Voice-based Malay command recognition for smart house applications

Abstract

This study is related to the voice-based Malay command recognition for smart house applications. Voice-based command recognition is commonly used for human-computer interaction in security systems, phones, household appliances and hardware designed for handicapped people. Most of the current research, study the voice command recognition for smart home in English. Lack of study for voice command recognition in Malay makes it difficult to apply the voice command services for smart home in Malaysia. In addition, speech recognition is a non-trivial task in natural language processing. The main objective of this project is to identify the command used for smart house appliances using Malay and design the algorithm for this system. After that, the proposed voice recognition algorithm will be deployed on a Raspberry Pi to see the performance of Malay command in accuracy and the suitability of the algorithm to be deployed on low cost embedded devices. In this project, light, fan, and television had been chosen as electrical appliances to build the command. The command will be given in standard Malay. The commands are ‘Buka Kipas’, ‘Tutup Kipas’, ‘Buka Lampu’, ‘Tutup Lampu’, ‘Buka TV’, ‘Tutup TV’, ‘Siaran 1’, ‘Siaran 2’, and ‘Siaran 3’. An algorithm that previously used to recognize songs, the robust quad algorithm, is used in this project for voice command recognition. The proposed voice recognition system has two main processes known as feature extraction and voice recognition. In the feature extraction process, the audio fingerprint will extract data from the command spectral peak. For voice recognition method, audio fingerprint matching will be used to analyze the audio commands. The outcome of this project is when the voice command is given by the user the command will activate or deactivate target application for example when the user said ‘open light’ the light should be open. The second outcome is the background noise that affects the system is reduced by choosing a suitable filter and increase the accuracy of the system. The results of this project have shown that the proposed algorithm is suitable to be implemented on a Raspberry Pi and achieve a high recognition rate with 87%. In the presence of noise with 15 dB, the proposed algorithm is able to maintain the high recognition rate with 82%.