independent, Turkish speech recognition in car / vehicle environment
  • İstanbul Kültür Üniversitesi Adresli: Evet

Özet

This paper represents the results of speaker independent Turkish speech control experiments in vehicle environments. Almost there are unlimited numbers of words in the Turkish language, it is impossible to use words as the basic unit in the system. In this case, assuming to use the sub-units is more efficient, we choose to work with context-dependent (different vocabulary context in training and test) tri-phones as the smallest units and modelled with Hidden Markov Model (HMM). Also to limit the complexity of the tri-phone models decision tree based state clustering is used. Proposed speech recognition system is able to recognize the speech waveform by translating the speech waveform into a set of feature vectors using Mel Frequency Cepstral Coefficients (MFCC) that are the typical recognition parameters. In experiments on hands-free in-car speech recognition with the microphone far from the talker, this framework is found to be effective in terms of recognition rate and computational cost under various driving speeds. To examine the recognition performance of the system, tri-phone based acoustic model is tested with different decision tree pruning factors. System experiments results had shown that the word correctness of system tests is between 50-86 percent. This ratio is fulfills the safety-critical operations requirements of a vehicle.