Jul 14, 2011
Talking Robot Mouth Mimics Human Speech
This mechanical talking robot mouth, developed by the Sawada Group at Kagawa University, features speech organs which imitate a humans.
"We didn't bring the air pump which works as the lungs this time, but air is sent from the pump, and goes through a section equivalent to the vocal cords, which vibrate. When that happens, resonance occurs in the parts that touch the mouth. So by moving the mouth, the robot changes the volume of air in the cavity and forms words."
The robots mouth and tongue are made from silicone rubber, and to enhance the sound that is made, several types of silicone were blended together. The inside of the nose is made of plaster, which give it similar characteristics to the human nose, which consists of bone and membranes.
"To pronounce "m" and "n" sounds, you need this nasal cavity, and the tongue is used to pronounce "r" sounds."
One special feature of this robot is that it can learn through auditory feedback. It listens to the sound it is making through a microphone, and then determines by itself what motions will make it sound more human.
"Even if a person doesn't know a word or what it means, they can make the sound of the word. That is, people use data they've already learned to guess how to move the inside of their mouth to make a certain sound. So we started researching autonomous learning. This robot can't be taught every sound, but it can be taught to some extent, and then it can use the results of its learning. So even if it hears a sound it doesn't know, it can estimate the movement needed to make that sound. Our research on autonomous learning has gone quite well. Even if the robot hears an unknown sound, it can estimate the movement that's needed, and reproduce the sound."
"We've given the robot the score for the song Kagome, kagome. It's already learned how to make the sound "ka," and it'll retrieve that information from its brain. The robot's also learned how to change a sound's pitch by changing the air flow, so it'll retrieve that data. In other words, it'll use the information already in its brain to make sounds, while estimating speech movements based on the score. So all we are giving it is the words and the pitch."
Professor Sawada has been doing this research for over a decade. He says the next step to make it's voice more like a human's will be to add teeth, so it can produce what are called fricative consonants, such as 'f' and 's'.
- This Week
- This Month
- All Time