22/10/2018 - JFK UNSILENCED PROJECT
JFK’s words in July 1963 negotiated the resolution to end the Cold War. His words provided federal support for the growing civil rights movement. The course of history was changed forever on November 22, 1963, as JFK made his fateful trip through downtown Dallas to deliver his speech at the Dallas Trade Mart and was shot dead. The assassination sent shock waves throughout the world and his speech was never to be heard, until now!
CereProc were approached by Rothco, a PR company in Ireland whose objective was to recreate an authentic JFK’s voice based on previous speeches that he had given so that he could finally deliver the speech he was meant to deliver in Dallas 55 years ago. The CereProc team over the duration of 8 weeks analysed recordings of 831 speeches and built the voice by splitting them into 116,777 small phonetic units.
To build the voice, Chris Pidcock co-founder of CereProc and project manager of “JFK Unsilenced project” states:
"We built parametric Deep Neural Network (DNN) and Unit selection voices (US), we built voices that could be blended into both techniques…the unit selection voices reproduce the character of the speaker and the DNN voices help guide the prosody modelling which is the intonation”
The biggest challenges that faced the project were capturing JFK’s individual speaking style and irregularity of the quality data which was recorded at different times, dates and on different types of equipment. Blending the segments of speech together and smoothing out the data was very difficult. To overcome this challenge, our team used the DNN modelling framework, de-noised the data to build the voice and reapplied some noise and reverberation to the final speech to make it sound more authentically 60’s.
Our partner in the project, Rothco, was delighted with the results:
“CereProc consistently found solutions to the toughest of challenges and were a pleasure to deal with through the long and arduous process.” (Al Byrnes, Executive Producer, Rothco 14 September 2018).
Together we created the world’s first AI audio speech made completely out of data. Not only was the project published on the front page of “The Times”
achieving a total editorial reach of 1 billion, but it also allowed his speech to be heard for the first time, touching on topics such as freedom, power and the dangers of populism which seems more relevant today than ever before.
“In a world of complex and continuing problems; in a world full of frustrations and irritations, America’s leadership must be guided by the lights of learning and reason, or else those who confuse rhetoric with reality, and the plausible with the possible, will gain the popular ascendancy with their seemingly swift and simple solutions to every world problem” (JFK Unsilenced)
Listen to the full speech here!
This project demonstrates how AI impacts on the lives of individuals and how, through our real characterful voices, it can touch human emotion. We were delighted to hear from Lisa Brown, the daughter of the speech-writer who wrote JFK’s final speech.
She supplied the following statement:
“as I heard Kennedy speaking my father’s words, I felt an extraordinary emotional reconnection both to the man I so loved and admired, and to Jack Kennedy, who was his friend. The feeling was like a shock wave. Strangely, I felt that I was hearing my father’s voice again through JFK’s voice. The power of the human voice and the spoken word is astonishing” (Lisa Brown).
The original JFK speech sent to us by Lisa Brown
The technology developed in this project can be seen in our new version of the system. Chris Pidcock states that, our parametric voices are now using the DNN technique to do their synthesis so their modelling of the prosody intonation is similar to the system implemented in the JFK voice. Furthermore, it pushed CereVoiceME to reduce the amount of data needed to clone a voice, resulting in the ability to create a better sounding voice with less data and the possibility to repair the recordings of our customers suffering from a degenerative illness.
To summarise, we at CereProc are very proud to be part of history and the way this project has impacted on advancing the development of TTS technology and the lives of individuals.