Meeting Intelligence System

Project Overview

Engineered a comprehensive meeting analysis system deployed on Jetson Nano hardware that processes real-time video and audio streams to extract actionable insights while maintaining strict privacy compliance. The system performs facial detection, emotion analysis, and speech-to-text processing, storing only anonymized metadata for dashboard visualization while ensuring complete deletion of sensitive data post-analysis.

Technical Implementation

• Embedded Vision Processing: Developed an optimized computer vision pipeline on Jetson Nano that performs real-time facial detection and cropping. The system implements efficient memory management for processing multiple faces simultaneously while maintaining processing speed. A custom face tracking algorithm generates unique identifiers for each detected individual, enabling emotion tracking across video frames without storing identifying information.
• Emotion Analysis System: Implemented a two-stage emotion detection pipeline where cropped facial images are processed through a deep learning model to extract emotion probabilities across seven basic emotions. The system maintains temporal consistency in emotion detection while operating within the hardware constraints of the Jetson Nano, achieving real-time performance through optimized model quantization and parallel processing.
• Audio Processing Pipeline: Developed a parallel audio processing system that performs real-time speech-to-text conversion, synchronizing timestamps with emotion data for comprehensive meeting analysis. The system includes noise reduction and speaker separation capabilities to improve transcription accuracy.

Project Achievements

• Internal Analytics: Deployed system successfully analyzes audience emotional responses during team presentations and internal events, providing real-time engagement insights.
• Performance Metrics: Generated analytics helped optimize presentation content and delivery methods, leading to improved audience engagement and comprehension rates.

Technical Stack

• Jetson Nano, CUDA, TensorRT, Docker
• Computer Vision Models, Emotion Detection Networks
• Privacy-Preserving Data Handling
• Azure Blob Storage, Real-time Data Streaming