Speech recognition starts sooner than expected

Linguistic signals are not first processed in the cerebral cortex

How does our brain process language? © Giii / istock
Read out

Do you understand what I'm saying? Our brain apparently processes language differently than thought. As a study suggests, speech recognition does not take place in the cerebral cortex. Instead, the processing of the hearing information already begins in the pathways on the way there. In the future, this knowledge could help with the further development of language assistants - but also improve the understanding of dyslexia.

Whether Siri, Alexa or the Google Assistant: Language assistants are an integral part of everyday life today. You turn on devices, present the latest news and inform you about the weather forecast. That these systems can recognize our language so well, we have to thank for advances in the field of artificial intelligence. Through the use of artificial neural networks, speech recognition has become better and better in recent years.

However, these systems are not as good as the natural model. "The human brain works much better than computer-based speech processing, and it will probably stay that way for a while. Because the exact processes of speech processing in the brain are so far largely unknown, "explains Katharina von Kriegstein of the Technical University of Dresden.

What is being said?

For example, where does the processing of language begin? Up to now, the language centers of the cerebral cortex have been the decisive points of action in this context. However, Kriegstein and her colleagues around first author Paul Mihai have now found evidence that speech recognition starts much earlier, namely on the way to the cortex.

For their study, the scientists investigated which regions of the brain are particularly active in speech processing. Thirty-three subjects were given short voice recordings by different speakers. They were given one of two instructions: Either they should find out if the heard voice was the same as the previous one, or they should compare their own syllables. display

Medial knee-raiser in the brains of test persons: Already here the speech recognition apparently begins. Mihai et al. 2019

Left tram in focus

The first task was therefore to identify the speaker, while the other was the focus of the content of the spoken language language had to be processed. Would the subject's brain activity differ depending on the task? The research team observed this using magnetic resonance imaging (MRI).

The results showed that when the study participants performed the speech recognition task, a structure in the left horseshoe, the so-called medial knee hocker, was surprisingly particularly active. The activity of this region was all the higher the better the subjects were in recognizing the syllables.

Processing starts earlier

According to the scientists, this indicates that the processing of hearing information already begins in the pathways from the ear to the cerebral cortex and not when they reach the cortex. "We have some evidence for some time now that the cableways are more specialized in language than previously thought. This study now shows that this is indeed the case, "commented Kriegstein.

"Much of the language research focuses on structures in the cerebral cortex. But also investigating the contribution of subcortical structures to speech perception is enormously important for our understanding of human speech recognition. This investigation brings us a decisive step in this direction ", explain the researchers.

Meaning for read spelling errors?

In the future, the new findings could not only help with the further development of computer-based speech recognition systems. The findings may also be relevant to some symptoms of reading-spelling disorders, the researchers report. For example, it is known that the left medial knee joint works differently in the case of those affected than in persons without this disturbance.

Among other things, the specialization of the left medial knee-joint on language could, in the opinion of the team, explain why dyslexia patients often have difficulty understanding speech signals in noisy environments. Further studies will investigate this relationship in the future. (eLife, 2019; doi: 10.7554 / eLife.44837)

Source: Technische Universität Dresden

- Daniel Albat