Abstract
Conference Title: 2016 IEEE 13th International Conference on Signal Processing (ICSP) Conference Start Date: 2016, Nov. 6 Conference End Date: 2016, Nov. 10 Conference Location: Chengdu, China Determining Voice Onset Time (V OT) in speech is a challenging problem, because it combines temporal and very short duration. VOT is important to detect and classify languages and dialects. This paper is important because of the lack of research in the field of speech processing in Arabic language. Detection of VOT by manually is feasible, but it becomes a time consume when the database is large. Any language has the phonemes classified into two main categories voiced and unvoiced sounds. VOT features appear only in Stop sounds. The benefit of VOT is to classify between voiced and unvoiced stop sounds. This paper focus in unvoiced sounds in Modern Standard Arabic (MSA) stop sounds which are /t/, /t?/, /q/ and /k/. We created the database using CV-CV-CV structure for each word. Our algorithm detects the VOT value automatically for unvoiced stop sounds in MSA language by relaying on reading the power signal, with high accuracy. The outcome error rate is around 2.15 % to 3.52 % for unvoiced stop sounds in MSA Arabic. The measurement of the accuracy depends on comparing the VOT value detected by the algorithm and the VOT value detected manually by reading the spectrogram. The average standard deviation of the manually assigned VOT values is the lowest for unique unvoiced stop sound /t?/ and the highest for sound /t/.