Institut für Nachrichtentechnik

Univ.-Prof. Dr.-Ing. Jens-Rainer Ohm

Navigation

• 1.1: Home
• 1.2: Team
  • 1.2.1: Prof. Ohm
  • 1.2.2: Dr. Rohlfing
    • 1.2.2.1: Publication Websites
      • 1.2.2.1.1: Adaptive coding of non-negative factorization parameters with application to informed source separation

      • Current page is 1.2.2.1.2: Audio source separation with magnitude priors: the BEADS model

      • 1.2.2.1.3: Quantization-aware Parameter Estimation for Audio Upmixing
      • 1.2.2.1.4: Very Low Bitrate Spatial Audio Coding with Dimensionality Reduction
      • 1.2.2.1.5: Generalized constraints for NMF with application to Informed Source Separation
      • 1.2.2.1.6: NMF-based Informed Source Separation
  • 1.2.4: Sekretariat Frau Szelag
  • 1.2.5: T. Claßen
  • 1.2.6: D. Mehlem
  • 1.2.7: C. Jansen
  • 1.2.10: I. Reißel
  • 1.2.11: G. Kaschel
  • 1.2.12: H. Flasche
  • 1.2.13: Alumni
• 1.4: Lehre
• 1.5: Forschung
• 1.6: Projekte
• 1.7: Publikationen
• 1.8: Software
• 1.10: Abschlussarbeiten
• 1.12: Bibliothek
• 1.13: Sammlung
• 1.14: FRIENT e.V.
• 1.15: Sitemap
• 1.16: English Pages

Audio source separation with magnitude priors: the BEADS model

Antoine Liutkus, Christian Rohlfing, Antoine Deleforge

Presented at the 43rd IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP 2018), 15-20 April 2018, Calgary, Canada
Conference homepage

Abstract

Audio source separation comes with the need to devise mul-tichannel filters that can exploit priors about the target signals. In that context, experience shows that modeling magnitude spectra is effective. However, devising a probabilistic model on complex spectral data with a prior on magnitudes is non trivial, because it should both reflect the prior but also be tractable for easy inference. In this paper, we approximate the ideal donut-shaped distribution of a complex variable with approximately known magnitude as a Gaussian mixture model called BEADS (Bayesian Expansion Approximating the Donut Shape) and show that it permits straightforward inference and filtering while effectively constraining the magnitudes of the signals to comply with the prior. As a result, we demonstrate large improvements over the Gaussian baseline for multichannel audio coding when exploiting the BEADS model.

[ Paper | Slides | Bibtex ]