As someone who has previously worked in assistive technology and worked with a lot of folks with moderate to severe speech disorders, I was very excited to discover this TED Talk – “Ruba Patel: Synthetic Voices, As Unique as Fingerprints”. It shows how the organization VocalID.org are creating unique voices for recipients, by combining the unique vocal source of the person with the natural speech of a surrogate donor voice.
Human voices work in a way similar to subtractive synthesis. A rich harmonic source, is filtered by the vocal tract (throat, nasal cavity, mouth) to form the unique sounds that make up speech. They discovered that although people with speech disorders couldn’t form words correctly, their source characteristics and vocalisations – pitch, loudness, tempo – could still be manipulated effectively. By combining the natural speech of the donor, with the vocalisations of the recipient, a unique and more natural sounding voice could be produced by the assistive speech device. The donor creates a database of speech which can be selected by the device by a process known as unit selection ( a type of concatenative synthesis).
The results are pretty remarkable and I imagine it would be incredibly exciting for someone to hear their own voice, speaking naturally for the first time (even though it’s produced by a machine). The How It Works section of the vocalID.org page provides an example. The project is ongoing and looking for assistance either from donors or experts who want to help provide unique voices to many of the millions worldwide who don’t currently have a (literal) voice of their own.