This article aims to develop an acoustic vowel space in Persian speech. There are several aspects in this survey which make it different from what has been done before. The first is related to the issue of speech material. The need for more natural choice in voice qualities in recent years exhort us not relying on citation form or artificial sound produced in laboratories. Furthermore, the formant frequencies were not extracted from specified vowels, in specified context. In contrast, we are interested in the shape of vowel space determined by extremely large collections of vowel tokens, with whatever distribution of categories and context they may have in the read text. Thirdly, the vowels selected in the database for calculation of the area of vowel space are being stratified for locating in stressed or unstressed syllables, or being uttered by male or female speakers. So, we are simultaneously dealing with four groups. But the most important aspect is related to the methodology used for better plotting of vowels. Either F1*F2 or F1 *F2-F1 is leaded to better vowel classification is a matter being evaluated by two parameters: (a) linear discriminant analysis and (b) scatter reduction.
Vowel Space, Outlier, Linear Discriminant Analysis, Scatter Reduction.