Authors: Albin Nilsson and Andreas Anderberg

Title: Concurrent Biometric Identification and Control Using Handclaps

Authors: Albin Nilsson and Andreas Anderberg

Supervisors: Anders Heyden and William Tidelund (Ericsson AB)

Examiner: Kalle Åström

Abstract:

This thesis explores the feasibility of using handclaps as a means of biometric identification and control in the context of home appliances. Utilizing the acoustic properties of handclaps, the study proposes a machine learning-based solution capable of identifying the individual performing the handclap as well as distinguishing between single, double, and triple claps. The solution utilizes well established audio descriptors in combination with classification models including random forest, support vector machines, K-number neighbours, and linear discriminant analysis. Results demonstrate that the linear discriminant analysis classifier in combination with the Mel-frequency cepstral coefficients, spectral density, cepstral density and spectral contrast features achieve the highest performance. Additionally, results show that handclap-based biometric identification is viable under controlled conditions and that a real-time implementation of the proposed predictor is feasible. However, further research is needed regarding the robustness in more diverse environments.