Enhancing Enterprise-Grade Video Communication Meetings Using an AI-Based Recurrent Neural Network Scheme

In this regard, this study presents an AI conceptual framework to optimize enterprise-level video communication meetings, addressing the challenges of remote and hybrid work. The system focuses on real-time speech recognition, reducing background noise, accurate speaker identification, and automatic meeting summarization. The methodology begins with the use of the Spectral Gating preprocessing to achieve a clear sound by effectively removing noise. Speech characteristics are extracted by the Mel-Frequency Cepstral Coefficients (MFCC) feature extraction method, and feature selection is carried out by the Principal Component Analysis (PCA). The combination of the MFCC and PCA was adopted to ensure a trade-off between the extraction of the perceptual speech features and computation power. MFCC preserves the human auditory properties and PCA eliminates the redundant coefficients, eliminates noise, and leaves the high-variance components. This dimensionality reduction enhanced the stability of RNN-Attention, minimized overfitting, convergence speed, and minimized inference latency which is important in real-time use in enterprises. To classify, an RNN with Attention is used to learn about sequential dependencies and enhance recognition accuracy. The methodology relies on the Speech Recognition and Speaker Diarization dataset provided by Kaggle, which will enable strong training and monitoring. Experimental results prove superior performance with an accuracy of 96.8%, precision of 95.5%, recall of 96.2% and an F1-score of 95.8%. The scalable and low-latency solution proposed fits perfectly into communications platforms, increasing intelligence and workforce collaboration that is more effective and efficient.