Periodic Reporting for period 5 - SOUNDSCENE (How does the brain organize sounds into auditory scenes?)
Reporting period: 2024-09-01 to 2025-05-31
Listening involves making sense of the numerous competing sound sources that exist around us. The neuro-computational challenge faced by the brain is to pull apart the sound mixture that arrives at the ear and reconstruct the original sound sources; this process is known as auditory scene analysis. While young normal hearing listeners can parse an auditory scene with ease, the neural mechanisms that allow the brain to do this are unknown – and we are not yet able to recreate them with digital technology. Hearing loss, aging, impairments in central auditory processing, or an inability to appropriately engage attentional mechanisms can negatively impact the ability to listen in complex and noisy situations and an understanding of how the healthy brain organizes a sound mixture. into perceptual sources may guide rehabilitative strategies targeting these problems.
While functional imaging studies in humans highlight a network of brain regions that support auditory scene analysis, little is known about the cellular and circuit based mechanisms that operate within these brain networks. A critical barrier to advancing our understanding of how the brain solves the challenge of scene analysis has been a failure to combine behavioural testing, which provides a crucial measure of how any given sound mixture is perceived, with methods to record and manipulate neuronal activity in animal models. In SOUNDSCENE we combine complex behavioural tasks, that mimic those that human listeners face in everyday situations, with methods to observe and manipulate neural activity. Our goal is to understand how a network of brain regions: auditory cortex, prefrontal cortex and hippocampus enable scene analysis during active listening. We will understand how processing within each area, and the interactions between these areas, underpins auditory scene analysis. This knowledge will increase our knowledge of fundamental brain function, and may contribute to biologically inspired machine listening devices, and improvements in hearing aid and cochlear implant signal processing methods.
While functional imaging studies in humans highlight a network of brain regions that support auditory scene analysis, little is known about the cellular and circuit based mechanisms that operate within these brain networks. A critical barrier to advancing our understanding of how the brain solves the challenge of scene analysis has been a failure to combine behavioural testing, which provides a crucial measure of how any given sound mixture is perceived, with methods to record and manipulate neuronal activity in animal models. In SOUNDSCENE we combine complex behavioural tasks, that mimic those that human listeners face in everyday situations, with methods to observe and manipulate neural activity. Our goal is to understand how a network of brain regions: auditory cortex, prefrontal cortex and hippocampus enable scene analysis during active listening. We will understand how processing within each area, and the interactions between these areas, underpins auditory scene analysis. This knowledge will increase our knowledge of fundamental brain function, and may contribute to biologically inspired machine listening devices, and improvements in hearing aid and cochlear implant signal processing methods.
During this project we developed multiple behavioural paradigms in an animal model, which provide us with a foundation to define the underlying neural mechanisms. Throughout the project, in parallel to our behavioural work in animals, we deployed near-identical tasks in humans. Each task pinpoints a particular aspect of auditory scene analysis. For example, in humans we explored how rapidly listeners can adapt to background sounds and whether the brain does this by building a model of the ‘noise’ as a distinct source. Consistent with this, we found that if the noise changed in type (for example from a lawnmower to running water) or in location, listeners required a similar period of time to adapt to the new noise background. We found that these abilities were preserved in listeners who were older, and in those with hearing loss.
In our animal model we used an identical task where animals identified brief speech sounds (artificial vowels), as well as tasks that required animals identify statistical regularities in sounds, or discriminated streams of speech in background noise. Through these tasks we were able to show that many of the listening abilities that humans deploy to make sense of sound in noisy situations are found in non-human animals. These include extracting statistical regularities, linking sounds from a common source according to their fundamental frequency, and preserving speech sound discrimination in background noise. We then recorded neural activity from a network of brain regions during behaviour in order to understand how the brain is able to ‘tune out’ background noises, and allow us to focus on sounds of interest.
We have published a number of scientific outputs (11 to date) including methods developments (see following section), behavioural results and some key findings from auditory cortical recordings. Multiple publications further manuscripts are in preparation for human work, and from neural data collected during task performance that explore how the auditory cortex represents multiple simultaneous sounds, and how auditory cortex is shaped by activity in other brain areas to allow us to select which sound in a mixture to listen to.
Our results have been disseminated through multiple scientific conference presentations, including keynote presentations and seminars by the PI, competitively awarded talks and oral papers, as well as poster presentations at national and international meetings around the world by the trainees funded by SOUNDSCENE. Trainees funded in whole or in part by SOUNDSCENE have written four PhD thesis, all of which are openly available on UCL Discovery.
Results from SOUNDSCENE are currently being used as a foundation for multiple grant applications and fellowship applications by the PI and trainees funded by the award.
In our animal model we used an identical task where animals identified brief speech sounds (artificial vowels), as well as tasks that required animals identify statistical regularities in sounds, or discriminated streams of speech in background noise. Through these tasks we were able to show that many of the listening abilities that humans deploy to make sense of sound in noisy situations are found in non-human animals. These include extracting statistical regularities, linking sounds from a common source according to their fundamental frequency, and preserving speech sound discrimination in background noise. We then recorded neural activity from a network of brain regions during behaviour in order to understand how the brain is able to ‘tune out’ background noises, and allow us to focus on sounds of interest.
We have published a number of scientific outputs (11 to date) including methods developments (see following section), behavioural results and some key findings from auditory cortical recordings. Multiple publications further manuscripts are in preparation for human work, and from neural data collected during task performance that explore how the auditory cortex represents multiple simultaneous sounds, and how auditory cortex is shaped by activity in other brain areas to allow us to select which sound in a mixture to listen to.
Our results have been disseminated through multiple scientific conference presentations, including keynote presentations and seminars by the PI, competitively awarded talks and oral papers, as well as poster presentations at national and international meetings around the world by the trainees funded by SOUNDSCENE. Trainees funded in whole or in part by SOUNDSCENE have written four PhD thesis, all of which are openly available on UCL Discovery.
Results from SOUNDSCENE are currently being used as a foundation for multiple grant applications and fellowship applications by the PI and trainees funded by the award.
Through the course of SOUNDSCENE we made a number of key technological advances that have allowed us to advance the state of the art. These include the use of optogenetics in ferrets to reveal that auditory cortex is necessary for listening in noise, the use of machine learning to analyse the rich behavioural datasets that we generate from trained animals in listening tasks, and methods for implanting state of the art recording electrodes (“Neuropixels”) in a way that allows free-moving behaviour and neural recordings that persist for months to years.