ACE Challenge Corpus: Acoustic Room Parameters from Speech and Impulse Responses

A corpus developed for the IEEE-AASP ACE Challenge to evaluate algorithms for blind acoustic parameter estimation from speech. It includes anechoic speech recordings from TU Delft and room impulse responses (RIRs) and noise recordings from seven rooms at Imperial College London. The dataset is described in a 2016 journal paper by Eaton et al. and includes multi-channel microphone configurations.

Use Cases

• Training models to estimate Reverberation Time (T60) directly from speech signals.
• Developing speech enhancement systems that require knowledge of the Direct-to-Reverberant Ratio (DRR).
• Benchmarking blind estimation algorithms for room acoustic parameters like reflection coefficients.
• Generating synthetic datasets for room acoustics research using the provided Matlab software.

Strengths

• Includes multi-channel room impulse responses (RIRs) from seven real rooms, with configurations ranging from single-channel to 32-channel.
• Provides anechoic speech recordings from multiple talkers (4 male for Dev, 5 male and 5 female for Eval) with transcriptions.
• Contains measured T60 and DRR values in fullband and ISO-266 frequency bands, along with room dimensions and microphone/source positions.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Last update date is unknown; freshness unverified.
• Data may reflect bias inherent to the specific rooms and recording setups at Imperial College London and TU Delft.

Provenance

Source

Imperial College London (RIRs), TU Delft (anechoic speech)

Collection Method

Recorded in anechoic chambers and real rooms using various microphone configurations.

Geography

London, UK (Imperial College); Delft, Netherlands (TU Delft)