‘Sophia’ the Robot sings through the barriers of technology by launching CereProc’s singing TTS synthesiser on ‘The Tonight Show’
We are proud to announce that CereProc has created their own singing TTS synthesiser, which premiered with ‘Sophia’ on ‘The Tonight Show’ on the 21st of November. Together with Hanson Robotics, we created Sophia’s characterful speaking voice and now CereProc has given her the ability to sing. ‘Sophia’ using CereProc’s singing synthesiser performed ‘Say Something’ by Christina Aguilera - the first robot-human duet. CereProc’s technology not only reflects the character of ‘Sophia’ but our synthesiser gives any individual the ability to sing with only a small amount of data.
Left to right: CereProc founders: Christopher Pidcock, Paul Welham, and Matthew Aylett
The project was started after receiving an email from David Hanson asking how difficult it would be for CereProc to get ‘Sophia’ to sing. Our lead engineer on the project, Chris Buchanan took on the challenge of adapting the CereVoice Deep Neural Network engine to produce singing instead of speaking. The TTS singing technology developed by CereProc involved re-building the ‘Sophia’ voice using Deep Learning which increased the control in her voice. To reproduce the expression, timbre and other characteristics associated with singing, CereProc trained a new Deep Neural Network (DNN) system and built a database specifically designed for singing synthesis.
Chris Pidcock, CTO CereProc commented that the most challenging part of the project was “to capture the steady-state vowel phonation whilst retaining naturalness and embedding the singing into computer -readable characters”
By taking on challenging projects CereProc has pushed its technology to the next level by creating the world’s first voice cloning TTS synthesiser. Not only does it make the capability to sing accessible to all individuals, but this technology has the potential to change the music industry.
Matthew Aylett CSO, CereProc, stated: “There isn’t a ceiling on the potential quality of the voice in singing TTS. This sort of technology could be used both as in instrument itself, a training aid for musicians, accompaniment for live performance and as a replacement for a human vocalist altogether”
CereProc is in the process of commercialising the singing system and expect to release a beta version for testing in 2019.
Paul Welham CEO, CereProc, concluded: “As an independent Scottish company, CereProc are proud of the innovation that we have achieved and our ability to disrupt the status-quo in speech technology. Hanson Robotics is a key partner of CereProc, together we believe strongly in continuous innovation. CereProc is sure to be partnering Hanson in the future, developing new expressive TTS voices with even more novel functionality”