AppTek Grows Its Sphere of Influence
AppTek made a series of key enhancements to MediaSphere, incorporating into its media monitoring solution the company’s hybrid machine translation system, new engines for non-broadcast sources, and increased coverage of dialects for automated speech recognition.
According to Hassan Sawaf, chief scientist at AppTek, the hybrid machine translation system has been optimized to deal with noisy data and the speech recognition engine has been optimized to better handle acoustic noise and different dialects.
“Instead of using a standard speech recognition engine we also optimized toward the noisy input,” Sawaf says. “Our overall speech recognition quality is very good.”
MediaSphere is a turnkey solution that makes use of AppTek’s speaker-adaptive speech recognition engine to offer improved speed and accuracy in media monitoring.
According to AppTek, the updated software provides a unified and scalable solution that offers text processing with speech-to-text transcription, machine translation of transcribed text, information retrieval with query translation, and automated name-entity detection.
All of these enhancements allow MediaSphere to generate transcripts from both broadcast and non-broadcast sources, translate the transcribed text, and deliver rich media content online. The solution allows users to set customized reporting and search parameters. With video and audio logging technologies, transcribed content is time-coded, analyzed, and indexed, allowing users to query the repository for terms of interest.
“We’re able to monitor additional broadcast types,” says Steve Cook, chief technical officer at AppTek. “Now we’re able to do much better detection and translation on man-on-the-street-type conversations…like roundtable discussions where you have people that start talking over each other, you have interrupted speech.”
According to Mudar Yaghi, CEO of AppTek, the company works with both commercial and government customers that monitor a variety of media sources.
“People are interested in talk shows, town meetings, editorials, interviews on the street, interviews with government officials—more than just the newscasts,” says Yaghi, who notes that when people monitor media, they are usually looking for opinions that are often expressed colloquially and in dialects.
And Cook says AppTek’s ability to deliver high-quality translation and recognition regardless of the background noise, dialect, setting, etc. has differentiated the company from its competition. “Now we’re able to capture a lot more information in audio/video” he says. “Now we can get higher-quality on just about everything that you get on the video broadcast.”
MediaSphere runs on Windows or Linux and is available in more than 23 language pairs, including multiple local dialects. MediaSphere also offers many special subject domains, including military, political, broadcast, law enforcement, technology, and medical.