guilinhu/libri_conversation

Libri Conversation FLAC Dataset

This dataset accompanies the paper:

Proactive Hearing Assistants that Isolate Egocentric Conversations
Hu et al., 2025

It contains ~234 hours of conversational-style audio derived from LibriSpeech-like sources, processed into 60-second multispeaker segments under different experimental conditions:

• libri_leaving — scenarios where one speaker intermittently leaves the conversation
• libri_multi — 3-speaker conversational segments

All audio is stored in FLAC format. Metadata files (JSON) are preserved exactly as in the original directory structure.

---

Dataset Structure

The dataset is organized into two main components:

libri_leaving/
  train/
  val/
  test/

libri_multi/
  train/
  val/
  test/

Because of Hugging Face API request limits, the dataset is packaged into .tar archives.
Each archive mirrors the original folder structure:

libri_leaving_train.tar
libri_leaving_val.tar
libri_leaving_test.tar

libri_multi_train.tar
libri_multi_val.tar
libri_multi_test.tar

---

Citation

If you use this dataset, please cite the following paper:

@inproceedings{hu2025proactive,
  title={Proactive Hearing Assistants that Isolate Egocentric Conversations},
  author={Hu, Guilin and Itani, Malek and Chen, Tuochao and Gollakota, Shyamnath},
  booktitle={Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing},
  pages={25377--25394},
  year={2025}
}