Evaluation of an HMM speech recognizer with various continuous speech databases