Welsh Language Technology Action Plan 2018–2024 — Open Welsh Speech, Voice-Banking and Text-to-Speech Tools

The Welsh Government's Welsh Language Technology Action Plan (2018–2024) funded open Welsh-language speech and language tools, many built by Bangor University, so that Welsh speakers can use digital services in their own language.
Reported outputs: crowdsourced Welsh voice data via Mozilla Common Voice (over 1,285 voices in Bangor's collaboration), used to build the Macsen personal assistant and the Trawsgrifiwr transcription tool, which the National Library of Wales uses to help volunteers transcribe audio; synthetic Welsh voices used by the RNIB to give visually impaired people Welsh text-to-speech; and the Lleisiwr voice-banking project, available through speech therapists so patients losing their voice can keep speaking Welsh.
Reach figures in the 2024 final-report blog: the Welsh interface of Office 365 reached 78,086 users in Welsh-medium schools, and the Cysgliad spelling and grammar package had 12,586 downloads. Bangor built domain-specific machine-translation engines for health and legislation that tend to beat general engines.
Limits: public sources give no speech-recognition accuracy rates or funding amounts, no independent evaluation, and Macsen's ChatGPT-based answers (for example yes/no questions) are still being improved with OpenAI.

Read the full analysis: https://digitalanddata.blog.gov.wales/2024/02/21/celebrating-the-achievements-of-the-welsh-language-technology-action-plan-2018-2024/