Evaluation and reconstruction of a blended string ensemble: Recent developments & findings

Title: Evaluation and reconstruction of a blended string ensemble: Recent developments & findings
Authors: Jithin Thilakan
TOR: Video Series
Series: Analysis, Creation, and Teaching of Orchestration Symposiums and Workshops
Source URL: https://www.timbreandorchestration.org/tor/modules/video-series/symposiums-workshops/y3/thilakan

Abstract:

The blending between sound sources in a joint performance is an important feature relevant in the evaluation and reconstruction of the sound field of an orchestra/ensemble in real life and also in virtual reality domain. It is also relevant in fields such as music composition & orchestration, stage/room acoustic adaptation, and music production. Followed by a macroscopic assessment of blending using a live listening test, microscopic level investigation, i.e. signal analysis of the blending is presented in this work. The violins played in a joint performance are recorded using spot microphones, and a listening test with Tonmeisters and musicians is carried out using sound samples extracted from the recording. Musical and audio signal features have been extracted from the sound samples to investigate those potential features that contribute to the impression of blending. The higher dimensional audio feature data is analyzed using statistical dimensionality reduction techniques such as PCA, LDA, etc. The first findings report that the MFCC features analyzed using LDA support a hypothesis to predict the blending impression in microscopic levels. The second part of this presentation includes an investigation of the conservation of blending impression in simulated environments. A room acoustic simulation of Detmold concert house is developed using the commercial software ODEON and optimized using the traditional room-acoustic parameters measured from the concert hall. A violin ensemble played in the concert hall is auralized using the room acoustic simulation. Finally, the sound field in simulations and real measurements are compared using perceptual evaluations to check whether the blending impression is conserved in simulations.