Voicebox

lastmod 2023-06-21created 2023-06-211 edits

Meta releases Voicebox, a generative AI model for speech

Features

• Like image/text generative AI, generates high-quality audio clips in a variety of styles
• Can also edit a given audio sample
• Not only synthesizes speech in 6 languages, but also does noise removal, content editing, style conversion, and diverse sample generation
• Previous generative AI for speech required carefully prepared training data for each task. Voicebox uses a new method to train on raw audio and transcription data
• Trained on more than 50,000 hours of public domain audiobooks in English, French, Spanish, German, Polish, and Portuguese

See also

• Meta (facebook)

Favorite site

• Introducing Voicebox: The first generative AI model for speech to generalize across tasks with state-of-the-art performance