Challenges To Revolutionise Hearing Device Processing

Lead Research Organisation: University of Nottingham
Department Name: School of Medicine

Abstract

Abstracts are not currently available in GtR for all funded research. This is normally because the abstract was not required at the time of proposal submission, but may be because it included sensitive information such as personal details.
 
Description We have run three competitions aimed at improving the signal processing in hearing aids. In each competition we supply a very large number of speech-in-noise signals, representing someone listening to a talker in a room but made difficult by there being other background sounds on at the same time, plus software for modelling what a hearing aid will do to the sounds and predicting how many words will be heard. Any research group around the world can enter; the challenge is to improve the number of words or how well they can match what real listeners do. Each competition is designed to be a harder situation than the preceding one.
Exploitation Route They will be useful to hearing-aid developers and also those interested in auditory theory and how hearing problems affect listening.
Sectors Digital/Communication/Information Technologies (including Software),Healthcare
 
Description We organised the first open Machine Learning Challenge for enhancing the processing of hearing aids in 2021. This was part of developing a new research area and growing the number of researchers considering hearing loss and speech technology. Non-academic entrants were from: Music Tribe; Google and Tancent. Academic entrants were: Brno University of Technology; ELO-SPHERES Consortium; University of Oldenburg; University of Sheffield; University Hannover and Shenzhen University. The results were presented at the The Clarity Workshop on Machine Learning Challenges for Hearing Aids (September 2021). We organised the first open Speech Intelligibility Prediction Challenge for people with a hearing loss listening via hearing aid. Ten entrants came from a variety of universities in Taiwan, China, Japan, Germany and UK. The results were presented at the The 2nd Clarity Workshop on Machine Learning Challenges for Hearing Aids (June 2022) We organised the second open Machine Learning Challenge for enhancing the processing of hearing aids in 2022. The second challenge substantially increased the difficulty of the problem by including multiple interferers and head movement during the stimulus. This second challenge was also used as the basis for an ICASSP Signal Processing grand challenge in 2022/23. The ICASSP challenge also included a more ecologically-valid and independent evaluation set. There were 10 teams between these two challenges. The results were presented at the 2022 2nd Clarity Workshop on Machine Learning Challenges for Hearing Aids (December 2022) at at ICASSP. www.claritychallenge.org
First Year Of Impact 2021
Sector Digital/Communication/Information Technologies (including Software),Healthcare
Impact Types Cultural
 
Title Dataset of British English speech recordings for psychoacoustics and speech processing research 
Description The Clarity Speech Corpus is a forty speaker British English speech dataset. The corpus was created for the purpose of running listening tests to gauge speech intelligibility and quality in the Clarity Project, which has the goal of advancing speech signal processing by hearing aids through a series of challenges. The dataset is suitable for machine learning and other uses in speech and hearing technology, acoustics and psychoacoustics. The data comprises recordings of approximately 10,000 sentences drawn from the British National Corpus (BNC) with suitable length, words and grammatical construction for speech intelligibility testing. The collection process involved the selection of a subset of BNC sentences, the recording of these produced by 40 British English speakers, and the processing of these recordings to create individual sentence recordings with associated prompts and metadata.clarity_utterances.v1_2.tar.gz contains all the recordings as .wav files, with the accompanying metadata such as text prompts in clarity_master.json. Further details are given in the readme.Sample_clarity_utterances.zip contains a sample of 10.Please reference the following data paper, which has details on how the corpus was generated: Graetzer, S., Akeroyd, M.A., Barker, J., Cox, T.J., Culling, J.F., Naylor, G., Porter, E. and Muñoz, R.V., 2022. Dataset of British English speech recordings for psychoacoustics and speech processing research: the Clarity Speech Corpus. Data in Brief, p.107951. 
Type Of Material Database/Collection of data 
Year Produced 2022 
Provided To Others? Yes  
Impact This dataset was used to enable the 1st Clarity Enhancement Challenge (CEC1) and will be used for further challenges within the project. CEC1 attracted participation from academic and industry teams from the US, Europe and Asia and resulted in a one day international workshop. 
URL https://salford.figshare.com/articles/dataset/Dataset_of_British_English_speech_recordings_for_psych...
 
Title 1st Clarity Enhance Challenge Software 
Description Software to support the Clarity Enhancement Challenge. Python tools are provided for Mixing scenes to generate the training and development data Running the baseline hearing aid processing using the Oldenburg/Hortech open Master Hearing Aid (openMHA) software platform. Running signals through a python implementation of the Moore, Stone, Baer and Glasberg (MSBG) hearing loss model Estimating intelligibility using the Modified Binaural Short-Time Objective Intelligibility measure (MBSTOI). 
Type Of Technology Software 
Year Produced 2021 
Open Source License? Yes  
Impact This software is being used by all participants in the 1st Clarity Enhancement Challenge, the first of the three rounds of machine learning for hearing aid processing challenges that this project is planning to deliver. The challenge and related software have only just been launched so it is a little early to assess impacts. However, it is already drawing interest internationally both from academic research groups and industry (e.g. Google). This broad participation is being directly enabled by the software we are making available. In particular, for groups who have not previously worked in the area of hearing impairment, the baseline hearing aid processor, hearing loss model and intelligibility prediction software are likely to be essential tools that they will not have previously had access to. 
URL https://zenodo.org/record/4593857
 
Title NOTE - Nottingham Online Testing Environment 
Description NOTE is an online platform for running audio experiments and questionnaires.Its main aims are [A] For audio, to present sounds in a reliable way (and to know just what code is doing the presenting) [B] For questionnaires, allow for various forms of responses including multiple choice and free text, etc. [C] Require active consent before proceeding [D] Tag data with participant codes and keeps all data secure.[E] Be open to participants when told the address, but passed-protected for data access and generally hidden from Google [F] Have a 'live' area of current experiments and a 'development' area for developing & debugging [G] to be stable on all modern browsers and on computers, tablets and phones. This award helped fund its ongoing development. 
Type Of Technology Webtool/Application 
Year Produced 2021 
Impact When finished we intend to make it available to other researchers and we are hopeful it will prove influential. 
URL http://surveys.hearing.nottingham.ac.uk/notedemo1
 
Title claritychallenge/clarity_CC: clarity_CC release v1.2 
Description Software to support the Clarity Enhancement and Prediction Challenges. Python tools are provided for • Mixing scenes to generate the training and development data • Running the baseline hearing aid processing using the Oldenburg/Hoertech open Master Hearing Aid (openMHA) software platform • Running signals through a python implementation of the Moore, Stone, Baer and Glasberg (MSBG) hearing loss model • Estimating intelligibility using the Modified Binaural Short-Time Objective Intelligibility measure (MBSTOI). This version includes • The first release of the code for supporting the recently opened 1st Clarity Prediction Challenge (CPC1) • Code for supporting the (now closed) 1st Clarity Enhancement Challenge (CEC1) 
Type Of Technology Software 
Year Produced 2021 
Impact ... 
URL https://zenodo.org/record/4593856