A perceptually orientated approach for automatic classification of timbre content of orchestral excerpts
Aurélien Antoine, Eduardo Reck Miranda · The Journal of the Acoustical Society of America · 2017
In this paper, we report on the development of a perceptually orientated and automatic classification system of timbre content within orchestral audio samples. Here, we have decided to investigate polyphonic timbre, a phenomenon emerging from the mixture of instruments playing simultaneously. Moreover, we are focusing on the perception of the entire orchestral sound, and not individual instrumental sound. For accessibility to non-Acoustics experts, we chose to use verbal descriptors of timbre quality, such as brightness and roughness, to represent the timbral content of the samples. We based our acoustic analysis on the existing research into the perception and description of timbre. However, with a lack of agreed metrics, we had to establish a comparative scale for each timbre attribute implemented in our system, which is based on an analysis of audio recordings, in order to identify the dominant timbral attribute. To improve the classification accuracy, the system continually calibrates this scale as new audio files are analyzed. Preliminary analysis of our results shows a correlation between the system’s classification and human perception, which is promising for further developments, such as standardizing metrics for perceived responses of timbral attributes or implementing systems for music production tasks.