About ÈIST

ÈIST [eːʃtʲ] (‘ayshch’) is short for Ecosystem for Interactive Speech Technologies, and means ‘listen’ in Gaelic. The project is funded by the Scottish Government and Bòrd na Gàidhlig, with key partners including the BBC ALBA, MG ALBA, the University of Glasgow and Tobar an Dualchais / Kist o Riches. It aims to support the revitalisation of the language through cutting-edge interactive technologies, including speech recognition.

Since launching in 2023, ÈIST has focussed on developing accurate speech-to-text for Gaelic. The initial aim was to produce a system that could generate Gaelic-medium subtitles for BBC ALBA and Radio nan Gàidheal. In a 2025 research paper, the team reports achieving nearly 90% accuracy for chat shows, news and current affairs programmes. Now the team is expanding the technology, making it suitable for more diverse contexts, ranging from old fieldwork recordings to Gaelic medium classrooms.

In January 2026, the team built a user interface and a API (Application Programming Interface) to provide access to its Gaelic speech recognition model. For the first time, developers and researchers were able to embed the technology into applications ranging from educational software to digital assistants. Our first model release was based upon OpenAI’s open-weight Whisper model and is now available here. We are currently working on more advanced models, including ones capable of real-time speech recognition.

Bridging Linguistic Gaps

Currently, Gaelic speech recognition systems struggle to transcribe younger speakers, whose speech patterns often differ from those of heritage speakers. ÈIST addresses this with a dedicated subproject, Recognising Children’s Speech, funded by Bòrd na Gàidhlig, which is gathering data from Gaelic Medium Education schools and units across Scotland. This new data will ensure that the speech of young learners is represented accurately.

The implications of this work are profound. Enhanced speech recognition can transform educational tools and provide more effective support for literacy and learning among children. In general, ASR can improve accessibility for Gaelic speakers of all ages.

A Community-Powered Initiative

Central to ÈIST’s mission is Opening the Well, a pioneering crowdsourcing platform that launched at the University of Glasgow in December 2025. This initiative is mobilising the Gaelic-speaking community worldwide to transcribe audio recordings of traditional narratives and oral history held on the Tobar an Dualchais portal, such as those originally from the School of Scottish Studies Archives (University of Edinburgh) and the National Trust for Scotland’s Canna House Archive. It takes its cue from Ireland’s successful Meitheal Dúchas crowdsourcing project. These transcriptions won’t just make Scottish heritage more accessible — they will also provide vital training data to enhance the accuracy and versatility of future speech recognition models.

Risks, Rewards and Responsibility

The team behind ÈIST is keenly aware that speech and language technologies are not neutral tools. Poorly trained models can misrepresent language, distort culture or reinforce social bias. That is why ÈIST places community involvement at the heart of its design process.

The researchers also promote best practices for ethical AI use in revitalisation: curate transparent training data, return outputs to the community (e.g., through the Digital Archive of Scottish Gaelic), and avoid relying on ‘big tech’ to solve our problems. It is a values-led approach as much as a technical one.