Exponential Sine Sweep Recordings of Varying Characteristics Injected with Noise Signals

General This is a dataset of Exponential Swept Sine (ESS) recordings intentionally injected with noise signals. ESSs are one commen way to measure Room Impulse Responses (RIRs) in the domain of room acoustics. The dataset was created in the context of robot-assisted acoustic measurements, as these pose particular challanges for noise detection algorithms: the distance between source and receiver varies between measurements, sources may be obscured for certain positions of the robot and the robotic platform itself may produce additional noise. For some additional information see Comparison of Noise Detection Methods for Exponential Sine Sweep Measurements (G. Stolz et al., Proceedings of DAS|DAGA 2025 March 17-20, 2025, Copenhagen, pp. 1549-1552) Measurement Setup A setup with three loudspeakers and a microphone array was chosen for this measurement series. The measurements where conducted in the Hörlabor at Technische Universität Ilmenau, a 8, 4m × 7, 6m × 2, 8m big room with a T60 of 0.25 s (200 Hz - 2.5 kHz). See measurement_arrangement.png for the arrangement of sound sources and receiver inside the room. Note that Loudspeaker LS2 was obscured by a mobile wooden wall, from the microphone's point of view. While the ESS was played back over one loudspeaker, one of the others played a noise signal (impulse or birds chriping). Additionaly, all measurements where carried out once with the robot underneath the microphone array switched off and once with it switched on, proucing static nose from the platform's LiDAR-Scanner. The robot did not move during the measurements, as this data set is only intended to investigate noise, not the influence of different positions. For more information about the acoustic measurement robot used, see Autonomous Robotic Platform To Measure Spatial Room Impulse Responses (G. Stolz et al. Proceedings of DAGA 2023, March 6 - 9, 2023, Hamburg, pp. 59-61). The following table condenses all the different parameters: Noise Parameters Values Gain [dB] 0*, -6, -12 Loudspeaker LS1, LS2, LS3 Soundfile Impulse.wav, Birds.wav, None Target Interval 1st, 2nd, 3rd third during sweep or during decay time Static Noise Robot On/Off *0 dB beeing the loudest possible playback volume for the measurement setup. In order to achieve a balance between the random timing of the noise signals and a manageable number of combinations of noise parameters, the span of time during the ESS was divided into three equal intervals. Within the corresponding third, the noise signal was played at a random point in time. The fourth interval is directly after the the end of the ESS, during the decay time. See noise_interval_illustration.png for an illustration of the time intervals. One importaint parameter for ESSs is the length of the sweep, directly impacting the signal to noise ratio of the resulting room impulse response. For automated measurements the length can usually not be increased arbitrarily, assuming that the overall measurement duration is limited. Therefore different lengths of sweeps were used in this series of mesurements to compare their influence on noise detection and compensation. Sweep Parameters Values Gain [dB] -6, -12 Loudspeaker LS1, LS2, LS3 Length [s] 4, 8, 16, 32, 64 File Structure All combinations of sweep and noise parameters where recorded for this dataset, resulting in 4320 7-Channel recordings. The audiofiles played back over the loudspeakers are all stored in input_files.zip, including all gain differences and noises played back over the loudspeakers. noises.zip containes the used noise sounds files seperatly. The measurements where conducted in two runs: once with the robot switched off (measurements_without_static_noise.zip) and once with it switched on (measurements_with_static_noise.zip). Each corresponding zip-File contains a playlist.csv which in turn contains all the metadata for all recorded ESSs (input file, recording time, sweep and noise parameters, etc.). The recordings themselfs are stored in output_files inside the zip-Files. Around every minute inbetween the ESS recordings, 10 second recordings of background noise where performed. These are stored in silence_recordings, while silence.csv contains some additional metadata to those recordings.