Human ability to attend to a single voice in the presence of background interference is remarkable. If this could be imitated in practical technology it would be of great benefit for automatic speech recognition and other applications. Researchers at Ohio State University have demonstrated computational methods for separating speech from interfering sounds that imitate human auditory processing, and that have achieved levels of performance clearly evident to the untrained ear. Aetion Technologies will partner with Ohio State to carry this research to commercial application. Aetion is well suited to the task by reason of its technical competencies, its physical proximity to Ohio State, its official status as a University Technology Commercialization Company, and its business competencies. The effort will focus on monaural processing.
Benefits: An effective system for separating speech from acoustic interference would greatly facilitate many applications, including automatic speech recognition (ASR) and speaker identification. Market demand for these applications is very great; the constraint has been the availability of effective solutions. Application areas are diverse, but attenuating acoustic interference comprises only part of the solution. It is unlikely to form the basis for a business-to-consumer start-up. Aetion will seek to fully develop the capability and license it as an OEM to companies developing full applications for speech to text , speech command, and improved intelligibility for voice communications. Likely customers include such premier names as IBM, AT&T, Microsoft, Motorola, and Sony. The potential of this market is likely to attract venture capital should it be needed to fully develop the capability. Abstract: monaural speech separation, auditory scene analysis, coupled oscillators, abductive inference, imitation of human speech processing
