Fraunhofer-Gesellschaft

Publica

Hier finden Sie wissenschaftliche Publikationen aus den Fraunhofer-Instituten.

Objective evaluation of speech intelligibility for speech masked by time reversed masker processed from the target speech and the effects of its parameters

 
: Wang, Yue; Leistner, Philip; Li, Ping

:

Acta acustica united with Acustica 98 (2012), No.5, pp.820-826
ISSN: 1610-1928
ISSN: 1436-7947
English
Journal Article
Fraunhofer IBP ()

Abstract
Time-reversed masker processed from the target speech, which is different from other maskers because of its unsteadiness, has the highest masking efficiency. For time and economical reason, other than subjective measurement, appropriate objective measurements of speech intelligibility for speech masked by time reversed masker processed from the target speech should be considered. Four objective evaluation methods are analyzed in this paper to find which of them has the highest correlation with speech intelligibility of speech masked by time-reversed masker processed from the target speech in open-plan offices. Results showed that although STI has high correlation with speech intelligibility for masking speech by steady noise, it cannot represent the subjective speech intelligibility very well for masking speech by time reversed signal. A Short-Time Objective Intelligibility (STOI), Modified normalized covariance method (mNCM) and Perceptually Evaluation of Speech Quality (PESQ) all has high efficiency to predict speech intelligibility of speech masked by time-reversed masker processed from the target speech. We further applied STOI to investigate the effect of two parameters (reversed frame length and Target-to-Masker Ratio) in this masker on speech intelligibility. The experiment results showed that reversed frame length between 120ms and 240ms and Target-to-Masker Ratio lower than -5dB will lead to poor speech intelligibility.

: http://publica.fraunhofer.de/documents/N-223351.html