Principal Researcher: Prof Chenghua Lin, University of Manchester (UK)
Scientific summary:
The primary purpose of this investigation is to address the communication barriers experienced by individuals with ataxia by developing an efficient and personalised automatic speech recognition (ASR) framework tailored to their unique speech patterns. By enabling accurate recognition of ataxic speech using minimal training data, the project aims to support improved communication, social participation, and quality of life for people living with ataxia.
Despite major advances in ASR for the general population, state-of-the-art systems (e.g. Whisper and Parakeet) perform poorly on ataxic speech due to two fundamental challenges: (i) high inter- and intra-speaker variability, and (ii) the scarcity of publicly available ataxic speech data. These limitations hinder effective model training, personalisation, and evaluation for this population, leaving people with ataxic speech largely underserved by current technologies. To address these challenges, this project aims to develop innovative and inclusive approaches that deliver high performance with minimal data requirements. The objectives include:
- Creating a new ataxia speech dataset using speech data collection processes, tools, and data provided by project partners;
- Develop methods to generate diverse synthetic speech data for Ataxia, augmenting the real-world speech data collected;
- Explore contrastive learning based techniques to mitigate data scarcity and enable efficient training;
- Propose novel parameter-efficient fine-tuning methods for personalised ASR tailored to ataxic speech.
This project aligns closely with the goal of enhancing ataxia patients’ quality of life through cutting-edge AI communication technologies. By working with the national ataxia centres and through lived experience advice, the project aims to ensure that the tools developed meet the practical needs of the ataxia community. We intend to work with Ataxia UK’s social outreach team to promote the project in order to attract more patient volunteers to contribute to the dataset. By leveraging NVIDIA’s platform and research capabilities, it aims to develop transformative, inclusive, and scalable AI solutions that improve accessibility using state-of-the-art techniques.
Lay summary:
Many people with ataxia experience difficulties with speech, which can make everyday communication challenging. Slurred or unclear speech can affect conversations with family and friends, reduce confidence, and limit social interaction and independence. While voice-controlled technologies are now widely used in daily life, they often struggle to understand speech affected by ataxia, meaning that many people are unable to benefit from these tools.
This project, which will be a collaboration between the University of Manchester, University College London, the University of Sheffield, the University of Strathclyde and NVIDIA, aims to improve this situation by developing speech recognition technology that is better suited to the unique ways people with ataxia speak. Because ataxic speech can vary greatly between individuals and change over time, a “one-size-fits-all” approach does not work well. Instead, this research will focus on creating speech recognition systems that can be personalised to an individual’s speech using only a small amount of recorded speech.
To do this, the project will work with people with ataxia to collect speech data in an ethical and supportive way. The collected data will help researchers develop and test new approaches that make speech recognition more accurate and more inclusive. In the longer term, this work aims to support the development of communication tools that help people with ataxia express themselves more easily, take part in conversations, and improve their quality of life.
This project is due to begin on 15 May 2026.
For more support or information please contact:
Ataxia UK, 12 Broadbent Close, London, N6 5JW
Website: www.ataxia.org.uk.
Helpline: 0800 995 6037 Tel: +44 (0)20 7582 1444
Email: helpline@ataxia.org.uk.
