Gaussian Framework for Interference Reduction in Live Recordings
By D. Di Carlo

Diego Di Carlo presents to ADASP a Gaussian framework to reduce source interferences in live recordings.


When recording a live musical performance, each instrument is usually captured by dedicated microphones. However it is difficult to acoustically shield these microphones and in practice other sounds of other instruments are also captured, leading to so-called “interferences”. Reducing this phenomenon is desirable because it opens new possibilities for sounds engineers and also it has been proven that it increase performances of music analysis and processing tools (e.g. pitch tracking). The aim of this work is to extend state-of-the-art methods, which exploiting some heuristics, lead to sub-optimal estimation. We show how a rigorous Gaussian framework may be used to yield provably optimal algorithms to learn all the parameters required for good interference reduction. We present 4 alternative algorithms to this ends, providing an open-source Python implementation. Moreover the discussed methods are compared with the state-of-the-art in a perceptual evaluation on real-world multi-track live recordings, form the Montreux Jazz Festival. According to this listening test, the results showed that the proposed methods behave well.