Abstract:

Our prime concern is with the recognition of vowels in sounds produced by the human vocal organs, and more specifically, the sounds, which are used in human speech. In this paper we tried to analyze and trace vowels in different phrases of the speech. An algorithm was previously designed which would automatically spot the vowels in the recorded phrase by using rectifier cutoff line and latter on reverse and forward tracking modules were introduced. This approach can further be improved if we could introduce rectification line for both positive and negative phase of the oscillogram (waveform). It has been observed by analyzing different oscillograms, that the stable portion of the vowels crosses the rectification level fixed at 50% of the maximum value (3FFFh for positive and C000h for negative portion) in 16-bit sample wave file. By the introduction of rectification level both in negative and positive portion of waveform about 80% of the vowel body can be identified. The algorithm of reverse and forward tracking based on turning points can identify the rest 20% portion. Thus by having information about start and end addresses of vowels, their formants can be determined with available software like praat. Vowels can be distinguished by the frequency values of the two or three formants, which are called F1 and F2.