Task

To develop an AI-based sound localization model that tracks the location of noise sources in the mouth and nasal cavities of subjects using an array of contact microphones. This information is then used to perform several diagnostics related to sleep quality and sleep apnea. The AI model we created effectively functions as a beamformer.

Challanges

• Noisy data
• Variability between subject anatomies, resulting in high data variance
• The human head is a complex acoustical environment with multiple transmission paths

How we helped

Our solution comprised a de-noising step, a delay estimation step, and the creation of a PyTorch machine learning model, which we trained using data supplied by the client.

Results

In this early stage of the project, our classification results were encouraging, with 91% precision and 85% recall, validating the approach taken by the client toward the commercialization of their product.

More information can be found here: Bairitone on LinkedIn