Experts at the University of California have created a device that converts the activity of a bird’s brain into a song.
In the new work, scientists first obtained a “pattern” of the brain activity of zebra finch birds using silicon electrodes implanted in the sensorimotor area. It is in this area that the muscles of the bird’s vocal apparatus are controlled.
The authors loaded the brain activity data into a neural network – it broke the recorded bird songs into small patterns and connected them with the patterns of the birds’ brain activity.
As a result, scientists literally read the thoughts of the bird and reproduced them in a trill, familiar to us. Moreover, the reproduction accuracy turned out to be very high – the device was able to recognize the smallest details of the bird’s vocals, right down to the pitch, timbre and volume of sounds.
If you need to simulate every nuance, every detail of the underlying sound, then the problem of reproduction becomes much more difficult. And with a simple view, the system can learn data – more reliable and suitable for a wider range of conditions and behavior.
Vikash Gilya, author of the work
At the moment, the device cannot work in real-time, because the birds, when singing, constantly change the vocal pattern of trills, adjusting to the surrounding conditions. The authors plan to solve this problem and in the future use a speech prosthesis for mute people.