There is a large quantity of information communicated through speech audio streams (phone calls, broadcasts, commentary etc.) across many industries. In many instances this information is of potentially high value but is discarded for lack of efficient and accurate information retrieval strategies. Speech recognition technologies partially solve this problem but the output texts are still high volume, noisy and difficult to interpret and summarize.

TopicListener combines several approaches and mechanisms including; - TopicListener integrates the power of automatic speech recognition and text based topic modeling technologies - Focuses on the most salient topics, events and issues TopicListener use a combination of; - Topic tracking can work in an incremental way so that as new data becomes available, the topics are updated thus providing an up-to-date and ‘live’ overview of the voice data streams. - A novel interface is designed to visualize topic evolution over time.

- The TopicListener project is currently designed around call centre data and online video transcript data, but can be extended over other large volume speech audio domains. - One operator can manage a specific problem but it is more valuable to find common issues or queries which might alert a company to put preventive actions in place before a problem escalates. - This knowledge would generally be unavailable in time based on other techniques. However, the TopicListener project makes this possible. - In the online video domain, TopicListener extracts the most salient topics from video transcriptions and offers support for advertising and marketing. - For general applications such as broadcasts and online video commentary, TopicListener offers a solution to observe core topics over huge volumes of audio content.