The ACE challenge — Corpus description and performance evaluation

Abstract

Knowledge of the Direct-to-Reverberant Ratio (DRR) and Reverberation Time (T <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">60</sub> ) can be used to better perform speech and audio processing such as dereverberation. Established methods compute these parameters from measured Acoustic Impulse Responses (AIRs). However, in many practical situations the AIR is not available and the parameters must be estimated non-intrusively directly from noisy speech or audio signals. The Acoustic Characterization of Environments (ACE) Challenge is a competition to identify the most promising non-intrusive DRR and T <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">60</sub> estimation methods using real noisy reverberant speech. We describe the ACE corpus comprising multi-channel AIRs, and multi-channel noise including ambient, fan and babble noise recorded in the same environment as the measured AIRs, along with the corresponding DRR and T <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">60</sub> measurements. The evaluation methodology is discussed and comparative results are shown.