We’re closer than ever to AI “immortality” thanks to Meta’s new voice technology

Rapid developments in the field of artificial intelligence are continuing and now meta Take an important step towards it “immortality” Through AI, with its presentation sound box, sophisticated one Text-to-speech form. This new AI aspires to do what ChatGPT did for text and Dall-E did for images, Produce incredibly realistic audio extracts from text. In theory, with Voicebox, the voice of anyone famous or otherwise could continue to be used even after their death.

According to the Meta, Voicebox acts as a “non-self-regressive flow-matching” model that is trained for speech completion, given voice and text context. The AI ​​was trained with more than 50,000 hours of audio data from publicly available audiobooks in various languages, including English, French, Spanish, German, Polish and Portuguese. This rich and diverse set of data allows Voicebox to generate speech that sounds like the real thing, regardless of language. In fact, impressively, the error rate for the synthesized speech Voicebox produced was only 1%, which is a huge improvement over the 45 to 70% attenuation seen in existing text-to-speech (TTS) models.

Introducing Voicebox, a new innovative speech system based on Flow Matching, a new method proposed by Meta AI. It can synthesize speech across six languages, perform noise removal, content editing, voice pattern transposition and more.

More details about this work and examples

—MetaAI (MetaAI) June 16, 2023

The innovative training process included, among other things, the model’s ability to predict parts of speech based on relevant context and text. As described by the Meta researchers, this allows the model to reproduce specific parts of the recording, without completely recreating the input. In addition, Voicebox can also edit audio clips, filter out noise, and even correct mispronunciation, just like using photo editing software to enhance photos. Overall, the Meta’s AI is said to far outperform the latest text-to-speech models in terms of intelligibility and “audio similarity” and runs up to 20 times faster.

but, Meta has decided not to launch an app for now or publicly sourced code, indicating potential abuse risks. You can listen to examples and learn more Click here.

Follow Unboxholics.com at www.unboxholics.com
google news
To be the first to know the latest news about technology, video games, movies and series. Follow Unboxholics.com at www.unboxholics.com FacebookAnd TwitterAnd
InstagramAnd Spotify
And Tik Tok.