Daily AI Newsletter: Meta AI unveiled Voicebox, a breakthrough generative AI for speech

Meta AI unveiled Voicebox, a breakthrough generative AI for speech

Adblock In Real Life and An embedding for all human beliefs and religions

By Saeed EzzatiPublished Jun 19, 2023, 11:01 PM PDTUpdated Sep 2, 2026, 6:33 PM PDT5 min read

Meta AI unveiled Voicebox, a breakthrough generative AI for speech

In today’s email:

🔊 Meta introduced: a breakthrough generative AI for speech
☯️ Tenet, A community project to embed all human beliefs.
📚 Tutorial: Creating an Autonomous HR Assistant with ChatGPT and LangChain
🛠 Various AI-related tools and platforms, including FableForge, Vid2Avatar, Chat Thing, Framer, AutoPod, Love Stories, Blush, Mano, MagicBrush, QR Craft, and more.

Highlights💡

Meta Introducing Voicebox: The First Generative AI Model for Speech [Link]

to generalize across tasks with state-of-the-art performance

Meta AI researchers have unveiled Voicebox, a breakthrough generative AI for speech that excels at tasks it was not specifically trained for, setting new performance standards in the field. Meta claims the model is so good that they’re too scared to make it publicly available. Unlike other models that need specific task training and carefully prepared data, Voicebox learns from raw audio and transcription, offering high-quality audio clips across six languages and capabilities in noise removal, content editing, style conversion, and diverse sample generation. The model outperforms the current leading English model, VALL-E and YourTTS, in zero-shot text-to-speech and cross-lingual style transfer regarding word error rates and audio similarity while also being up to 20 times faster.

Voicebox is built upon the Flow Matching model, a non-autoregressive generative model, and trained with more than 50,000 hours of recorded speech and transcripts, making it capable of versatile applications such as in-context text-to-speech synthesis, cross-lingual style transfer, speech denoising and editing, and diverse speech sampling. However, due to potential misuse risks, Meta is not releasing the Voicebox model or code publicly but will share audio samples and a research paper outlining their approach and results. The team believes Voicebox signifies a significant step forward in generative AI research and looks forward to future applications and developments in the audio domain.

Alexandria Index released Project Tenet, a community project to embed all human beliefs.[Link][GitHub]

Open-source embeddings for 10+ major religious texts (over 15m tokens, 20 billion vector dims)

This all started with a question: What would we find if we embedded all religions? What parts of human belief are the same? Which are different?

So far, we've seen fascinating results. No conclusions just yet, but feel free to see for yourself.

More than insights, Tenet has clear application as a tool.

To demonstrate this, we've built http://tensor.church, where you can ask questions and search for wisdom across all religions. Hear from Marcus Aurelius or Moses, Krishna, or Confucius. (ChatGPT plugin coming soon)