The 2025 LRAC Challenge
Table of Contents
Introduction
Call for Participation
We invite you to participate in the 2025 Low-Resource Audio Codec (LRAC) Challenge, an exciting opportunity to advance the state-of-the-art in neural audio coding for resource-constrained devices. This inaugural edition of the challenge will focus specifically on low-resource neural speech codecs designed to operate under everyday noise and reverberation conditions, while meeting strict constraints on computational complexity, latency, and bitrate.
See the full Call for Participation here: LRAC-Challenge-Call-for-Participation.pdf.
Motivation
The growing demand for speech interfaces across a wide range of resource-constrained environments—including embedded systems, mobile devices, and general-purpose computing platforms with limited processing or power budgets— requires speech codecs for real-time communication that are efficient in terms of compute, bitrate, and latency, while also delivering high-quality speech under real-world conditions. Although recent neural coding methods excel at optimizing for these constraints individually, addressing all three simultaneously remains an open challenge.
Aims
The first edition of the LRAC Challenge is aimed at reducing this gap and advancing the state-of-the-art by fostering research on neural speech codecs under joint resource constraints, including in realistic noise and reverberation. This edition will also explore the combination of speech enhancement and coding, provide datasets for crowdsourced evaluation of neural speech codecs to support comparable and reproducible research, and share insights into the effectiveness of objective metrics across real-world conditions.
The outcomes will be disseminated through an ICASSP 2026 Satellite Workshop in May 2026 (subject to workshop proposal acceptance) and via related publications.
Tracks
The challenge comprises two tracks:
Track 1 – Transparency Codecs
Development of low-complexity, low-latency and low-bitrate speech codecs designed to preserve the perceptual transparency of input speech, including under mild noise and reverberation conditions.
Goal: Minimize perceptual speech degradation introduced by speech codecs while meeting the complexity, latency and bitrate constraints detailed under challenge Rules.
Focus Areas
-
Performance in clean conditions
-
Robustness to mild real-world distortions, e.g., background noise, reverberation
Track 2 – Speech Enhancement Codecs
Development of low-complexity, low-latency and low-bitrate enhancement codecs, i.e., those that in addition to coding and compression perform denoising and dereverberation, including under challenging acoustic conditions.
Goal: Enable transmission of intelligible and natural-sounding speech including under challenging acoustic conditions while meeting the complexity, latency and bitrate constraints detailed under challenge Rules.
Focus Areas
-
Transparency in clean conditions
-
Robustness to everyday distortions
-
Denoising performance in additive noise, including under challenging codnitions
-
Dereverberation performance, including under challenging conditions
We welcome end-to-end approaches, as well as modular pipelines, e.g., codecs with enhancement front-ends.