https://github.com/mim/messl

Model-based EM Source Separation and Localization
https://github.com/mim/messl

Last synced: 14 days ago
JSON representation

Model-based EM Source Separation and Localization

• Host: GitHub
• URL: https://github.com/mim/messl
• Owner: mim
• Created: 2012-03-05T02:58:35.000Z (over 12 years ago)
• Default Branch: master
• Last Pushed: 2017-01-13T02:32:09.000Z (almost 8 years ago)
• Last Synced: 2024-08-02T07:19:42.705Z (4 months ago)
• Language: Matlab
• Homepage:
• Size: 485 KB
• Stars: 36
• Watchers: 5
• Forks: 9
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

• awesome-speech-enhancement - [Code

README

        
# Model-Based EM Source Separation and Localization

Copyright 2006-2009 Michael I Mandel and Ron Weiss, all rights reserved

 and 

Last updated 2009-08-20

## Basic usage to separate two sources:

```matlab

addpath('./plottools/')

% % Load stereo wav files of the same length and mix them

% [y1 fs] = wavread('data/src1.wav');

% [y2 fs] = wavread('data/src2.wav');

% lr = y1' + y2';

% Load pre-mixed version of those two files (don't forget the transpose)

[lr fs] = wavread('data/mix.wav');

lr = lr';

% Derive grid of tau values (use your numbers here)

tau = tauGrid(0.15, fs, 31);

% Run MESSL

[m,p] = messl(lr, tau, 2, 'vis', 1);

% Reconstruct wavforms from masks

yhat1 = reconstruct(m, lr, 1);

yhat2 = reconstruct(m, lr, 2);

```

## Fancier usage

Initialized from PHAT-histogram:

```matlab

% Localize and then run MESSL

tdoa = phatLoc(lr, tau, 2, 1024, 1);

[m,p] = messl(lr, tau, 2, 'vis', 1, 'tauPosInit', tdoa);

```

Even fancier usage, garbage source and ILD prior (better in reverb,

but only when using dummy-head recordings):

```matlab

[m,p] = messl(lr, tau, 2, 'vis', 1, 'ildPriorPrec', 3, ...

              'GarbageSrc', 1, 'sr', 16000);

```

Can also use prob2mask to make the mask more definitive, i.e. closer

to binary, but not binary.

```matlab

m2 = prob2mask(m);

yhat1 = reconstruct(m2, lr, 1);

yhat2 = reconstruct(m2, lr, 2);

```