Our Low-cost, high accuracy Speech2Topics technology can process live or stored streams (audio or video) that contain speech data and produce a time-coded set of topics. These topics are taken from a “topic set” that currently features a wide range of consumer-oriented terms, sports and place names and travel destinations. For a live stream, the topics are returned in real time, with a user-defined frequency (e.g. once every 30 seconds) and short lag time. For stored streams, the topics returned are coded to the elapsed playtime of the stream and can thus be indexed against user playback. Topic accuracy can exceed 90% with low-background noise signals and typical North American accents.
Our topic sets and tailored speech recognition models manage ambiguities well in most cases, with interpretation focused on topic-related speech. As regards distinguishing a tennis court from a food court from a court of law, our topic engine does provide for such disambiguation generally, but for higher accuracy we may need to customize topic sets based on customer requirements. Topics not included in a customer’s topic set are not recognized.
As an example, the customer might have a set of video files that will be web streamed. Using Yactraq’s Speech2Topics, each file is pre-processed to produce a map of the topics covered in the video. Using our web API, this JSON data can be accessed by the embedded player streaming the video and keyed to the file’s playback. Using the Speech2Ads service (again via our API) we can then deliver topic-specific, location-aware ad content directly to the user’s web page or mobile viewing platform.
For customers with high volumes and dedicated processing, we can provide a Linux executable that will process the speech embedded in media files or streams to be run on the customer’s servers. This executable engages with the Speech2Topics and Speech2Ads services again via our RESTful API layer. In the future, we plan to allow our customers to choose from a set of predefined topic categories and even to customize their own topic set.