抽象的

Implementation on FPGA and Evaluation Of a Prosody Modification of Speech for Impaired Persons using DWT-OLA

L. Bendaouia, S.M. Karabernou, L. Kessal, H. Salhi

In this paper, we describe the design of a platform on Field Programmable Gate Array (FPGA) for a Prosody Modification of Speech (PMS). As it is not possible to reproduce the hearing capacities to the deficient ear, the speech signal can be shaped on the temporal or spectral domain to preserve the relevant information which could be inaccessible for the auditing system of the hearing-impaired person. So, in designing such platform, we aim to analyze the functioning of the cochlea and to perform efficient processing for improving intelligibility by shifting the frequency components of certain defective bands to different healthy ones. The design combines both Discrete Wavelet Transform (DWT) and Over-Lap and Add (OLA) technique and permits to analyze the input data at different segmental length in order to detect and manipulate the pitches. The DWT-OLA gives efficient real time results as compared to classical designs. We conducted experiments on speech data base get from Arctic corpus and used Mean Opinion Score (MOS) tests to subjectively evaluate the speech intelligibility. We obtained a gain improvement of speech intelligibility reaching 70%. Moreover, the proposed FPGA platform involves significantly fewer resources, reduced memory-size and less dynamic power consumption as compared to some previous wavelet-based implementations.