OpenX-Sound is an audio-augmented robot manipulation dataset in RLDS / TFDS format.
Each episode keeps the original trajectory structure and adds an episode-level audio field. Existing code that reads steps, observations, actions, language, and metadata can continue using the same paths, while audio-aware pipelines can additionally consume the synchronized waveform.
The dataset is organized as standard TFDS directories:
//… See the full description on the dataset page: https://huggingface.co/datasets/biubiu2/OpenX-Sound.
