Beamforming Device and Architectures for Large Language Model Enabled Personal Monitoring

Background:

Traditional diet-tracking apps rely on manual logging, which often results in incomplete or inaccurate data, particularly for users struggling with consistent healthy eating habits. Sensor-based approaches such as cameras or EMG sensors are often intrusive, uncomfortable, or unreliable in real-world settings. Accurate, long-term monitoring of eating behavior is essential for managing and preventing conditions such as obesity, diabetes, and other diet-related health issues. There is a clear need for a non-intrusive, automated system that can track dietary intake reliably across diverse, real-world environments.

Technology Overview:

University at Buffalo technology WisperWatcher uses a wearable microphone array combined with beamforming to capture high-fidelity chewing and swallowing audio, even in noisy environments like busy meals. The recorded audio is processed either on-device or in the cloud by a Large Audio-Language Model (LALM), which classifies sounds to identify food type, meal timing, and user disposition, such as stress levels. Advanced algorithms optimize microphone configuration for source separation, and the system is designed to integrate seamlessly with future smartwatches or other wearable form factors. The processed data is presented through a web application, providing real-time or logged insights to users or healthcare providers. WisperWatcher also allows flexible adaptation to new dietary analysis tasks without retraining.

Advantages:

WisperWatcher automates dietary monitoring, reducing reliance on self-reporting and improving data completeness and accuracy. Beamforming significantly enhances noise resistance, making it effective in real-world conditions.

Applications:

• Personal diet tracking for weight management and healthy eating.
• Clinical monitoring of patients requiring supervised dietary intake, such as those with diabetes or eating disorders.
• Telehealth and nutrition coaching platforms for remote guidance.
• Occupational health monitoring in workplaces requiring dietary compliance.
• Integration with future wearable devices like smartwatches.

Intellectual Property Summary:

Provisional patent application 64/007,905 filed March 17, 2026. 

Stage of Development:

• Prototype wearable system demonstrated; algorithms validated through simulation and real-world audio recordings. (TRL 4–5).
• https://en.wikipedia.org/wiki/Technology_readiness_level

Licensing Status: 

Available for licensing or collaboration.