What if the speech is already affected ?

Some individuals with neurodegenerative disease may already have speech symptoms at the time of the recording. In that case, the voice cloning process will also replicate these symptoms in the synthesised speech. However, another advantage of the speech technology used by the University of Edinburgh is that the synthetic voice can be easily modified in order to change some of its characteristics such as the speaking rate, the intonation and even the articulation. Therefore the synthetic voice of a patient can be "repaired" by modifying some of these characteristics so that it can generate natural sounding speech while keeping speaker identity.

This process, known as "voice repair", relies on the healthy speech data stored in our voice catalogue. From the voice catalogue, we select a voice donor whose vocal characteristics are most similar to the patient and build a synthetic version of his voice. Then, the features of the patient's synthetic voice that are affected by the symptoms are replaced by those of the donor's. In that way, the symptoms can be removed from the synthesised speech of the patient. In any case, only a few features of the donor's voice are ever used so that the vocal identity of the patient is retained and cannot be confused with that of the donor.

Ideally the voice donor should share the gender, age range and regional accent of the patient since these factors are likely to contribute to the characteristics of the voice. This is why we need to record a large number of healthy voice donors with a variety of age and regional accents.

Voice repair analogy
The patient you are hearing in this video was part of a trial experiment.
However, at the current stage of the project, our digitalised voices are not yet ready to be installed on a communication aid as we are still working on improving the quality of the synthesised speech and solving the compatibility issues.