You are here

Giving Voice: Voice banking, building and reconstruction

Research Stream
  • Phillipa Rewaj (University of Edinburgh)
  • Shuna Colville (University of Edinburgh)
  • Christophe Veaux (University of Edinburgh)
  • Gergely Bakos (University of Edinburgh)
  • Denise Cranley (University of Edinburgh)
  • Lynda Smyth (University of Edinburgh)
  • Junichi Yamagishi (University of Edinburgh)
  • Prof Simon King (University of Edinburgh)
  • Prof Siddharthan Chandran (University of Edinburgh)
Funding / Sponsorship: 
This study was funded by The Euan MacDonald Centre, The Anne Rowling Regenerative Neurology Clinic, MRC, and MNDA (UK)

The onset of speech impairment and reliance on AAC for people with neurodegenerative diseases can be one of the most difficult symptoms to manage; the inability to accept an alternative voice has been cited as reason for AAC abandonment. The Voicebank Project aims to address this by using a new speech synthesis technique to create personalised synthetic voices for use in high-tech AAC devices.


Participants read aloud between 100-400 sentences designed to identify speaker accent. Recordings are 'banked' and features unique to the individual's voice are synthetically reproduced in a process called 'voice cloning'. When recorded speech is affected by mild to moderate impairment, it is also possible to 'repair' the voice in the synthesis process using donor voices to alter affected features.


Preliminary feedback from fifteen patients has been positive. Participants rated similarity of their synthetic voice to original to voice with an average score of 3.3/5 and all participants expressed a preference for their personalised synthetic voice over a pre-existing generic alternative.


This new speech synthesis technique provides accepted personalised synthetic voices for use in communication aids using minimal speech recording and in the presence of speech impairment, helping to preserve identity.

Level of Session 
Age Group