Voicebox, a Meta AI technology, can do in-context text-to-voice synthesis, speech editing and noise reduction, cross-lingual style transfer, and diversified speech sampling.
- The model now speaks English, French, German, Spanish, Polish, and Portuguese.
- Voicebox now produces more natural-sounding speech even with loud or grammatically wrong input.
- The model now better understands the context of the input text and generates more accurate and relevant speech.
Meta is in development and not yet accessible. The corporation will release the model later.
Voicebox uses include:
Accessibility: Voicebox can read friends’ typed messages aloud.
Virtual assistant: Voicebox gives virtual assistants genuine voices.
Non-player characters (NPCs): Voicebox may give Meta verse NPCs natural voices.
Audio editing: Voicebox can remove noise and correct mispronounced speech.
Diverse speech sampling: Voicebox can simulate real-world speech. The enormous dataset of audio recordings from varied speakers, settings, and speaking styles trains the model. Voicebox generates more lifelike and diverse speech than text-to-speech technologies. Voicebox can produce voice samples for customer support professionals. This would help salespeople sound more natural and engaging with consumers.
Cross-lingual style transfer: Voicebox transfers speech styles between languages. Voicebox can synthesize speech in a similar manner from a sample in one language. This might help with voice translation and multilingual virtual assistants. Voicebox can translate speech styles. The translator could provide more natural-sounding translations.
In-context text-to-speech synthesis: Voicebox generates in-context text-to-speech. Voicebox generates grammatically accurate, natural-sounding voice from a text paragraph. Voicebox is trained on a vast dataset of audio recordings and text excerpts to do this difficult task. Voicebox can produce text-appropriate speech for teachers. This would help teachers develop more interesting and interactive classes.
