Automatic Continuous Speech Segmentation Using Level Crossing Rate

In this paper, a new algorithm to automatically segment a continuous speech signal into phonemes is presented. The proposed method is based on the Average Level Crossing Rate (ALCR) of the speech signal, which is motivated by auditory models.1 It portrays that time information is more suitable to handle non-stationary signals. Based on this assumption, we proposed a non-uniform quantization function which dynamically assigns the levels depending on the importance of the given amplitude range. The parameters, to calculate the importance of given amplitude segment approximately, are estimated by incomplete beta function and logarithmic rules. Experimentation results conducted on TIMIT database show good approximation of segmentation obtained from the TIMIT labeling. The proposed method has a success rate of about 79% for 20ms tolerance using logarithmic rule method