Cloud-Based NLP Framework for Automated Speech-to-Content Structuring in Live TV Broadcasts

Main Article Content

Saraschandra Arveti
Anish Hadkar
Mani Teja Nutalapati

Abstract

There is a variety of difficulties associated with the analysis of live television content, especially because of uninterrupted speech, many speakers at the same time, and high levels of background noise. However, efficient structuring of the content is important in order to make such tasks as subtitle generation, highlight creation, and topic summarization more feasible. Being motivated by the necessity of developing a technology which will be able to perform the described functions efficiently, it is suggested to apply a cloud-based system using ASR in combination with NLP techniques. The presented research paper discusses such a system and its main components, namely ASR and NLP technologies for converting audio streams into texts and segmenting topics, as well as algorithms of content structuring and cloud-based storage and streaming. In particular, it is supposed to use self-supervised learning and contextual adaptation to analyze the audio stream efficiently despite noise and multiple speakers. For evaluating the system in question, it is planned to perform real-life tests using samples from live TV broadcast and measure WER, topic segmentation accuracy, and processing latency of the proposed technology.

Article Details

Section
Articles