December 2024 Summaries

OCTAVE (Omni-Capable Text and Voice Engine) is a next-generation speech-language model that combines the capabilities of various systems, including OpenAI's Voice Engine, Elevenlab's TTS Voice Design, and Google Deepmind's NotebookLM. It can generate voices and personalities from prompts, clone voices from recordings in one step, interact with users in real-time, and create multiple interacting characters. OCTAVE maintains comparable performance to similar-sized LLMs on language understanding tasks while ensuring its intelligence is coherent and nuanced. The model is currently being evaluated for safety and effectiveness by trusted partners before broader availability is rolled out in the coming months.

Voice Control is a novel interpretability-based method that brings precise control to AI voice customization, allowing developers to create one-of-a-kind voices without the risks of voice cloning. The tool gives users control over 10 voice dimensions, including attributes such as gender, assertiveness, and confidence, which can be adjusted relative to the base voice's characteristics using intuitive sliders. This approach enables continuous adjustments along these dimensions, making voice modifications reproducible across sessions and allowing for precise control. Voice Control is released in beta and aims to transform how custom AI voices are created through interpretable, continuous controls, providing developers with a new artistic tool to design their own unique voice in seconds.