Stable Audio Open

·2024.07.19 16:26

Key point

Stable Audio Open, an open-weight text-to-audio model trained on Creative Commons data, has been released.

Details

Most text-to-audio models are operated as closed source, making it difficult for researchers and artists to build upon them. To address this, a new open-weight model trained using Creative Commons data, Stable Audio Open, has emerged.

Stable Audio Open has proven performance on par with state-of-the-art (SOTA) across various evaluation metrics. In particular, the FDopenl3 results show that this model has the potential to synthesize high-quality 44.1kHz stereo sound.

generative-ai open-weights text-to-audio stable-audio-open audio-synthesis

Read originalShare on Threads

This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.

Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.