Contactless Respiratory Rate estimation using acoustic beamforming and spectral techniques

Respiratory Rate (RR) is an important physiological parameter for assessing respiratory health and detecting early signs of ill-health. Microphone arrays offer a promising solution for capturing respiratory sounds in enclosed environments for non-contact health monitoring. However, monitoring RR in a closed environment presents significant challenges, especially when multiple subjects are in proximity, leading to overlapping acoustic signals and spatial interference that make signal separation challenging. This study proposes an RR estimation method based on Direction-of-Arrival (DoA) estimation and beamforming techniques to address this challenge. Acoustic signals were recorded using a circular microphone array, and preprocessed with a Butterworth bandpass filter (100–3000 Hz) to suppress noise and preserve relevant physiological information. A time-frequency representation was obtained via the Short-Time Fourier Transform (STFT), followed by computation of the spatial covariance matrix to characterize inter-channel dependencies. DoA estimation was estimated using the Steered Response Power with Phase Transform (SRP-PHAT) method, and Minimum Variance Distortionless Response (MVDR) beamforming with diagonal loading was applied to isolate individual sound sources spatially. The beamformed signals were reconstructed in the time domain using inverse STFT. Post-processing combined spectral subtraction, Wiener filtering, harmonic enhancement, and adaptive gain control. Signal quality was monitored using voice/activity detection (VAD) based on median and median absolute deviation (MAD) thresholds in the log-energy domain. RR was extracted using Hilbert transform-based envelope detection after bandpass filtering in the respiratory frequency range. Results from the proposed method showed a mean absolute error (MAE) of 1.62 bpm and a root mean square error (RMSE) of 1.93 bpm.