| Summary: | 碩士 === 朝陽科技大學 === 建築及都市設計研究所 === 100 === So far, English-based rapid speech transmission index (RASTI) is the most common method in Taiwan on evaluating sound fields to test for speech intelligibility in halls. Chen and Chan (2005) found that the minimum value of the effective duration of running autocorrelation function, (τe)min, is correlated with the articulation scores on single-syllable words collected in a hall, but not with the sound field uniformity.
Miyoshi and et al. (2004), Zeng and Wu (2007) use cepstrum analysis to effectively determine the variations within speech signals, where the interference of noise is adequately reduced or filtered out, resulting in clearer speech signals.Hung et al. (2009) found that performance on recognition tasks decreases when there is interference of noise, To solve this problem, one can use cepstrum calculation and substract background noise, improving speech recognition performance.Therefore, this study will delve into the cepstral analysis of speech, and study the correlations between cepstral capacities of single syllables and articulation scores in each survey points of a hall.
This study continues to analyze the samples of 6 sound fields recorded by Chen et al (2005). Results from the comparative study of 76 survey points showed significant correlation between the cepstral energy of one-syllable words and articulation scores collected in all halls (R2≧0.5).
|
|---|
