An MIT study has taken a significant step towards the development of computational models capable of mimicking the structure and function of the human auditory system. The models could have applications in the production of improved hearing aids, cochlear implants, and brain-machine interfaces. The researchers discovered that modern machine learning-derived computational models are progressing towards this objective and demonstrated that the majority of these models generate internal representations similar to those seen in the human brain when listening to the same sounds.
This study was praised for being the most comprehensive to date in comparing these kinds of models to the human auditory system. It found that models trained on auditory input including background noise more accurately mirrored the activation patterns of the human auditory cortex. The researchers analysed nine deep neural network models which had been trained to perform auditory tasks and developed 14 of their own. These models were mostly trained to perform a single task, such as word recognition or identifying sounds.
Experimentation showed that the models which best represented the human brain had been trained on more than one task and included background noise in their auditory training. This multi-task and noise inclusion training made the models better at predicting human brain responses. This corresponds with the knowledge that human hearing often needs to process sounds in noisy environments.
Furthermore, the study suggests the human auditory cortex has a hierarchical organisation. Researchers found that earlier stages of the model most closely matched the primary auditory cortex, whilst later stages were more similar to the areas beyond the primary cortex. Models trained for different tasks better replicated different aspects of audition.
The MIT team now aim to use their findings to create even more successful models in mirroring human brain responses. Producing models capable of predicting brain responses could open many pathways for further research, leading to a better understanding of how the brain operates. Ultimately, these models could be utilised for the development of superior hearing aids, cochlear implants and brain-machine interfaces, optimising functionality for users.
