Hardware validation for intelligibility improvement of NELE on DSP processor

Abstract

In communication systems, the speech signal is affected by background noise which reduces perception of the received speech. In the near-end scenario, noise can't besuppressed as it is generated in real time (background) and it reaches human ear directly. Smart way is to enhance the speech samples with respect to varying background noise. This fact opens a space for developing an algorithm that would be capable of sensing the magnitude of noise and adjusting the amplitude of the speech samples according to near-end (background) noise. Based on the noise level, speech samples can be enhanced dynamically to make it more acoustic. This work focuses on averaging the gain obtained by analyzing the speech and noise energy frame-wise to obtain the smooth gain and then implementing the Near End Listening Enhancement (NELE) algorithm on DSP TMS320C6713 processor for hardware validation. The intelligibility of the unprocessed and processed speech samples is assessed using Speech Intelligibility Index (SII).