A Simple Time Alignment Algorithm for Spoken Arabic Digit Recognition

Abstract. The problem associated with spectral sequence comparisonfor speech comes from the fact that different acoustic renditions, ortokens, of the same speech utterance are seldom realized at the samespeed across the entire utterance. In this paper a simple and effectivetime alignment was introduced for spoken Arabic digit recognitionsystems. We meant with simplicity here not only in its need for lowcomputational power, but also simplicity to understand, to implement,and to explain to others. While high power computers are availabletoday, time alignment algorithms, such as dynamic time warpingalgorithm and hidden Markov models need relatively high CPU time,which should be reserved for other complicated tasks. This algorithmhas a high accuracy rate considering the very limited number offrames taken from input utterances to be used in training or testing.An artificial neural network based speech recognition system wasdesigned and tested with automatic Arabic digit recognition to test thistime alignment algorithm. The system is an isolated whole wordspeech recognizer and it was implemented in a multi-speaker mode(i.e., the same set of speakers was used in both the training and testingphases). During recognition process, digitized speech was cleaned ofnoise, then the signal was pre-emphasized and it was windowed andblocked by Hamming window, the time alignment algorithm was usedto compensate for the differences in the utterance length andmisalignments between phonemes. Frames features were extractedusing MFCC coefficients to reduce the amount of the information inthe input signal. Finally, the neural network classified the unknowndigit. This recognition system achieved 99.48% correct digitrecognition while using only seven frames in the time alignmentalgorithm.