Ava (A Social Robot): Design and Performance of a Robotic Hearing Apparatus

Socially cognitive robots are supposed to communicate and interact with humans and other robots in the most natural way. Listeners turn their heads to-ward speakers to enhance communicative attention; this is also an act of appreciation to the speaker. In this paper we have designed and implemented a robotic head, “Ava”, which turns toward the speaker in noisy environments. Ava employs a Speech Activity Detection system which differentiates speech segments of non-speech. Then the speech segments are processed to reduce different kinds of noise levels. The speaker localization system then finds the speaker position in the azimuth plane and commands motors to turn horizon-tally toward the speaker in a smooth trajectory. Ava has two built-in micro-phones inside its ears and employs three different algorithms simultaneously for feature extraction and a two-layer perceptron neural network for localization. Ava operates real-time and updates the position even in its moving phase. Experiments show a precision of +/-5 degrees in white noise in SNR of 10 dB.