Anechoic or semi-anechoic instrument recordings are readily available for academic purposes on a few different sites online. Anechoic recordings are commonly used in auralizations, which today practically means convolving recordings with simulated or measured room impulse responses. Besides the possibility of being used as such, these recordings offer other possibilities for the generation of test stimuli. Many studies, such as, studies on auditory distance perception or source separation, would benefit from available experimental materials which would not be strictly musical but could still be linked to the perception of musical stimuli. The goal of the current investigation is to develop a procedure for generating such materials, i.e., an anechoic audio corpus which can be used in the future investigations of room acoustics and in related fields. Moreover, the aim is to provide a framework for further development of processes where a large number of stimuli can be generated in a systematic way. In this study, the proposed framework is instantiated by producing two sets of stimuli by either directly segmenting anechoic music or randomly combining different segments of anechoic instrument tracks. Music information retrieval (MIR) approach is used to calculate 14 musical features of the generated sets of stimuli. Principal component analysis is used to analyse the sample spaces enabling the experimenter to select a small number stimuli with desired characteristics. The benefits and drawbacks of this stimuli generation approach including some important theoretical underpinnings of experimental design are also discussed.
|Otsikko||EAA Joint Symposium on Auralization and Ambisonics, Berlin, 3.-5.4.2014|
|Toimittajat||Michael Vorländer Stefan Weinzierl, Hans-Joachim Maempel Franz Zotter|
|Kustantaja||Universitätsverlag der TU Berlin|
|Tila||Julkaistu - 2014|
|OKM-julkaisutyyppi||A4 Artikkeli konferenssijulkaisuussa|
- auralization, perception, listening tests, acoustics, audio, music, concert halls