By Stefan Deichmann, Oliver Jojic, Akash Nagle, Scot Zola, Tom Des Jardins, Robert Rubinoff, Amit Bagga, Comcast Labs
V-REX is a new platform Comcast is building to provide speech-based applications for television control and other areas. V-REX applies automated speech recognition, natural language processing, and action resolution modules to interpret the user’s request and identify the appropriate response. We describe here how we use V-REX to support an iPhone/Android app that allows users to control their cable set-top boxes by speaking into their phones. The primary focus of the work involves building grammar rules and dictionary entries for the range of requests the app can handle. We use the grammar and dictionary both to guide ASR and to allow NLP to extract the actions and entities in the request. We then convert these results into appropriate database queries that extract the information the user needs