Research: Computer Audition

Ohnishi Lab.

Lab. of Biological Information Engineering
Introduction of Research

Computer Audition

Auditory system is of equal importance as visual system which conveys a lot of information, because it has the advantages that are not present in vision. Audition is omni-directional, acts in the darks, and can make us notice with sound sources in hiding or other rooms. The time resolution is quite high, and the low computational cost make it possible to understand a scene in a faster way. Moreover, audition plays an important role on early perception of changes in the environment, such as in the case where we usually look back when a sound is heard behind us.

We have the ability to identify and localize sound sources. In the case of human voice, we can understand what one says and even recognize one's emotion.

We are conducting researches on localization, separation and recognition of multiple sound sources as an understanding of scene with audition, and Kansei information processing based on audition for higher level process.

Research theme Multiple Sound Source Localization & Separation from their Mixture Localization of multiple sound sources in the environment using temporal disparity between microphones and the preceding effect, and separation of sound sources based on the localization result from their mixture. Echo Canceler Using a Single Microphone Removing echoes in acoustic signal captured by a single microphone without knowledge about the environment. A Mobile Robot with Ears and An Eye The developed robot can estimate the direction of sound source with microphones as ears, move towards the source avoiding obstacles, and discriminate whether the source is a human or not by a camera as an eye. Extraction of Kansei Information from Speech We are studying how to estimate speaker's Kansei Information - happiness, anger, sadness, fear, disgust - from speech not relying on uttered words.