Val Kilmer has created his own audio deepfake, thanks to the wonderful advancements in technology. An audio deepfake is created when artificial intelligence is used to create content that is meant to deceive. It has often been used to create a fake voice to exploit others, by allowing a person to pretend to be someone else. In this case however, the audio deepfake will be used to accomplish something good. Kilmer can no longer use his natural voice, but this technology will enable him to communicate with those around him again using a voice model.
Val Kilmer lost his voice to cancer, and had it returned to him by British AI Company Sonantic. The sixty-two year old actor is best known for his role in movies such as: Top Gun, The Door, True Romance, and Tombstone. Kilmer was first hospitalized in 2015. Kilmer originally denied cancer rumors, but in 2017 it was revealed that he had undergone a two-year battle with throat cancer. Kilmer had two tracheotomies and underwent chemotherapy. His voice was left ravaged by the cancer; he could only talk in a rasp and would quickly fall short of breath. That was when Sonantic stepped in, and created a model of Kilmer’s voice. Kilmer will be able to use his new voice both personally and professionally.
Software firm, Sonantic, is well known for creating realistic sounding voices for video games. Sonantic creates their models by having actors read hours’ worth of scripts. Then the audio is ingested into the company’s voice engine to create a model that can basically say anything. The hard part of creating a custom voice model for Kilmer was that there wasn’t much high-quality audio with minimal background noise to use from his film career.
Sonantic, using an algorithm, attempted to create a clear, clean audio model. They removed as much background noise as possible, but ended up with a model with only one-tenth of what they would need. The team were not happy with the results, so they created new algorithms that would require less data. In the end they created over forty models of Kilmer’s voice. The team picked the best model to generate Kilmer’s voice from.
Kilmer will be able to use a desktop application to type what he wishes to say and the model will express his ideas through a voice that sounds just like the voice he lost to cancer. He can even adjust the pitch and pacing of his speech to better convey his thoughts.
Artificial voices have come a long way since the creation of Siri and Alexa. Voices are becoming more realistic and less robotic. It is becoming harder to tell real voices from artificial ones, because artificial voices can mimic tone and expression just like humans can. Sonantic audio engineers only require three hours of audio to recreate a voice, and they can do this all in a span of only twenty-four hours. But new technological advancements always come with legal, ethical, and economic concerns. Audio deepfakes have been used for foul deeds in the past. Yet, an audio deepfake also allowed Kilmer’s voice to be returned to him, and gave him the opportunity to communicate and work again.