Multi-Lingual Speaker Verification Using Malay, English, Mandarin and Tamil Languages for Door Security System

Door security system has evolved rapidly in the past twenty years, since door plays an important role on home security. Therefore, the paper proposes a door security system using multi-lingual speech recognition techniques for Malay, Chinese, Tamil and English languages. Speech is used as a tool for speech recognition since it contains features that could be trained by neural network to distinguish every voice pattern for users to perform classification. There are several factors that affects the performance of voice recognition, one of them is the background noise. This door security system that uses multi-lingual speech recognition techniques would classify the sound of the speaker by letting them voice out the name of fruits using any commonly used language in Malaysia, and the door security system will perform classification in order to identify whether the voice pattern comes from one of the users in the database. In order to implement this multi-lingual speech recognition door security system, MATLAB software is used as the development platform of this project. Besides, Mel-Frequency Cepstral Coefficient (MFCC) was used as the feature extraction technique and Convolutional Neural Network (CNN) was used as the training and matching of system while GoldWave application was used as the recording of audio signal. The experiments run on small dataset shows that the accuracy ranges from 71% to 100%.