The Third Clarity Prediction Challenge (CPC3)
In the Clarity prediction challenges, we are encouraging the development of novel approaches to the intelligibility prediction of speech processed by hearing aids.
Machine Learning applied to problems in audiology
Machine learning models of hearing demonstrate the limits of attentional selection of speech heard through cochlear implants
Machine learning models trained on simulated auditory nerve representations demonstrate the limits of attentional selection of speech in cocktail party scenario when heard through a cochlear implant.
A definition of lyric intelligibility by hearing aid users: A focus group study
This study explored how hearing-impaired listeners define lyrics intelligibility in music, providing a working definition for evaluating improvements and developing new signal processing strategies.
Automated scoring of speech audiometry for cochlear implant listeners using artificial intelligence and flexible grading methods
This project explores the use of Automatic Speech Recognition (ASR) models to transcribe and score open-set speech audiometry responses from cochlear implant users, demonstrating that ASR models can approximate benchmark scores while significantly reducing transcription time.
Exploring the standard Hughston-Westlake protocol in tone audiometry using Gaussian process regression: a virtual patient framework
Intermediate data generated in pure-tone audiometry, by our patient response simulator (Hughson-Westlake protocol), help to understand the probabilistic and continuous representation of a patient's hearing ability derived from Gaussian process inference used in computational audiometry.
Enhance, Remix, Compress: Comodulation Effects of Dynamic Range Compression in Speech Enhancement pipelines
Does applying dynamic range compression separately to speech and noise improve intelligibility compared to joint compression, when using separated sources in non-stationary noise?
Decoding the infant brain: Neural predictors of language and learning difficulties in neurodiverse population
This work examines how neural signals recorded during infancy can serve as early biomarkers for predicting speech, language and learning difficulties in neurodiverse populations.
Audiometrically normal, but attending an audiology clinic: using machine learning to learn more about these clients
Using machine learning, an analysis of a large clinical audiology database, throws light on clients who have normal hearing thresholds, but report hearing difficulties.
Using CARFAC-JAX, a fast, differentiable model of the human cochlea, to efficiently fit personalized hearing loss
When implemented in JAX, CARFAC not only displays rapid calculation time but also allows differentiation relative to its parameters. This facilitates the fast and efficient generation of personalized models of hearing impairment.
Speech Stimuli Creation using Deep-Learning Based Voice Conversion
A deep-learning-based pipeline for effectively disentangling speaker-specific attributes from linguistic content, enabling the creation of parametric voice stimuli at high sampling rates.
Modeling Auditory Attention with Deep Neural Networks
In this work, we instantiate a deep learning model of feature based attention that performs on-par with human listeners in the cocktail party problem.
From business need to model monitoring: Optimizing manufacturing of custom hearing aids with machine learning
This presentation will introduce a framework used for developing and deploying cloud-based machine learning models in the context of a project aimed at optimizing manufacturing of custom (in-the-ear style) hearing aids by improved detection of electroacoustic test failures due to feedback.
Neural Networks Hear You Loud and Clear: Hearing Loss Compensation Using Deep Neural Networks
This study investigates using deep neural networks to provide hearing loss compensation for people with hearing loss, showing promising results in speech intelligibility and perceived quality.
Introduction to Computational Audiology: Past, Present, and Future Applications
A journey around the world of AI and Audiology within 80 slides!
Analyzing brain connections in sound tolerance disorders using fMRI and machine learning
Machine learning models identify altered spontaneous brain connections in sound tolerance disorders
Podcast Episode 3: A holistic perspective on hearing technology
n this episode, Brent Edwards from NAL and Stefan Launer from Sonova take us through their careers and share lessons and perspectives on the development of hearing technology. We discuss how the development of technology becomes more holistic, design thinking, standardization, and what's needed to get to new service models and innovation.
Podcast Episode 2: Automated Speech Recognition (ASR) for the deaf
Automated Speech Recognition (ASR) for the deaf and communication on equal terms regardless of hearing status. Episode 2 with Dimitri Kanevsky, Jessica Monaghan and Nicky Chong-White. Moderator Jan-Willem Wasmann You…
Podcast Episode 1: Bayesian Active Learning in Audiology
Active Learning in the auditory domain A round table with Bert de Vries, Josef Schlittenlacher and Dennis Barbour. Moderator Jan-Willem Wasmann Audio version: Audiovisual version on Youtube: Josef Schlittenlacher is…
The goal is to find the model instance (folder from shelf y) that has the maximum likelihood to have generated the experimental data set d. The stimulus that leads to the smallest variance of parameter estimates is presented next. The process is repeated until the termination criterion is met.
