Loudness Assessment of Band Noise of Varying Bandwidth and Spectral Shape. An Evaluation of Various Frequency Weighting Networks

Abstract

The ability of different evaluation methods to predict loudness was assessed. In Experiment I, 12 subjects matched a band noise (900–1060Hz) to equal the loudness of each of 45 different noises covering the 15–1060 Hz frequency range, cut from three band spectra: Spectrum ‘STRAIGHT’ had its energy evenly distributed across frequency, ‘INVERSE’ was unbalanced towards the low-frequency bands and ‘A’ was unbalanced towards the high. Six frequency weighting networks: dB(A), dB(B), dB(C), dB(D), PLdB and PhondB were assessed. dB(B) fared best, followed by dB(D)and PhondB. dB(A) underestimated an dB(C) overestimated loudness especially when the noise only contained lower frequency bands. In Experiment II, 12 subjects matched the loudness of 7 noises cut from power spectrum ‘STRAIGHT’ at three different levels, all containing the 15–100 Hz low-frequency band. PLdB and PhondB predicted loudness best, followed by dB(D) and dB(B). Level proved to be of minor importance. In neither experiment did dB(A) predict the loudness of any noise better than dB(B) or dB(D).