[AUDITORY] [CfP] Room Acoustics and Speaker Distance Estimation Challenge at GenDA 2025 (ICASSP Satellite Workshop)

["Kim, Minje" <minje@xxxxxxxxxxxx>]

(Apologies for cross-posting)
 
Dear Colleagues,

We are delighted to invite you to take part in the Room Acoustics and Speaker Distance Estimation Challenge, hosted as part of the Generative Data Augmentation (GenDA) Workshop at ICASSP 2025 in Hyderabad, India. This challenge spotlights the potential of generative methods to produce synthetic room impulse responses (RIRs) as a data augmentation tool for a particular downstream task, speaker distance estimation (SDE) in real-world environments.

Challenge Overview

Task 1: Augmenting RIR Data
	•	Develop an RIR generation system that can produce realistic RIRs at new source-receiver positions based on just a handful of enrollment RIRs.
	•	Generate RIRs for 20 different rooms, each with limited initial RIR recordings.
	•	Submissions will be evaluated on how accurately the generated RIRs match key acoustic parameters (T60, DRR, EDF).

Task 2: Improving Speaker Distance Estimation
	•	Use your augmented RIR dataset from Task 1 to fine-tune a pre-trained baseline SDE model we provide (rather than developing a new SDE system from scratch).
	•	Evaluate on a test set of 480 reverberant speech audio samples to measure distance estimation performance.
	•	No external RIR data is allowed during fine-tuning, ensuring the focus remains on your system’s data-generation quality.

We encourage all participants to complete both tasks so we can fully assess the impact of your generated RIRs to augment the training datatset for the downstream task.

Why Participate?
	1.	Benchmark Your RIR Generation Methods: Compare your approach against our open-source baseline.
	2.	Advance Speaker Distance Estimation: Demonstrate how synthetic data can help refine state-of-the-art models in challenging acoustic scenarios.
	3.	Shape Future Directions: Contribute insights on generative acoustics modeling and share cutting-edge techniques in the growing field of data augmentation.

Important Dates
	•	December 23, 2024: Submission system opens
	•	March 12, 2025: Deadline to submit your two-page summary and challenge files
	•	Early April 2025: Results announced (prior to the main ICASSP 2025 conference dates)

Submission & Participation Details
	1.	Create a new submission on ICASSP 2025 CMT under
“Satellite Workshop: Generative Data Augmentation for Real-World Signal Processing Applications”
with the subject area “Challenge: Room Acoustics and Speaker Distance Estimation.”
	2.	Upload a two-page report (up to four pages allowed) describing:
		•	Your RIR generation system and training data
		•	The subset of enrollment data used
		•	Fine-tuning protocol of the baseline SDE model
		•	(Optional) New SDE architecture details if necessary
	3.	Include all your generated files (.wav for Task 1 and a .csv file with distance estimates for Task 2) as supplementary material.

All technical information (baseline code, data, instructions) is provided in our challenge website (https://sites.google.com/view/genda2025/room). Further technical details can be found in our paper (https://minjekim.com/wp-content/uploads/icasspw2025_jlin.pdf).

Contact
	•	Jackie Lin (jackiel4@xxxxxxxxxxxx)
	•	Minje Kim (minje@xxxxxxxxxxxx)

Join us at GenDA 2025 to explore how cutting-edge generative techniques can revolutionize acoustic modeling and speaker distance estimation. We look forward to your innovative submissions and seeing you in Hyderabad!

Best Regards,
GenDA 2025 Organizing Committee
- Minje Kim, University of Illinois at Urbana-Champaign / Amazon Lab126
- Dinesh Manocha, University of Maryland
- John Hershey, Google Research
- Trausti Kristjansson, Amazon Lab126
- Jaesung Bae (UIUC; Task Captain of the Zero-Shot TTS and PSE Challenge)
- Jackie Lin (UIUC; Task Captain of the Room Acoustics and Speaker
Distance Estimation Challenge)