Supervised speaker recognition is suitable for identifying famous people. Using this method, by feeding many views of people, a classifier can be trained. Such a supervised method is only useful for recognizing people in the domain.

In daily life or TV News, we do not necessarily meet the famous people. The goal in this project is to recognize people using the information in a video. Video includes appearances of people, their voices, text on the video, and closed captions. We have used TV News in our experiments.

A critical stage is determining the names of people. Names can be obtained from closed captions and video captions. We analyze the names that show up while a person speaks. We also check the most common name that appears while a person speaks. Our research also involves determining whether a person talks or not.

Abstract

Television (TV) networks produce a tremendous amount of information every day. Identifying the speakers throughout a video would help to analyze and understand the video content. Previous research has identified speakers on pre-trained faces for TV shows and movies. News videos are challenging because new faces often appear. By using an unsupervised method, this paper proposes to label speakers using just the available information in the news video without external information. Our proposed framework segments the audio by speaker, parses closed captions for speaker names, identifies talking persons, and performs optical character recognition for speaker names. Our framework utilizes face recognition, face clustering, face landmarking, natural language processing tools, and speaker diarization. Our results indicate 63.6% accuracy for identifying speakers for CNN news.

Citation

Daniel Woo, Ramazan Aygun, "Unsupervised Speaker Identification For TV News", IEEE MultiMedia, , no. 1, pp. 1, PrePrints PrePrints, doi:10.1109/MMUL.2016.3