Structural Usable Speech Measure Using LPC Residual
Ananth N. Iyer, Melinda Gleiter, Brett Y. Smolenski, Robert E. Yantorno · International Symposium on Intelligent Signal Processing and Communication Systems · 2003
In an operational environment speech is de- graded by many kinds of interferences. The operation of many speech processing techniques are plagued by such interferences. Usable speech extraction is a novel concept of processing degraded speech data. The idea of usable speech is to identify and extract portions of degraded speech that are considered useful for various speech processing systems. The performance reduction of speaker identiflcation systems under degraded condi- tions and use of usable speech concept to improve the performance has been demonstrated in previous work. A new usable speech measure, based on the structure of Linear Predictive Coding (LPC) residual is developed to identify usable speech frames. It is shown that the method has 72% success in identifying the usable frames with 28% false rate.