SPEAKER RECOGNITION USING MOBILENETV3 FOR VOICE-BASED ROBOT NAVIGATION

• Main Author: Mawadda Warohma, Ayu
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/83741
• Institution: Institut Teknologi Bandung

Giving robots the ability to recognize the different persons they are speaking to is a first step toward enhancing their perceptual and thinking abilities. In this context, ourresearch is implemented on a delivery robot that requires constraints on control and interaction. Implementing a speaker recognition system for a delivery robot is designed to ensure that the robot only executes commands from authorized speakers while avoiding receiving commands from unauthorized speakers. This motivates our research to address text-independent speaker recognition in the context of human-robot interaction. To develop the speaker recognition system, we used the d-vector embedding speaker representation with MobileNet V3 architecture and compared the performance of our proposed method with Fast ResNet-34 architecture. Tests were also conducted on MFCC and Mel-scaled spectogram feature extraction representations to determine which feature representation is suitable for our architecture. The proposed system has been evaluated on Indonesian datasets with various acoustic environments. Fast ResNet-34 achieves an AER of 5.756% with an accuracy of 94.78%, whereas MobileNet V3 achieves an AER of 7.014% with an accuracy of 93.88%. Despite Fast ResNet- 34 showing better performance, the MobileNet V3 approach improves computational efficiency by 98.27%, reduces model size by 87.47%, and speeds up inference time by approximately 7 ms compared to Fast ResNet-34.