Kernel Additive Modeling for interference reduction in multi-channel music recordings

Abstract

When recording a live musical performance, the different voices, such as the instrument groups or soloists of an orchestra, are typically recorded in the same room simultaneously, with at least one microphone assigned to each voice. However, it is difficult to acoustically shield the microphones. In practice, each one contains interference from every other voice. In this paper, we aim to reduce these interferences in multi-channel recordings to recover only the isolated voices. Following the recently proposed Kernel Additive Modeling framework, we present a method that iteratively estimates both the power spectral density of each voice and the corresponding strength in each microphone signal. With this information, we build an optimal Wiener filter, strongly reducing interferences. The trade-off between distortion and separation can be controlled by the user through the number of iterations of the algorithm. Furthermore, we present a computationally effective approximation of the iterative procedure. Listening tests demonstrate the effectiveness of the method.

Guitar String Separation Using Non-Negative Matrix Factorization and Factor Deconvolution

“…These tracks are not ideal for use as ground truth because the goal is to estimate the signals produced by each string individually. Conveniently, the dataset includes tracks de-bleeded using Kernel Additive Modeling for Interference Reduction (KAMIR) [14]. These tracks were used for creating ground truth for objective evaluation.…”

Applying Source Separation to Music

“…Dealing with multitrack recordings featuring some amount of bleeding has always been one of the daily duty of professional sound engineers. In the recent years, some research was conducted to design engineering tools aimed at reducing those interferences (Kokkinis et al, 2012;Prätzlich et al, 2015). In the remainder of this section, we briefly present the Multitrack Interference RemovAl (MIRA) model presented in (Prätzlich et al, 2015).…”

“…First, all the recordings x i are assumed independent. This amounts to totally discard any phase dependency that could be present in the recordings and proved a safe choice in case of very complex situations such as the real-world orchestral recordings considered in (Prätzlich et al, 2015). Second, each recording is modelled using the Gaussian model presented in chapter ??…”

Exploring sound source separation for acoustic condition monitoring in industrial scenarios

“…KAMIR is based on the Kernel Additive Model (KAM) proposed in [18] and assumes that each source is predominant in its dedicated channel. By using an approach based on a generalized Wiener filter, KAMIR can estimate the sources in an iterative manner [17].…”

“…This section briefly describes the three methods for SSS which are applied in the experiment described later in Section III, namely Azimuth Discrimination and Resynthesis (ADRess) [15], Frequency Domain Source Identification and Manipulation in Stereo Mix (FDSI) [16], and Kernel Additive Modeling for Interference Reduction (KAMIR) [17]. Even though these algorithms were developed within the music separation community, the fact that they can handle stereo signals without making strong assumptions about the harmonicity or continuity of the target source, makes them promising candidates for industrial condition monitoring applications.…”