Model-based selection of most informative diagnostic tests and test parameters
The model should conduct the experiment because it knows best which condition is going to be the most informative for confining its free parameters, at least in theory
Detecting hearing loss from children’s speech using machine learning
Here we explore the potential for using machine learning to detect hearing loss from children's speech.
Use of a deep recurrent neural network to reduce transient noise: Effects on subjective speech intelligibility and comfort
Transient noise reduction using a deep recurrent neural network improves the subjective speech Intelligibility and comfort.
(A) We process signals with a binaural processing stage. The signal is converted to phoneme probabilities using a DNN. The degradation of these is used to predict binaural SI. (B) Results for subjective data, our model (BAPSI) and baseline models in terms of speech recognition threshold (SRT).
Binaural prediction of speech intelligibility based on a blind model using automatic phoneme recognition
In this study, we show that phoneme probabilities from a DNN can produce good estimates of speech intelligibility when combined with a blind binaural processing stage.
Estimated distortion component of hearing impairment from the model's prediction errors as a function of the average hearing loss.
Estimating the distortion component of hearing impairment from attenuation-based model predictions using machine learning
Attenuation-component-based model predictions of speech recognition thresholds like FADE seem to facilitate an estimation of the supra-threshold distortion component of hearing impairment.
Figure 1 A: Data streamed from the hearables to a PC running a recording script. B. Data from two different activities, recorded over 5 minutes each. C. Results from a classification using Naïve Bayes classification and a 5-fold cross-validation procedure.
Automatic detection of human activities from accelerometer sensors integrated in hearables
Using a Naive Bayes classifier, we could show that twelve different activities were classified above chance.
Schematic showing experimental set-up including the placement of the participant, loudspeakers and two observers.
A classification approach to listening effort: combining features from the pupil and cardiovascular system
Training k-nearest neighbor classifiers to predict intelligibility, social state and participant perception of a listening task
Overview of the decision tree to diagnose somatosensory tinnitus
A data-driven decision tree for diagnosing somatosensory tinnitus
Based on the results of an online survey, we developed a decision tree to classify somatosensory tinnitus patients with an accuracy of over 80%.
Clarity Challenge Enhancement Challenge
The Clarity project is running a series of machine learning challenges to revolutionise signal processing in hearing aids.
Opening the Black Box of Binaural Neural Networks
Simulated binaural neural networks show that sharp spatial and frequency tuning is needed to accurately localize sound sources in the azimuth direction.
Modeling speech perception in hidden hearing loss using stochastically undersampled neuronal firing patterns
Modeling speech perception in hidden hearing loss using stochastically undersampled neuronal firing patterns
Prediction of speech recognition by hearing-aid users: the syllable-constituent, contextual model of speech perception
Speech perception by hearing aid (HA) users has been evaluated in a database that includes up to 45 hours of testing their aided abilities to recognize syllabic constituents of speech, and words in meaningful sentences, under both masked (eight-talker babble) and quiet conditions.
On nonlinearity prediction of Hearing Aid behavioural measures using Speech Auditory Brainstem Responses and Decision Trees
PTA is more enlightening than speech-ABR to predict aided behavioural measures.
Predicting tinnitus from auditory brainstem responses using machine learning
This study used machine learning to identify normal-hearing listeners with and without tinnitus based on their ABRs.
Machine Recognition of Overlapping Speech Encoded in the Midbrain
A system that predicts and identifies neural responses to overlapping speech sounds mimics human perception.
Use of air conduction thresholds to predict bone conduction asymmetry and air-bone gap
This study used machine learning methods to predict bone conduction abnormalities from air conduction pure tone audiometric thresholds.
Random Forest Classification to Predict Response to High-Definition Transcranial Direct Current Stimulation Therapy for Tinnitus
A random forest classifier can predict response to high-definition transcranial direct current stimulation treatment for tinnitus with 82.41% accuracy.
Computational modelling of the human auditory brainstem response to natural speech
The computational model consists of three main parts (auditory nerve, inferior colliculus and cochlear nuclei). The figure shows the input (natural speech) and the neural outputs at the different levels.
Final masked AMLAG results for one participant (127) with a left cochlear implant and no residual hearing. Red diamonds denote unheard tones and blue pluses denote heard tones. The most intense tones at lower frequencies in the left ear were effectively masked.
Dynamically Masked Audiograms with Machine Learning Audiometry
Dynamically masked audiograms achieve accurate true threshold estimates and reduce test time compared to current clinical masking procedures.
Predicting Hearing Aid Fittings Based on Audiometric and Subject-Related Data: A Machine Learning Approach
A machine learning model is trained on real-world fitting data to predict the user's individual gain based on audiometric and further subject-related data, such as age, gender, and the acoustic environments.
Computational modeling of attentive voice tracking
This work presents a CASA model of attentive voice tracking.