Meta Unveils New Audio AI Model

Meta Platforms on Tuesday unveiled a new audio transcription model, Muse Voice Transcribe, that CEO Mark Zuckerberg said can transcribe speech to text and segment audio based on who is speaking.

According to Zuckerberg’s announcement in a Threads post, the model was trained on more than 70 languages and can handle when speakers switch between two languages mid-sentence as well as hour-long sessions with more than 20 speakers. The model also powers voice input prompts in Muse Code, Meta’s coding model.

Zuckerberg announced Monday that Muse Code is out of beta and is now available through a tiered paid subscription model, with power users paying up to $50 a month for higher usage limits.

The rollout comes as Meta has released new AI models in recent months as it seeks to close the gap with rivals OpenAI and Anthropic. In August, the company released its latest model, Muse Spark 1.2, and said it would open-source a smaller model called Muse Glimmer. Meta also plans to release its consumer AI agent, internally known as Hatch, in the coming weeks.