Auditory Psychophysics Research Study

As for my final year research project from December 2019 - March 2020, I worked with the Auditory Cognition Group at Newcastle University Medical School. The research title given to me is: Mechanism for hearing speech in noise. I conducted an Auditory Psychophysics experiments on participants in Newcastle upon Tyne, United Kingdom.

Let me break down the title. We all understand ‘mechanism’ is a process by which something takes place or brought about. So, what is ‘hearing speech-in-noise’ ?

Have you ever been in a noisy social setting? Picture yourself in a cocktail party with your work colleagues, and you are having an interesting and clear conversation with your colleague across the table from you, despite the background noises around you such as the musicians, barista and other chatters. This situation is well-known as the cocktail party effect; the phenomenon of being able to focus one’s auditory attention on a particular stimulus while filtering out a range of other stimuli.

It is not unusual for normal hearing individuals to experience hearing difficulty in these noisy situations however, it was commonly misunderstood as a hearing loss and would seek clinical help. Thus, this recurring concern led to a prevalence of clinical referrals in the UK of 5% for difficulties in speech-in-noise perception despite having a normal hearing threshold. ‘Normal’ hearing is usually defined from the audiometric evaluations in an audiogram as shown below, but there is no standard procedure to assess normal hearing individuals with speech-in-noise difficulties. A normal healthy hearing threshold is from 0 to 20 decibels (dB) and from 250-8000 Hz as indicated by the red line. So, at any range of frequency above ≥20 dB is classified as mild to profound hearing loss according to the British Society of Audiology recommended procedure.

In this study, we used a series of psychophysics behavioural tests to assess discrimination tasks and speech-in-noise (SPIN) task performance. An established behavioural test developed by Teki et al (2011); Stochastic figure-ground (SFG) that was applied in this study for discrimination tasks to assess grouping mechanisms that require high-level analysis in the brain cortex. This spectrogram is only for illustrative purposes.

However, we applied the new method developed by Holmes & Griffiths (2019) where a ‘gap’ occurred in the ‘figure’ component within the ‘background’ component, as shown in this right spectrogram. The black dashed line represents the ‘background’ chords; made of random frequency tones and orange dashed line represents as ‘figure’ chords; made of pure tones of fixed and constant frequency.

In this SFG Discrimination tasks, a ‘figure’ will pop out from the ‘background’ chords with or without ‘gap’ component. The idea of the ‘gap’ stimulus is that normal speech consists of a few pauses, thus successful participants that able to discriminate ‘gap’ in the ‘figure’ component of the stimulus corresponding closely to the speech-in-noise perception. To assess SPIN (speech-in-noise) tasks, the Oldenburg matrix set combined with either 16-talker or 3-talker babble noise, and participants were asked to report sentences. The SFG stimulus in this study has a fixed setting for ‘figure’ component lasting on average 42 chords (2100ms) and ‘background’ lasting exactly 70 chords (3500ms).

Hypothesis

So, we hypothesised that performance in SFG discrimination tasks predicts performance in SPIN tasks and would also be a better predictor of speech-in-noise tasks with 3-talker babble noise (SPIN3). As our ‘background’ structure of the synthetic stimuli is more alike to the 3-talker babble noise, it was predicted that the performance of the SFG tasks would be similar and a better predictor of the SPIN3.

Here is the study flowchart, a total of 65 participants (19-70 years old) were recruited from Newcastle Upon Tyne. Participants were self-reported to meet the criteria of the study and written informed consent was obtained. The full experiment lasted for approximately 1.5hours and was conducted in a soundproof room located in the Medical School. Participants were assessed in both ears at baseline by using a diagnostic audiometer and plotted in an audiogram as shown previously. Participants with a normal hearing threshold of an average of ≤20 dB across 6 tested frequencies (0.25-8 kHz) were eligible to proceed to the behavioural study (computer-based task). Online structured questionnaires were used to obtain information on medical history mainly focusing on ear health, spoken language(s), and any musical experience.

SFG Discrimination Tasks

A previous study by Holmes & Griffiths (2019) showed a correlation between our initial test of figure-ground analysis and speech-in-noise perception. Here, we are testing new variations on the original test to see if we can demonstrate a better correlation. We applied the original figure-ground test from the previous study that only tested on TMR (target-to-masker ratio), but in this study, we manipulated with 2 different aspects of SFG; coherence and gap duration. So, a total of 3 SFG discrimination tasks and each adapting on either:

1. TMR: target-to-masker ratio equals the manipulated intensity of the foreground figure

2. Coherence level: manipulated the number of frequency elements in the figure

3. Gap duration: manipulated the length the gap in the figure

In each of these SFG discrimination tasks including SPIN tasks, we used an adaptive tracking procedure applied to the independent variable in order to establish a threshold. This allowed us to test the hypothesis, portraying the different ways of defining the threshold correlate with SPIN performance for every participant. In the behavioural study, the participant’s task was to detect which interval in each SFG discrimination task contained a gap in the ‘figure’ component by selecting the answer on the response screen. The SFG discrimination task results will give three different sets of thresholds (SFG TMR, SFG Coh and SFG Gap).

