Researchers Present the First Working Speech Recognition System for Scottish Gaelic

Screenshot of video with audio transcription created by the ASR system

 

An interdisciplinary University of Edinburgh research team led by Dr Will Lamb has developed an automatic speech recognition (ASR) system for Scottish Gaelic. 

The system has a wide range of applications, from providing television subtitles and controlling smart devices, to supporting Gaelic-medium schoolchildren with dyslexia. The research team hopes that the system will provide new opportunities for the endangered language by widening access to it. 

The research team have most recently reported an accuracy level of 71.8% for predicted transcription, and have shared subtitled videos to showcase the system’s current performance. 

The ASR system uses neural network artificial intelligence developed by Quorate Technology Ltd in Edinburgh, and has been developed using more than 60 hours of speech training data. The team have begun working with collaborators to provide initial transcriptions of interviews and traditional stories recorded from Gaelic speakers, and hope to eventually make it freely available online.   

Led by Dr Lamb as Principal Investigator (School of Literatures, Languages and Cultures), the team also involves co-investigator Dr Beatrice Alex (School of Literatures, Languages and Cultures; School of Informatics), specialists, advisors and research assistants. 

The project is funded by the Data-Driven innovation Initiative as part of their ‘Building Back Better’ open funding call, and received generous support from Soillse, the National Research Network for the Maintenance and Revitalisation of Gaelic Language and Culture.