The Ambisonic Recordings of Typical Environments (ARTE) Database

Abstract

Everyday listening environments are characterized by far more complex spatial, spectral and temporal sound field distributions than the acoustic stimuli that are typically employed in controlled laboratory settings. As such, the reproduction of acoustic listening environments has become important for several research avenues related to sound perception, such as hearing loss rehabilitation, soundscapes, speech communication, auditory scene analysis, automatic scene classification, and room acoustics. However, the recordings of acoustic environments that are used as test material in these research areas are usually designed specifically for one study, or are provided in custom databases that cannot be universally adapted, beyond their original application. In this work we present the Ambisonic Recordings of Typical Environments (ARTE) database, which addresses several research needs simultaneously: realistic audio recordings that can be reproduced in 3D, 2D, or binaurally, with known acoustic properties, including absolute level and room impulse response. Multichannel higher-order ambisonic recordings of 13 realistic typical environments (e.g., office, cafè, dinner party, train station) were processed, acoustically analyzed, and subjectively evaluated to determine their perceived identity. The recordings are delivered in a generic format that may be reproduced with different hardware setups, and may also be used in binaural, or single-channel setups. Room impulse responses, as well as detailed acoustic analyses, of all environments supplement the recordings. The database is made open to the research community with the explicit intention to expand it in the future and include more scenes.

Measuring Speech Intelligibility and Hearing-Aid Benefit Using Everyday Conversational Sentences in Real-World Environments

“…The ECO-SiN corpus comprises 192 naturally spoken sentences, in which four lists of 16 sentences were spoken with three different vocal efforts. The average sentence length is 6.3 words, and an example sentence is “That discovery was like really interesting for me.” In brief, the sentences were extracted from two people engaging in unscripted conversation while they listened to three different realistic background noises from the ARTE database ( Weisser et al, 2019b ); a church, an indoor café, a busy food court (see Table 2 ) via highly open headphones. The background noises were selected based on the conversational speech levels determined by Weisser and Buchholz (2019) .…”

“…The background noises were drawn from the ARTE database ( Weisser et al, 2019b ), which were recorded with a 62-channel hard-sphere microphone array and encoded into the higher-order Ambisonics (HOA) format. They were then decoded here for simulated playback with the spherical 41-channel loudspeaker array inside the anechoic chamber of the Australian Hearing Hub, Macquarie University.…”

“…To maximize the realism of the ECO-SiN sentences, and thereby to optimize their perceptual integration with the background noise, realistic room reverberation was added by convolving the individual sentences with multi-channel Room Impulse Responses (RIRs). The RIRs were taken from the ARTE database ( Weisser et al, 2019b ) and measured in the real-world environments with a loudspeaker at a distance of 1.3 m in front of the 62-channel microphone array. As for the noise recordings, the measured RIRs were encoded into the HOA format and decoded for simulated playback with the 41-channel loudspeaker array.…”

“…The free-field levels of the reverberant ECO-SiN sentences were slightly higher than the values shown in Table 2 , the reverberation providing an increase in the effective test SNR by: + 0.8 dB, church: + 0.1 dB, living room: + 1.2 dB, café: + 0.8 dB, dinner party: + 1.6 dB, and food court: + 0.5 dB. For a detailed description of the microphone array recording, HOA encoding and decoding, and the RIR manipulation process see Weisser et al (2019b) .…”

SAQAM: Spatial Audio Quality Assessment Metric

“…For the perturbation set, we consider all degradations commonly found in various audio processing tasks including additive noise, speech distortions (e.g. clipping and frequency masking, frequency resampling, pitch shifting), compression (e.g., mu-law and MP3), and recorded binaural sounds [30,31]. We also use the data collected using the binaural multichannel wiener filter (MWF) [32] algorithm, and find that adding datasets and perturbations with subtle differences increases the robustness of our model to small differences.…”