Real-time Parental Voice Recognition System For Persons Having Impaired Hearing

Abstract

Persons having impaired hearing do not live a comfortable life because they can’t hear sounds when they are asleep or alone at home. In this study, a parental voice recognition system was proposed for those people. Persons having impaired hearing are informed by vibration about which one of their parents is speaking. By this means, the person having impaired hearing real timely perceives who is calling or who is speaking to him. The wearable device that we developed can real timely perceive parental voice very easily, and transmits it to person having impaired hearing, while he/she is asleep or at home. A wearable device has been developed for persons having impaired hearing to use easily at home environment. Our device is placed on user’s back, and just a ring-sized vibration motor is attached to the finger of person. Our device consists of Raspberry Pi, usb sound card, microphone, power supply and vibration motor. First of all, the sound is received by a microphone, and sampling is made. According to the Nyquist Theorem, 44100 samples are made per second. Normalization during preprocessing phase, Mel Frequency Cepstral Coefficients (MFCC) during feature extraction stage, k nearest neighbor (knn) during the classification phase were used. Statistical or Z-score normalization was used in the pre-processing phase. By means of normalization of the data, it is ensured that each parameter in the training input set contributes equally to the prediction of the model. MFCC is one of the feature extraction methods that are frequently used in voice recognition applications. MFCC represents the short-time power spectrum of the audio signal, and models the manner of perception of human ear. Knn is an educational learning algorithm, and its aim is to classify the existing learning data when a new sampling arrives. The sound data received via microphone were estimated through preprocessing, feature extraction and classification stages, and the person having impaired hearing was informed through real time vibrations about to whom this voice belongs. This study was tested on 2 deaf, 3 normal hearing persons. The ears of normal hearing persons were covered with a earphone that gives out loud noise. Persons having impaired hearing estimated their mothers’ voice by 76%, and fathers’ voice by 81% accuracy in real-time tests. The success rate decreases due to the noise of environment especially while watching tv. In the tests performed while these persons are asleep, a person having impaired hearing perceives his/her mother’s voice by 78%, and father’s voice by 83% accuracy. In this study it was aimed for persons having impaired hearing to perceive their parents’ voice and accordingly have a more prosperous standard of living.