DocumentCode
2424236
Title
An improved STI method for evaluating Mandarin speech intelligibility
Author
Yang, Lin ; Zhang, Jianping ; Yan, Yonghong
Author_Institution
Inst. of Acoust., ThinkIT Speech Lab., Chinese Acad. of Sci., Beijing
fYear
2008
fDate
7-9 July 2008
Firstpage
102
Lastpage
106
Abstract
Speech Transmission Index (STI) is a physical metric predicting the speech intelligibility. It shows a good correlation with the subjective intelligibility when used for English, but it has not been reported to be applicable to such tonal language as Mandarin. In this study, the traditional STI method is improved by adopting an accumulative normal distribution assumption between the signal-to-noise ratios and intelligibility, in which some characteristics of Mandarin speech are taken into considered. Results from psychoacoustic perceptual experiments show that the improved method is more suitable for objective evaluation of Mandarin speech intelligibility in noisy and reverberant listening conditions compared with the traditional method. The correlation coefficient between the subjective intelligibility and the objective STI is improved from 0.97 to 0.98; and the standard error decreased from 15.67% to 6.58%. This method is also expected to be applicable to other languages.
Keywords
natural languages; speech intelligibility; speech processing; Mandarin speech characteristics; Mandarin speech intelligibility; accumulative normal distribution; correlation coefficient; noisy listening condition; psychoacoustic perceptual experiment; reverberant listening condition; signal-to-noise ratio; speech transmission index; subjective intelligibility when used tonal language; Acoustic measurements; Acoustic testing; Artificial intelligence; Frequency; IEC standards; Natural languages; Noise measurement; Signal to noise ratio; Speech analysis; Speech enhancement;
fLanguage
English
Publisher
ieee
Conference_Titel
Audio, Language and Image Processing, 2008. ICALIP 2008. International Conference on
Conference_Location
Shanghai
Print_ISBN
978-1-4244-1723-0
Electronic_ISBN
978-1-4244-1724-7
Type
conf
DOI
10.1109/ICALIP.2008.4590080
Filename
4590080
Link To Document