The ICASSP 2024 Audio Deep Packet Loss Concealment Challenge

TL;DR

The paper presents the ICASSP 2024 Audio Deep PLC Challenge to push ML-based packet loss concealment under realistic long-burst losses and full-band audio, evaluated with the ITU-T P.804 multi-criteria perceptual protocol. It details a dataset built from real-world loss traces and diverse speech sources, with careful preprocessing and a staged release of validation and blind sets, and it uses crowd-sourced subjective ratings to assess performance across multiple quality dimensions. Nine systems are compared, eight meeting real-time constraints, and the results show statistically significant differences, including joint first place for two teams, highlighting progress and remaining challenges in practical PLC. The work provides a concrete benchmark and evaluation framework to drive advances in ML-based PLC for real-time communications.

Abstract

Audio packet loss concealment is the hiding of gaps in VoIP audio streams caused by network packet loss. With the ICASSP 2024 Audio Deep Packet Loss Concealment Grand Challenge, we build on the success of the previous Audio PLC Challenge held at INTERSPEECH 2022. We evaluate models on an overall harder dataset, and use the new ITU-T P.804 evaluation procedure to more closely evaluate the performance of systems specifically on the PLC task. We evaluate a total of 9 systems, 8 of which satisfy the strict real-time performance requirements of the challenge, using both P.804 and Word Accuracy evaluations.