Several models have been developed to calculate the acoustic roughness, but there is no official standardization yet.
In MOSQITO, the code is based on the algorithm described in Daniel and Weber, 1997. A step by step description of how to use MOSQITO to calculate the roughness is given in tutorial n°6
H.Fastl and E.Zwicker proposed some reference values for the roughness of amplitude-modulated tones in figure 11.2 of Zwicker and Fastl, 2007.
In Mosqito the test AM tones are generated using the 'test_signal_generation' script, in accordance with the equation (1) from the Daniel and Weber article.
The plot below compare different roughness implementations' results for amplitude-modulated tones (carrier frequency of 250, 1000, 4000 Hz and modulation frequency from 10 to 400 Hz):
- Daniel and Weber implementation described in the 1997 article (results from figure 3)
- Psysound3 software
- Mosqito
Mosqito implementation give similar results as other implementations. Zwicker and Fastl recommand that any roughness calculation should give results within a +/- 17% range around their reference values. Note that none of the 3 implementations respect this criteria. However, Daniel and Weber algorithm is considered as a reference. Any improvement of the roughness assessment is welcome to be implemented in Mosqito.
All the plots and scripts for more detailed validation can be found in this folder.
P.Daniel and R.Weber: Psychoacoustical roughness: implementation of an optimized model, Acustica acta.acustica, 1997.
H.Fastl and E.Zwicker: Psychoacoustics, Springer, Berlin, Heidelberg, 2007.