Human Ear Implants Based on Feature Extraction (DSP strategies)

Index:

4005

Year:

2003

Thesis Title:

Human Ear Implants Based on Feature Extraction (DSP strategies)

Student Name:

Daisy Armin Tadros

Supervisor:

Dr. Iman Abuel-maaly

Department:

Electrical

Abstract

 

Cochlear (inner ear) implants have been very successful in restoring partial hearing to profoundly deaf people. The objective of this project is to partially simulate a DSP strategy for the extraction of certain features (frequencies) in the sound. The algorithm used is the linear predictive analysis algorithm. LPC is a way of encoding the information in a speech signal. LP is utilized in all speech applications that involve the speech spectrum. In our case it is based on finding the spectral envelope of speech from which formants are extracted.

The idea of the implant is to bypass the outer and middle ear and to communicate with the Auditory nerve directly through the electrodes inserted in the inner ear. The electrodes are simulated by certain frequencies (formants) extracted from the speech heard.

The first step was to record a speech sound via a microphone and a voice card into the computer.

The second step was to find the spectrum of the time waveform of the speech signal recorded. This was done using MAT Lab simulation. Frames from the speech signal are multiplied by a hamming window then the discrete Fourier transform of each frame was calculated.

The final step is to extract certain frequencies from the spectrum using linear predictive algorithm. The frequencies are called formants which appear at the resonance of the vocal tract, these frequencies are very important to the ear in understanding speech. These steps were simulated using MATLAB due to its greater flexibility.

The results obtained were good and especially for vowel sounds, the analysis showed the formant frequencies, the pitch frequency and the relative amplitudes of the formants. The results showed the difference between voiced and unvoiced sounds.