Novel Approach towards Speaker Identification in Meetings

Effective documentation and script recording of official proceedings, meetings and conferences is essential for decision-making. But manually recording who says what can be time-consuming and prone to mistakes. In order to automate the process of identifying speakers during meetings, we propose in this paper a speaker identification system that makes use of deep learning techniques. Our system extracts features from audio recordings using Melfrequency cepstral coefficients (MFCCs) and Long Short-Term Memory (LSTM) networks. We validate and experiment a great deal to show that our method works well for reliably identifying speakers in real-time scenarios. This study helps to improve efficiency and accuracy in recording and analyzing meeting content by streamlining the formal meeting documentation process.