DEVELOPMENT OF AUTOMATIC SPEECH RECOGNIZER AND IOT DEVICE CONTROL SYSTEM USING PERSONALIZED VOICE COMMANDS

• Main Author: Basuki, Kevin
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/51251
• Institution: Institut Teknologi Bandung

Voice command has been one of the easiest ways for users to command IoT devices. Although it offers ease of usability and convenience, commands are usually predefined by the service providers which makes them quite rigid for users. To provide a more flexible user experience, it would be better if commands are personalized. With personalized commands, users can use their desired phrases to give command to IoT devices. In this work, an Automatic Speech Recognizer (ASR) and a mobile application are built as proposed solution by handling personalized commands. This research begins with gathering voice data and examples of voice command sentences from several speakers. Then, an ASR is built using n-gram language model with experiments on acoustic models by comparing GMM-HMM, DNN-HMM, and TDNN models to produce a good accuracy for voice command. After that, a mobile application is built to register personalized action keywords, lamp names, and command sentences; give commands to lamp devices; and update the language model based on registered personalized commands to maintain the recognizer’s accuracy. ASR using TDNN acoustic model produces best result with 0.52% average WER. When applied to the mobile application, it produces 99.58% task accuracy for commanding lamp devices.