Analysis of classroom speech intelligibility using fuzzy linear regression model

Abstract

Speech is defined as a process of producing sounds confirming to an accepted code by the human vocal apparatus. These sounds are perceived by the ear-brain system with the primary purpose of conveying thoughts. The intelligibility of speech refers to the accuracy with which a normal listener can understand a spoken word or phrase. In a classroom, a teacher talks to a group of students who are intended to hear everything the teacher says. The achievements and behavior of the students inside the classroom is mainly influenced by the speech intelligibility factor. For achieving the highest possible speech intelligibility, the acoustical design of classrooms should be based on all the listeners in the classroom. This research work investigates the effect of Signal to Noise Ratio (SNR) on speech intelligibility in University classrooms. The speech intelligibility level of a classroom depends on the room volume, source receiver distance, back ground noise level, RT, SNR. A set of Malay words with CVC format is complied and a speech signal data base is created. In a classroom, the speech signal is presented at a level of 47 to 72 dB with +/- 2dB and the noise at levels of 55, 60 and 63 dB are electrically mixed to yield a signal-to-noise ratio of -16 to +16 dB. The sound pressure levels are then measured at different classroom positions. From the measured sound pressure level, the speech intelligibility at various listeners' positions is determined and a simple neural network model is developed to predict the speech intelligibility at various listeners' positions of a classroom for various speech levels.