These are just the first steps
Modern neural networks are based on the idea of being similar to the neural networks of the human brain. Scientists from Indiana University in Bloomington decided to see if it was possible to get rid of the need to create something similar and just take the original. To do this, they grew human brain organoids and tested whether they could be used, just as we use modern neural networks in conventional computers. That is, replace silicon with organics.
This is only the first experiment, and the authors of the project do not draw any big conclusions, but at least they received some positive results.
They called their system Brainoware. The experiment showed that brain organoids can be taught to recognize speech, just as artificial neural networks do, although with a number of reservations.
The article used a Japanese vowel sound database to test speech recognition. The researchers took 240 audio files of isolated Japanese vowel sounds from eight different male voices. These audio files were converted into sequences of electrical stimuli that were delivered to the brain organoid. These actions caused a certain activity in the organelles, which was recorded. In fact, this was the process of training the neural network.
This process and testing process included the following steps:
- The audio files have been converted into electrical impulses that can be interpreted and processed by the brain organoid, since it itself does not have the biological means to perceive sound. Simply put, he has no ears.
- The Brainoware system was then trained on neural network activity data to recognize and classify vowel pronunciations from different speakers.
- After training the system, testing was carried out using the same vowels from the same speakers, but to see how accurately the system could recognize vowels without taking into account differences in pronunciation.
- After training and testing, accuracy in speech recognition was assessed. The process involved analyzing an error matrix to determine how accurately the system recognized vowels.
In summary, the results showed that Brainoware could improve its speech recognition accuracy after repeated training. Simply put, the organoid’s neural network learned to do a specific job better without being initially programmed for anything.