Meta announced the launch of a new project called “NotebookLlama,” an open-source version of interactive episodic podcasts using artificial intelligence.

The project relies on Meta’s AI models, known as “Llama,” to implement a large portion of the processing needs.

NotebookLlama can generate audio summaries in a talk style that resembles podcast episodes based on text files uploaded to it.

Initially, NotebookLlama converts text into audio files, such as news articles or publications, and then adds “dramatic touches” and interactions that vitalize text before sending it to open forms to convert text into speech.

In Meta, the researchers noted that “the text-to-speech model is the main determinant of the nature of the sound”, as it is possible to take another approach to generate podcasts by having two AI agents discuss the topic and develop the structure of the dialogue, while one model is currently used to write the blueprint.

Source: Qatar News Agency