Title: An effective frame-based high frequency speech transposition by using neural network

Authors: Prashant G. Patil; Arun K. Mittra; Vijay S. Chourasia

Addresses: Department of Electronics Engineering, Manoharbhai Patel Institute of Engineering and Technology, Gondia (MS), India ' Department of Electronics Engineering, Manoharbhai Patel Institute of Engineering and Technology, Gondia (MS), India ' Department of Electronics Engineering, Manoharbhai Patel Institute of Engineering and Technology, Gondia (MS), India

Abstract: This paper investigate design methodology and performance of neural network based frequency transposition algorithm for hearing aid users. High frequency hearing loss associated with hearing disabled person is promising issue for research. Frequency compression and frequency transposition schemes are key solution to overcome high frequency hearing loss. Neural approach to frequency transposition makes algorithm more sensitive, accurate and specific towards processing. The proposed neural network frequency transposition (NNFT) algorithm is based on framing of speech into feature vector for NN with comprehensive training and processing. The parameter to set in NNFT algorithm was calculated by evaluative study. Using these algorithm Marathi alphabets, words, confusing words are efficiently classified. Classification will improve acceptance and rejection rate for FT processing. Validation and testing result of algorithm shows improvement in sensitivity, accuracy, specificity of NNFT method compared to FT method.

Keywords: neural network; frequency transposition; speech frames; FFT; hearing loss; sensitivity.

DOI: 10.1504/IJISDC.2018.092561

International Journal of Intelligent Systems Design and Computing, 2018 Vol.2 No.1, pp.88 - 98

Received: 18 Nov 2017
Accepted: 28 Feb 2018
Published online: 24 Jun 2018 *