Irving, TX (PRWEB) April 02, 2014
Resolvity, Inc. announced that it was awarded U.S. patent 8,682,660 “Method and System for Post-Processing Speech Recognition Results”. The patent describes an innovative use of Machine Learning for reducing systematic recognition errors in automatic Speech Recognition.
One of the inventors, Dr. Jacek Jarmulak, Resolvity’s VP of Product Development, explained: “Our invention addresses a common scenario where a speech recognizer with Statistical Language Models is used. Such a recognizer may, for a number of reasons, produce incorrect classifications of certain input utterances. Typically, these errors are not random but have a systematic bias due to various factors in the audio channel, SML implementation, training corpus selection, etc. What our invention does is correct for those systematic errors.”
The post-processing system is fully automatic and is trained using data collected during normal speech recognizer operation. This training corpus does not have to be fully transcribed, simple category labeling is sufficient, which keeps the cost low.
“The invention allows us to improve recognition accuracy for complex speech interactions. One example where the invention improves speech recognition accuracy is in Natural Language Call Routing applications,” said Mr. Mahesh Rajagopalan, co-founder & President of Resolvity.
About Resolvity: Resolvity’s cloud speech recognition software helps companies automate a wide variety of inbound and outbound customer support calls. Resolvity automates several million calls on a monthly basis for leading Fortune 1000 companies in the US using its cloud-based speech recognition IVR.
Resolvity solutions range from natural language call-routing to sophisticated customer care, billing, and technical support calls. The company has deep domain expertise in cable & telecommunications, financial services, healthcare, high-tech, and utilities market place. For more information, please visit http://www.resolvity.com.