First version of a synthetic Norwegian child voice

A preliminary test version of a synthetic Norwegian child voice has now been developed. User testing is planned to start in February next year.

Story by: Magne Lunde - 14.12.2011

Today children without speech must use adult synthetic speech in their communication with others. The Child Voice project is working to develop the first synthetic Norwegian child voice and is supported by IT Funk and Blindemisjonen IL.

Adaptation

It is more difficult to develop good child voices, than to develop adult voices. Modern synthetic speech development is based on data from large quantities of recorded text, and a young child is not able to read aloud such large quantities of structured text. A child's voice also has different properties to that of an adult, a child's voice is lighter and more dynamic.

A synthetic child voice is being developed using the adaptation techniquie. A short sound recording of a child's voice was carried out in the autumn working together with NorDubb. The project developer, Linguit, added a master voice to this data, using Lingit's adult synthetic voices and mixing these with the data from the recording. In this way they were able to create a synthetic child voice with very little recorded data.

Small adults

The implementation of the adaptation technique described here requires several weeks work on the computer. A complete adaptation is not yet ready, but an adaptation from a smaller data set has recently been completed and was presented to the project group on December 8th. The adult nature of the data used so far has resulted in a preliminary test version which still has a distinct adult quality, and also obvious technical deficiencies due to the limited amount of data used. The first complete version is expected in the New Year.

Increased text input

Further work remains, when the adaptation will be based on a recording of children's texts, to investigate if this will give the voice a little less maturity. Using their experience with artificial adult voices (Ingrid and Isaac) Lingit plan to make technical adjustments to improve the quality of the voice.

Internal testing

A new beta version of the child's voice will be ready in February when we will conduct internal testing to find out how the speech functions with the tools / software that children with speech disabilities use to communicate with others, and to make the first assessment of the quality of the voice.

Visually impaired and dyslectic children

Haukåsen school, Statped West and Møller-Trøndelag centre will participate in the implementation of user testing in February and March. Visually impaired and dyslexic children will also participate to see if these groups will benefit from using a synthetic child voice to read text.

We welcome feedback and suggestions for the work in this project.
Ring 21 53 80 10 eller send en e-post på adressen info@medialt.no, dersom du har noe du vil si oss.

News archive