Gaelic speech‑to‑text technology
More about Gaelic speech-to-text technology
In May 2026, ÈIST launched a free web app that turns spoken Gaelic into written text. It is for anyone who works with Gaelic audio or video: broadcasters, teachers, researchers, podcasters or families with old recordings.
How it works
Upload a video or audio file and the app does the rest. Our speech recognition model listens to the recording and writes out what it hears in Gaelic. The text appears as timed subtitles that play along with the video. You can correct mistakes directly and preview the result before using it. A short training video in Gaelic walks through the process. Its subtitles and translation were made with the app itself.
Behind the scenes, the app uses a version of OpenAI's open Whisper model, which we have trained on hundreds of hours of transcribed Gaelic speech. Before transcribing, it filters out silence and background noise so that only speech is processed. The app and all data processing are hosted in Europe.
Tip: if the app hasn't been used for a while, it may take a minute or two to start up.
How accurate is it?
On clear broadcast speech, such as news and chat shows, the system gets about 90% of words right. That is a strong result for a language with so little training data available. Accuracy is lower for more challenging material:
- Archival recordings: about 82% of words are correct for older recordings from the School of Scottish Studies Archives.
- Children's speech: early results are about 71%, as children's Gaelic differs considerably from adult speech. This is an active area of our work.
- Poor-quality phone audio: results vary with recording quality.
We are gathering more training data in each of these areas, and the models improve as that data grows. Always check the output before publishing it.
Why it matters
Subtitling Gaelic by hand is slow and expensive, so much Gaelic video goes without subtitles. Automatic transcription cuts that work substantially:
- Some BBC staff now use the tool daily, especially for subtitling short online videos.
- Production companies, podcasters and content creators report large time savings. Some are using it to help produce commissioned Gaelic television programmes.
Does this put professional transcribers out of work? Very few people transcribe Gaelic professionally, and there is far more Gaelic audio and video than they could ever get through, in broadcast archives, sound archives and new productions alike. The tool does not replace their expertise, because its output still needs checking by a fluent speaker. What it does is take on the slowest part of the job, so that skilled transcribers can cover more material.
Subtitles also open Gaelic content to people who are hard of hearing, and to learners following along. Transcribed text is searchable, so recordings become easier to find and study. The same technology helps volunteers on Opening the Well / Fosgladh an Tobair transcribe traditional recordings for Tobar an Dualchais, and their corrected transcriptions in turn help train better models.
This is technology built with and for the Gaelic community.
ÈIST is funded by the Scottish Government and Bòrd na Gàidhlig.