Machine learning for Audio Scene Analysis

A user centric circum-aural headset with an internal and an external microphone that enhances user experience in very harsh environmental noise. The system continually measures the dB SPL noise level at the user’s ear (dosimetry), monitors the quality of the seal at regular time intervals, and, enhances speech in noise by selecting the optimum pre-set (default or user) for best user experience using machine learning for Audio Scene Analysis (ASA). The headset uses the classification of the user’s environment to control filter shapes, gain and enhancements for that specific scene.

The dosimetry data is stored locally on each headset and on a server on the IoT cloud. The user is notified when the daily noise exposure dosage is reached. The supervisor is also able to monitor workers for the noise exposure and issue a change of activities before any permanent hearing damage occurs.

A good seal around the ear is required for hearing protectors to attenuate sound. Therefore, an initial test of the seal at power on, and periodic checks would monitor and if necessary, alert the user of a bad seal; such that the user would be able to rectify the poor seal.