New project to provide synthetic Welsh language voice for those in danger of losing their speech

The Language Technologies Unit (LTU) at Canolfan Bedwyr, Bangor University, has won a grant of £20,000 to develop a programme that will record the voices of people who are in danger of losing their ability to speak because of throat cancer or other health issue, and produce a synthetic naturally sounding version of their own voice. At present this service is only available to English speakers, but this will for the first time allow patients to continue to speak Welsh with their own voice, rather than a generic, synthetic voice that sounds robotic or like someone else. The LTU will work with health authorities and speech therapists to offer the service throughout Wales.   

Delyth Prys of the LTU said: “We demonstrated our Welsh language text-to-speech resources at the Science and Technology Pavilion at the National Eisteddfod in August where we recorded and built a number of new Welsh text-to-speech voices for visitors.  A number of people mentioned that friends and family would need such technology to help them retain their ability to communicate in Welsh with their own voice after its loss”

The LTU have been innovative in speech technologies for Welsh since a number of years. The unit are also developing a Welsh language voice personal digital assistant called Macsen again with the grant aid from the Welsh Government.

Dewi Bryn Jones, LTU’s lead software engineer said “since Welsh is a lesser-resourced language, we’ve had to devise new methods for creating language resources. Thanks to hundreds of contributors from the general public, we have successfully crowdsourced a speech corpus via the Paldaruo ap, the results of which will be used in this project.”

The new service will be called ‘Lleisiwr’, and patients will be able to start recording and build their own personal voices in the New Year. The first versions will speak with their own voice any message that they type at a keyboard. Subsequent versions will allow patients to use their voices on mobile devices and with popular Welsh language keyboard assistants.

See also this other project funded under the same Funding stream.

Publication date: 28 November 2017