Formant Based Direction of Voice for Smart Microphone Array
Kaluri V. Rangarao, Atul Negi · 2024
Locating a person by analyzing the direction of their voice and studying the formant frequencies is a fascinating challenge. A person’s formant changes over time, creating a wide-spectrum signal, whereas most direction-finding methods typically require a narrow-band signal. Moreover, the ratio of interelement distance to wavelength $\left(\frac{d}{\lambda}\right)$ varies over time. We track the formant using Discrete Fourier Transform (DFT) and Wavelets. We employ a new streamlined easy MUltiple SIgnal Characterization (MUSIC) method to determine the direction. Our research focuses on capturing the formant range and determining the direction using this new approach. This formant stream and Direction of Arrival (DOA) provide crucial Machine Learning data. We have chosen the $\frac{d}{\lambda}$ parameter to be between 0.1 and 0.3, with the interelement spacing at a practical distance for this audio formant frequency. Before conducting accurate data testing, we simulated the array using authentic voices, which presented a significant challenge.