Architecture for Automatic Speaker Recognition in Voice User Interfaces

EasyChair Preprint no. 14094

Abstract

With the rapid advancement of Voice User Interfaces (VUIs), the need for sophisticated automatic speaker recognition systems has become increasingly vital. This paper presents an architecture for enhancing automatic speaker recognition within VUIs, focusing on improving accuracy, scalability, and user experience. The proposed architecture integrates several key components: a feature extraction module utilizing advanced signal processing techniques, a robust speaker modeling framework leveraging deep learning algorithms, and a dynamic adaptation system for personalized speaker identification. The system employs a hybrid approach combining traditional Gaussian Mixture Models (GMMs) with state-of-the-art Deep Neural Networks (DNNs) to capture both acoustic and speaker-specific characteristics. Additionally, it incorporates a real-time adaptation mechanism that refines speaker models based on ongoing interactions. Evaluation on diverse datasets demonstrates that the proposed architecture achieves superior performance in speaker recognition accuracy and adaptability compared to existing methods. This work highlights the potential of integrating advanced machine learning techniques in VUIs to deliver more secure and personalized user experiences.

Keyphrases: automatic speaker recognition, deep learning, feature extraction, speaker identification, speaker verification, Voice User Interfaces

@Booklet{EasyChair:14094,
  author = {Kayode Sheriffdeen},
  title = {Architecture for Automatic Speaker Recognition in Voice User Interfaces},
  howpublished = {EasyChair Preprint no. 14094},

  year = {EasyChair, 2024}}