Speech Technology Magazine

Artificial Solutions Enhances ASR

Revamped platform used to build NLI solutions.
Posted Nov 14, 2012
Page1 of 1
Bookmark and Share

Artificial Solutions, a natural language interaction provider, announced new developments in its Teneo platform that help address the limitations of automatic speech recognition (ASR).

The platform combines ASR with natural language interaction (NLI), delivering a more humanlike experience, the company said. ASR is used to turn the speech into a textual input that the NLI engine processes in three steps. First it analyzes the query using powerful linguistic understanding libraries that understand and derive the meaning. It then interprets this using advanced linguistic and business rules that simulate intelligent thinking, allowing it to reason like a human and determine the most appropriate action. Finally it performs the necessary action—for example, giving a response or delivering the requested information.

"While speech enablement is not a new concept in consumer devices, the user experience has, and still is, proving to be quite restrictive in that it's two- or three-word-command–based; there's no conversational flow based on the natural language that you would use if you were talking to another person," said Andy Peart, chief marketing officer, Artificial Solutions, in a statement. "Our Teneo technology is powered by our NLI engine in order to deliver intelligent conversations between consumers and the everyday devices they use. Imagine having an intelligent virtual assistant on your smartphone, Smart TV, SatNav, games console, laptop, or tablet—not command-based but able to hold two-way conversations using everyday language."

Improvements to the underlying algorithms used by Teneo ensure that it is able to cope with typical ASR errors, such as poor grammar, fragmentary input, and superfluous small words. The improvements also allow Teneo to handle the differences between spoken and written language.

"There are many factors influencing the quality of ASR implementations, some of which depend on the user and the context, and some on the type of ASR system used," Peart said. "It's impractical to expect a user to come up with perfectly formed, grammatically correct, nonfragmentary sentences. By combining the capabilities of Teneo with the capabilities of ASR systems, the user has a far superior experience when talking to an application or device."


Page1 of 1