At its annual Re-Mars conference in Las Vegas today, Rohit Prasad, Amazon’s senior vice president and principal scientist for Alexa, announced new and upcoming features for the company’s smart assistant. The most prominent twist of the bunch was a potential new feature that can synthesize short audio clips into long speech.
In the scenario presented in the incident, the voice of a deceased loved one (in this case a grandmother) is used to read a bedtime story to the grandchildren. Prasad notes that, using the new technology, the company has been able to achieve some very impressive audio output using just one minute of speech.
“It required inventions where we had to learn to produce high quality sound with hours of recording in less than a minute versus studio recording,” the executive notes. “The way we’ve done it is framing the problem as a voice conversion task, not a speech generation path. We are suspiciously living in the golden age of AI, where we have dreams and science.”
Details are few at the moment. There’s no timeline or further specifics, but — at the very least — it’s news that will invite all manner of scrutiny on potential applications, from something as simple or heartwarming as reading to a child. beyond. Wizard of Oz,