DocumentCode
607558
Title
Speaker dependent changes in formants based on normalization of vowel triangle
Author
Stanek, M. ; Sigmund, M.
Author_Institution
Dept. of Radio Electron., Brno Univ. of Technol., Brno, Czech Republic
fYear
2013
fDate
16-17 April 2013
Firstpage
329
Lastpage
333
Abstract
In smoothed LPC speech signal spectrum, the local peaks are called formants. Their frequency and level depend on three basic aspects, namely on spoken phoneme, individual speaker as well as emotional state of speaker. This paper is focused on speaker analysis based on comparison of vowel formants with their normalized values. The novel method described here uses for formant normalization statistical values of formants presented in the form of formant planes. Formant normalization is made by average formant position given for Czech vowels. For each speaker the center of gravity of two vowel triangles are calculated; for the small vowel triangle created by /a/, /e/, /o/ vowels and for the big vowel triangle created by /a/, /i/, /u/ vowels. The differences between the individual and normalized center of gravity for both vowel triangles are represented as resulting vectors. Parameters of these vectors were analyzed in more details for statistical evaluation for a set of 12 male speakers. Best results were achieved for the /a/, /e/, /o/ triangle normalized by average formant position in the F2-F3 plane, where the vector angle and the vector length were totally different within analyzed speakers.
Keywords
emotion recognition; natural language processing; speaker recognition; statistical analysis; Czech vowel; formant normalization statistical value; formant plane; individual speaker; local peaks; normalized value; smoothed LPC speech signal spectrum; speaker analysis; speaker dependent change; speaker emotional state; spoken phoneme; statistical evaluation; vector parameter; vowel formants; vowel triangle normalization; Analysis of variance; Databases; Gravity; Speaker recognition; Speech; Speech processing; Vectors; formants; normalization; speaker recognition; speech processing; vowel triangle;
fLanguage
English
Publisher
ieee
Conference_Titel
Radioelektronika (RADIOELEKTRONIKA), 2013 23rd International Conference
Conference_Location
Pardubice
Print_ISBN
978-1-4673-5516-2
Type
conf
DOI
10.1109/RadioElek.2013.6530941
Filename
6530941
Link To Document