An iterative approach to monaural musical mixture de-soloing

Abstract

In this article, we introduce a novel approach for monaural source separation with the specific aim to separate a polyphonic musical recording into two main sources: a main instrument (or melody) track and an accompaniment track. To that aim, we propose to model the power spectral densities (PSDs) of both contributions with a source/filter model for the main instrument while retaining a model emphasizing temporal repetitions of the musical background. We show that improved source separation performances can be obtained by a two-step estimation strategy where the model parameters are re-estimated in a second stage by adequately exploiting the main melody line estimated in a first stage. The experiments conducted on several monaural signal databases show that our system achieves state-of-the-art performances compared to other unsupervised source separation algorithms.

Source/Filter Model for Unsupervised Main Melody Extraction From Polyphonic Audio Signals

“…We adapted the IMM model in order to better fit the task at hand and also included a second parameter estimation step, which takes advantage of the estimated melody. The details of the implementation are given in [24]. On a database described on http://perso.telecom-paristech.fr/grichard/icassp09/, we obtain results comparable to [22] in terms of SDR [25]: 8.8dB of SDR gain for the separated main voice and 2.6dB of SDR gain for the accompaniment (see details in [24] 2) Multipitch Tracking: Multipitch tracking is a related task for which one desires to transcribe all the fundamental frequencies within each analysis frame of a polyphonic music signal.…”

Multimodal similarity between musical streams for cover version detection

“…In the approach proposed in this paper, each song is decomposed into 3 different "modalities": its original mixture, the main melody and accompaniment as estimated using [8]. Each modality is then compared using the method proposed in [1] which obtained the best results at the international evaluation MIREX 2007 campaign.…”

“…To execute separated analysis on the melody and the accompaniment, we first use a main melody extraction technique [8]. Each song is assumed to be the mixture of two contributions: the leading voice, usually a singer, which is modeled by a source/filter model, and the accompaniment, which is modeled using Non-negative Matrix Factorization (NMF) formalism.…”

“…Many reasons can be stated to explain this fact, from the quality of the separation to the heterogeneity of the representations involved in respectively describing the melody and the accompaniment. Nevertheless, we show that the use of a leading voice separation algorithm [8] can help identifying an invariant aspect among the two separate components as well as the mixture. Promising results are obtained and it is, in particular, shown that early and late fusion strategies that combine information from the mixture and the accompaniment lead to enhanced performances on a public database (e.g.…”

Concurrent estimation of singing voice F0 and phonemes by using spectral envelopes estimated from polyphonic music

“…A great deal of research has been conducted on lyrics or phoneme recognition in the polyphonic singing voice [2,3,4,5], and vocal F0 estimation [6,7,8,9,10]; however, to our knowledge, no studies have concurrently estimated both F0 and phoneme. The approaches of these studies differ from ours in that they either ignored the influence of accompaniment sound or segregated the singing voice from other instruments.…”