Skip to content

Dataset · treble-technologies

Room acoustics simulations for speech recognition

treble-technologies/Treble10-RIR

Room impulse responses at 32 kHz for training ASR models to operate in echo conditions.

What's inside

The dataset contains between 1,000 and 10,000 examples (audio, images, and English text) with high-precision acoustic simulations of 10 furnished rooms. The sample includes two bathrooms, two bedrooms, two living rooms with a corridor, two living rooms without a corridor, and two meeting rooms with volumes from 14 to 46 cubic meters. Their reverberation time ranges from 0.17 to 0.84 seconds, and the audio sampling rate is 32 kHz.

What you can do with it

  • Augmenting clean voice recordings to improve ASR model robustness against echo and reverberation
  • Training neural networks for speech dereverberation from room acoustic distortions
  • Testing voice assistants across different room types
Languages
English
Size
1K–10K rows
Downloads this month
1.2K
Likes
34

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("treble-technologies/Treble10-RIR")