Predição de intensidade sonora percebida (loudness ) para áudio espacial

Abstract

Loudness control for brodcasting is a common and legally required practice since the International Telecommunication Union (ITU) Recommendation ITUR BS.1770 for objective measurements in multichannel audio. Recommendations and regulations based on the ITU-R algorithm have been published worldwide, including Brazil. There is scope for improving national regulations in light of recent contributions to the field, and also for adapting the ITU-R model to measurements in advanced audio systems. This work pursues these two goals by testing the parameters of the Brazilian standard with a real-time loudness controller using short-form descriptors and by developing a new objective measurement model adapted to the new spatial audio formats. The proposed method performed well compared to other loudness models, although it was purely signal processing based and its readings were not very close to subject responses. The potential benefits of a more perceptually motivated model led to a PhD placement in the Institute of Sound Recording at the University of Surrey (UK), where listening tests were conducted to assess positional parameters of distance, azimuth and elevation, whose results served as a basis for deriving gain correction curves and a new directional weighting for the ITU-R model. General results point to advancements in the regulatory and standardization fronts, either by the elaboration of a strategy to improve the Brazilian standard of loudness, or by comparing this new prediction method with the critical fortune of loudness models through measurements on audio content for multichannel reproduction systems. The developed model resulted in the best trade-off between prediction errors (RMSE*), correlation between predictions and subject responses, and mean run time.