SPIN Tasks

SPIN tasks adapted only on TMR and stimuli settings were set accordingly. TMR was adapted in between the sentence and the respective 16-talker or 3-talker babble noise. Participants will hear a recorded male voice with a British accent speaking different sets of sentences that are presented as targets and were adapted from the Oldenburg matrix set. Each sentence was presented simultaneously with either 16-talker or 3 talker babble noise. The format of the 5-word sentences are “<Name><verb><number><adjective><noun>” as shown below. These are the ten different word options given for each word. Participant’s task was to listen and report the correct words by selecting the given options in any order that were presented on the response screen. For example, “Nina sees nine old windows”. The results will give two sets of thresholds; SPIN16 (speech-in-noise with 16-talker babble) and SPIN3 (speech-in-noise with 3-talker babble).

Results

Statistical analysis was performed on Minitab software. The conclusions for the correlation analysis were based on the regression analyses and P-values. The values obtained were either Pearson’s or Spearman’s correlation coefficient values and displayed on the respective bar graphs. We analysed with Pearson’s correlation coefficients when the threshold data was a normal distribution (parametric) while Spearman’s rho correlation for a not normal distribution (non-parametric). Spearman’s rho correlation is used when either of the parameters in a bivariate correlation is not normally distributed (non-parametric).In both SPIN16 and SPIN3, only SFG Coherence and SFG Gap discrimination threshold were analysed with Spearman’s correlation and the rest were Pearson’s correlation. Asterisks indicate the significance level of the correlation coefficient. Here are the results of the correlation between i) SPIN3 and ii) SPIN16 with each of the SFG discrimination tasks, audiometric thresholds, and age.

As you can see that in the respective SPIN results (SPIN3, SPIN16), both SFG TMR and SFG Coherence showed no significant correlations indicated by an asterisk but there are significant correlations for Audiometric thresholds and age variable. It was expected that age variable results will show significant positive correlations as older age participants performance are usually worse in speech-in-noise perception. Thus, our findings of SFG-Gap stimuli would be able to predict speech-in-noise performances (SPIN16 and SPIN3) alongside with audiometric threshold evaluations. However, SFG discrimination tasks would not be a better predictor for SPIN3 as all 3 SFG discrimination thresholds results were not significant enough due to SFG TMR and SFG Coherence Discrimination non-significant thresholds thus affecting the outcome of the results giving no positive and significant correlations.

Conclusion

In conclusion, our significant results supported SFG-Gap tasks as a predictor in speech-in-noise performance (SPIN16 and SPIN3) but SFG discrimination task could not be a better predictor for SPIN3 as mentioned above. The mechanism for SPIN perception involves both peripheral and central auditory processes such that people with normal hearing faced difficulties in speech-in-noise for different reasons, for example; age, language experience (lack of English language literacy), peripheral deficits etc. We also pointed out that SFG-Gap tasks could work alongside clinical procedure to identify speech-in-noise deficits by assessing with both peripheral (audiometric threshold) and central (grouping) processes.

References

1. Bee M, Micheyl C. The cocktail party problem: What is it? How can it be solved? And why should animal behaviorists study it?. Journal of Comparative Psychology. 2008;122(3):235-251.

2. Hind S, Haines-Bazrafshan R, Benton C, Brassington W, Towle B, Moore D. Prevalence of clinical referrals having hearing thresholds within normal limits. International Journal of Audiology. 2011;50(10):708-716.

3. British Society of Audiology. Recommended Procedure: pure tone air and bone conduction threshold audiometry with and without masking and determination of uncomfortable loudness level (2004).

4. Teki S, Chait M, Kumar S, von Kriegstein K, Griffiths T. Brain Bases for Auditory Stimulus-Driven Figure-Ground Segregation. Journal of Neuroscience. 2011;31(1):164-171.

5. Holmes E, Griffiths T. ‘Normal’ hearing thresholds and fundamental auditory grouping processes predict difficulties with speech-in-noise perception. Scientific Reports. 2019;9(1).

I would like to thank Professor Timothy Griffiths and Dr Pradeep Dheerendra for their guidance and support throughout my research project. I truly enjoyed the experience of working with researchers and neuroscientists. And to all my friends for the support in my research study as my participants and/or proofreading my 5k word research report!

Working from home, that is pretty much all I do now (don't mind the vacuum behind me!)

I love my current Starbucks mug that my brother got for me during his Japan trip early this year! (way before COVID-19 pandemic)

Follow me on my

LinkedIn & Instagram: camilarahim