Skip to content

Latest commit

 

History

History
52 lines (33 loc) · 2.58 KB

description.md

File metadata and controls

52 lines (33 loc) · 2.58 KB
Raw

IBM2

Antoine Liutkus, Fabian-Robert Stöter Inria and LIRMM, University of Montpellier, France antoine.liutkus@inria.fr

Additional Info

  • is_blind: no
  • additional_training_data: no

Supplementary Material

Method

Introduction

The Ideal Binary Mask (IBM) is a hard-assignment method that design a filter as a binary classification problem. This version computes the mask based on a ratio of power spectrograms.

Notations

We write $x$ for the 3-dimensional complex array obtained by stacking the Short-Time Frequency Transforms (STFT) of left and right channels of the mixture. Its dimensions are $F\times T\times 2$, where $F,T$ stand for the number of frequency bands and time frames, respectively. Its values at Time-Frequency (TF) bin $(f,t)$ are written $x(f,t)\in\mathbb{C}^2$, with entries $x_i(f,t)$ for $i\in{0,1}$. The mixture is taken as the sum of the sources images: $x(f,t)=\sum_j y_j(f,t)$, which correspond to the isolated instruments and are also stereo.

Filtering method

The IBM method computes source estimates by processing independently left and right channels of the mixture and multiplying it by a binary mask $M(f,t)\in{0,1}$: $\hat{y}{ij}(f,t)=M{ij}(f,t) x(f,t),$

where $M(f,t)$ is the parameter to be estimated. A classical reference for the IBM is for instance:

Wang, DeLiang. "On ideal binary mask as the computational goal of auditory scene analysis." Speech separation by humans and machines. Springer, Boston, MA, 2005. 181-197.

Binary mask estimation

This submission is an oracle, meaning that it knows the true sources to compute the binary mask.

Given the true sources $y_j$, the mask is computed very simply as:

$M_{ij}(f,t)=\frac{v_{ij}(f,t)}{\sum_j' v_{ij'}(f,t)}>0.5,$

where $v_{ij}(f,t)=\left|y_{ij}(f,t)\right|^2$ is the power spectrogram of $y_ij$ at time frequency bin $(t,f)$, hence the name IRM2 for this submission

References

  • A. Liutkus and F.-R. Stöter, The 2018 Signal Separation Evaluation Campaign, Proceedings of LVA/ICA, 2018

@inproceedings{sisec2018, title={The 2018 signal separation evaluation campaign}, author={A. Liutkus and F.-R. St{"o}ter and N. Ito}, booktitle={International Conference on Latent Variable Analysis and Signal Separation}, year={2018}, }