The acoustical data sets were measured outdoors at 2 residences (H1–H2) located 980 m (H1), 1.3 km (H2), from the nearest wind turbine of South Australian wind farms (See *.zip files). The data sets contained a total of 6,000 10-s audio files of WFN which were randomly extracted from the measured data sets. There are an equal number of .wav files for each residence. Audio files have been renumbered for the purposes of establishing the benchmark dataset.
The data measured at H1 and H2 were used for establishing benchmark AM characteristics as well as training and validating the AM detection algorithm (10.1016/j.apacoust.2021.108286).
The WFN benchmark data set was primarily scored by a single scorer using a validated rating experiment procedure based on detection theory. AM presence was rated based on confidence level which varied from high confidence of AM absence (rating ‘1’), to uncertainty between AM presence/absence (rating ‘3’), to high confidence of AM presence (rating ‘5’) (See *.csv files; the numbering in the rating files is consistent with that used for the .wav files). A MATLAB GUI was designed for the experiment and a horizontal slider was used which allows users set a value by moving an indicator.
Quantifying the impact of wind farm noise on rural communities