Voicebox by Meta

Meta AI researchers have unveiled Voicebox, a cutting-edge generative AI model for speech that sets new standards in the field. Voicebox leverages a novel approach called Flow Matching to learn from raw audio and transcriptions, enabling it to modify any part of a given audio sample. It has outperformed existing models like VALL-E and YourTTS in terms of intelligibility, audio similarity, and processing speed. Voicebox has been trained on 50,000 hours of public domain audiobooks in multiple languages and can perform diverse tasks such as cross-lingual style transfer, noise removal, and content editing. Despite its capabilities, the model or code is not publicly accessible due to potential misuse, though Meta has shared audio samples and research papers detailing its functionalities.