Children's voices

AssistiveWare and Acapela Group collaborated to create the first genuine children's Text to Speech voices.

The problem

For decades, Augmentative and Alternative Communication (AAC) solutions have given a voice to children who cannot speak. However, until recently, children with communication difficulties only had access to adult Text to Speech (TTS) voices or artificially emulated children’s voices that sounded very robotic and not genuine.

Listen to the artificially emulated children’s voices: Kenny Kenny Nelly Nelly

Rising to the challenge

Collaboration

Together with Acapela Group, one of the leading Text to Speech companies, we picked up the challenge to develop genuine Text to Speech children's voices. After all, a child using an AAC system to communicate deserves to sound just like their peers, not like a grown man or a woman on helium. In addition, we also have bilingual American Spanish-English children's voices available, so bilingual kids can finally speak with the same voice in both Spanish and English.

How does it work?

Text to Speech voices are synthesized using a large database of actual recordings from a voice talent. When text is spoken, each word is composed of different sound combinations recorded in the studio from a long script. This script is designed so that as many sounds and sound combinations as possible are recorded in a minimal amount of time. Still, the recording process requires several days in the studio. During the speech synthesis process, any word can be created by combining sounds together. To learn more about this process, watch the short video, “Text to Speech, how does it work?”, below. Creating the first two voices took more than 2,500 hours of research, recording, processing and development.

Voice talents

One of the most difficult aspects of creating children's voices was finding voice talents sounding young enough, yet were good readers and had the stamina to spend hours in the studio for several days. Just identifying the right talents and reviewing voice samples took a number of months. Watch "The Making of the New British Children's Voices Harry and Rosie" and “The Making of the New American Children’s Voices Josh and Ella” below to see the voice talents in action.
 

New voices

Once the recordings are done, the new voices are still far from ready. All the raw material needs to be processed and then repeatedly tested to make sure words are pronounced correctly and sound natural. The difference between the new, real voices and the old, simulated voices is striking. Listen to the new children’s voices in American English, British English, Australian English, German, Swedish and bilingual American Spanish-English. Note that for the German and the bilingual American Spanish-English voices we additionally collaborated with Tobii Technology.

Designed for AAC

The children's voices are also the first voices specifically designed to support AAC users. The specially adapted recording script, as well as the testing process, focused on optimally supporting vocabulary used in communication. What is more, a large set of expressions and sounds were recorded by the voice talents in support of the ExpressivePower™ feature of Proloquo2Go.

Life-changing

With these new genuine children's voices, children with communication difficulties can finally speak their thoughts and needs with a voice that resembles their own. They are one step closer to communicating as any other child. “These new voices will make such a difference to them and to their families - in particular, in terms of enabling them to be included, to have normal relationships with their peers and enjoy everything that so many other children take for granted,” said Anna Reeves, England’s National AAC coordinator. To understand the importance of this project, watch the video, “Giving Children a Voice of Their Own”, below.

Related blog post