Researchers at the University of Ghent, Belgium, have combined a convolutional neural network with computational neuroscience to create a model that mimics the mechanics of a human snail. The details are published by the journal Nature Machine Intelligence.
Over the past several decades, great strides have been made in speech and voice recognition technology. But they all have one thing in common: no matter how they look, none of them work in real-time. Each is based on hardware and software that processes what is heard. In a new study, scientists suggest that the problem with existing devices is the complexity of the computations that need to be performed. To solve this, they created a model that simulates hearing in humans, which is based on combining the best characteristics of convolutional neural networks with computational neuroscience.
Convolutional neural network – a special architecture of artificial neural networks, proposed by Jan Lekun in 1988 and aimed at efficient pattern recognition, is part of deep learning technologies.
Hearing in humans depends on different parts of the ear. Sound enters the ear canal and meets the eardrum. It vibrates in response, sending signals to the bones in the inner ear that create ripples in the fluid in the cochlea. This fluid mixes the hair cells that line the cochlea. The movement of hair cells stimulates ion channels, which in turn generate signals sent to the brain stem. Researchers in Belgium have created an artificial intelligence (AI) system that has been taught to recognize sound and then decode it in the same way. They then connected their system to a model based on human anatomy. They called their system CoNNear, a working snail model.
Testing has shown that the system is capable of converting 20 kHz sampled acoustic waves into cochlear basilar membrane waveforms in real time, outperforming modern conventional systems by a wide margin. CoNNear acts as a cochlea 2,000 times faster than modern hearing aids. The researchers speculate that their findings will lay the groundwork for a new generation of human hearing aids or devices with enhanced hearing and speech recognition.