快猫短视频

WHEN Hideyuki Sawada says he wants his robots to be articulate, he鈥檚 not talking computer-generated voice synthesis. Rather, he wants them to talk just like humans. So he鈥檚 designing an entire artificial voice system, complete with its own lung, windpipe, vocal cords and throat.

The idea is to make interacting with robots more natural. Unlike the stilted metallic utterances of a 鈥淩obbie the Robot鈥� type droid, Sawada鈥檚 creations at Kagawa University in Japan will have a natural-sounding voice that produces sound in the same way as the human vocal tract.

To mimic a human lung, Sawada uses a compressed air tank which forces air into a plastic voice-box chamber, where it makes rubber 鈥渧ocal cords鈥� vibrate (see Diagram). The basic sounds generated in the voice box are then fed to a flexible tube that mimics a human vocal tract.

The sound this produces depends on the speed of the airflow, the tension of the rubber vocal cord and the shape of the vocal tract鈥檚 cross-section. The tract is made from a flexible silicone tube, so that motor-powered rams positioned along it can alter its shape. Our throats and mouths work in a similar way to damp out certain frequencies generated by the vocal cords.

To control his robot鈥檚 vocal system, Sawada uses a neural network that can learn how to produce particular sounds by listening to its own utterances and then adjusting the airflow and the shape of the vocal tract to get closer to a desired sound. Sawada has already taught his system a full range of vowel sounds, and has found that the shape taken up by the artificial vocal tract as it makes the sounds is similar to the shape seen in X-rays of a person making the same sounds. 鈥淚t shows we are on the right track,鈥� he told 快猫短视频.

Consonants are harder, Sawada wil tell the International Conference on Robotics and Automation in Washington, DC next week. The end of the artificial vocal tract behaves much like a pair of lips and allows it to pronounce the sounds 鈥減鈥� and 鈥渢鈥�. But to extend its repertoire further, Sawada will first have to give the system a tongue, something he鈥檚 not yet sure how to do. He has, however, added a vent to the resonance tube that does the same job, vocally speaking, as the nasal tract, making the vowel sounds more authentic.

Sawada鈥檚 system, which he has built with Shuji Hashimoto at Waseda University, can鈥檛 match the quality of digital voice synthesisers connected to loudspeakers. But he claims that in time he will be able to make his system more realistic. Eventually he plans to install a complete version into a humanoid robot.