A special hardware and software complex recognizes the speech of people even in noisy places.
The Russian company Promobot, a manufacturer of service robots and a Skolkovo resident, has created a special version of the microphone array that serves as ears for the robot. With the help of the device, robots hear and recognize human speech in noisy places – an airport, an MFC or a shopping center. The array includes a complex of eight microphones and a software algorithm. It has no analogues in the robotics market.
Development and testing lasted a whole year – from January 2020. The microphone array consists of eight microphones that function as the outer ear in the human body. With their help, the robot detects noise at a distance of up to 2 m from it, the detection range is 360 degrees, like a human. The algorithm provides capture of analog data from microphones, their conversion, echo cancellation and data transmission. The program “separates” human speech from noise – just like the inner ear works in our body. Then the algorithm recognizes phrases, thanks to which the robot enters into a dialogue. Unlike voice assistants, which react only to clear speech of a person, Promobot reacts to all the noise – and already “inside it” it looks for a human voice, both for children and adults.
Unlike other systems, Promobot speech recognition practically does not depend on distance: over the year, the percentage of recognition at a distance of 1 m from the robot has grown from 25 to 75% – three times. This is the highest indicator in the world for service robots, and this, according to the developers, brings the microphone array as close as possible to the human ear.
“The peculiarity of speech recognition in service robotics is that our devices operate in crowded places. Microphones in smartphones, headphones and smart speakers are personal devices, they are designed to work only with the owner, often only at home, says Oleg Kivokurtsev, Development Director of Promobot. – Unlike a smartphone, a robot works in difficult noisy environments – an airport and a shopping center. He not only needs to hear a person’s speech – it is not so difficult, but also to process it and recognize, as the human ear does. This is what our team has been working on for the last year.”
The device was patented in the fall of 2020. Chinese robot manufacturers have become interested in the microphone array – the Russian company plans to start selling the device in January 2021.