EchoVista: A Multimodal Room Acoustic Dataset

EchoVista is a multimodal room-acoustic dataset for research on room impulse response (RIR) modeling, acoustic scene understanding, and audio-visual room acoustics. Each sample links a room/scene image, a measured RIR signal, and geometric metadata describing the microphone, loudspeaker, and room dimensions. We also provide an additional acoustic text description and a reverberant/noisy speech example for each sample; these auxiliary resources may also be generated by users if needed. The dataset contains 691 scene-level samples from 49 room IDs. Images are provided as RGB JPG files, RIR signals are provided as mono WAV files at 22050 Hz with a duration of 5.94 seconds, and geometry metadata is provided in metadata.csv. The dataset will be made publicly available after the paper is accepted.