Traffic noise and perceived soundscapes: a case study G. Licitra, G. Memoli ARPAT (Regional Agency for Environmental Protection of Tuscany), Via V. Veneto 22, 56127 Pisa (I), {g.licitra, g.memoli}@arpat.toscana.it , D. Botteldooren, B. De Coensel Acoustics Group, Department of Information Technology, Ghent University, St. Pietersnieuwstraat 41, 9000 Gent (B), d.botteldooren@ugent.be The present study will deal with a practical problem of noise mapping: the acoustical classification of roads in the medium size municipality of San Giuliano Terme, in Tuscany (I). At first, using a common approach in literature, the main road infrastructures have been classified in three clusters, applying threshold methods to traffic flow measurements. The time history of L Aeq was then acquired in selected sites over a continuous period of 24 hours and the power spectrum G(f) was then calculated from dB values of L Aeq over 15 minute intervals. A power law was fitted to G(f) in the range [0.02, 0.2] Hz, obtaining two parameters - B(t) and A(t) - over a complete day. Hierarchical clustering was finally performed and the clusters obtained resembled the ones based on traffic flow. The values of B(t) and A(t) have been compared with other indicators: comparison with Number of Noise Event (NNE) and L 10 -L 90 has been reported here. New fitting possibilities for G(f) has also been explored and discussed in this work. Finally, statistical analysis has been used to get further information on the meaning of B. 1 Introduction The search for new indicators to distinguish soundscapes is crucial for the noise control in existing quiet areas, as prescribed by the 2002/49/EC END, and for the drawing up of cost/effective action plans. In this direction, some studies classify external environments acquiring their “acoustical characteristic” and comparing it to people’s perceptions both in “quiet” and “disturbed” areas. The final goal is to identify new indicators which could help to design new areas and improve existing ones, directing technical efforts to achieve the “ideal characteristic” of the site, as defined by the expected utilization. In 1978, R. F. Voss & J. Clarke [1] studied long term variation of loudness, for different kind of man- produced music, in order to find a better way to generate stochastic computer music. They compared the loudness power spectrum of signals acquired during 12 hours in three different radio channels (classical, rock, news) with the ones due to a few classical masterpieces, finding that all these sources shared an 1/f dynamical behaviour, where the “frequency” f is not related to the signal emitted every second, but to the occurrence of events in the time history (i.e.: if the explored range were [0.002¸0.2] Hz, target events would occur in the time interval between 500 ms and 5s). Their observations suggested that musical pieces where the frequency and duration of each note had been determined by 1/f noise sources sounded “pleasing”, while those generated by 1/f 2 (Brownian noise) sources sounded too correlated (“boring”, in other words). The modern theory of stochastic chaos has linked 1/f power spectrums to “self-organized criticality” [2]. As the latter description covers a great number of environmental situations, it was thought to study the long-term noise dynamics in urban and rural scenarios, seeking in it a finite set of indicators to characterize the site [3]; [4]. A previous study [5] described different soundscapes with 13 parameters calculated from the long-term dynamics of three different signals: L Aeq (t), Zwicker loudness and instantaneous pitch. After acquiring a noise signal in a definite site (15 min. duration during daytime, in 1/3 octave bands), the power spectrum G(f) of Zwicker’s loudness and pitch in the range [0.002¸ 5] Hz was fitted in [5] with: (1) B f A f G ⋅ = ) ( Hierarchical clustering (H-C, in the following text) based on “within-group linkage” was then performed on the data, using Pearson’s correlation coefficient r as a distance, obtaining two large clusters that could be characterized as “pleasing” and “predictable”. In more recent works [6] [7], the correlation between the clustering and the soundscape perception was investigated, with particular interest in sites infected by traffic noise. In particular, in [7] it was shown that a fair correlation exists between the perceived soundscape (assessed by simple questionnaires) and clustering results, obtained using the values of B at fixed moments during the 24 hours. They also showed that, for each site, the fitted values of B did not depend directly on the corresponding L Aeq , but that they gave complementary information, to be interpreted through a deeper study. It was also seen that, when the traffic was low or in night hours, the G(f) dependence in eq. (1) failed to describe the acquired power spectrums over the frequency range. After dealing with a practical case, using H-C and calculated values of B to classify the main roads in a 1875