|
c@0
|
1 %% Multichannel audio source separation evaluation framework (MASSEF)
|
|
c@0
|
2 %
|
|
c@0
|
3 %% Overview
|
|
c@0
|
4 %
|
|
c@8
|
5 % <<../MASSEF.png>>
|
|
c@8
|
6 %
|
|
c@0
|
7 % The multichannel audio source separation evaluation framework is designed
|
|
c@9
|
8 % to facilitate the development and evaluation of audio source separation
|
|
c@9
|
9 % algorithms. The framework generates the mixture(s), provides the
|
|
c@9
|
10 % mixture(s) to the separation algorithm(s), and evaluates the outputs of
|
|
c@9
|
11 % the separation algorithm(s). The framework can also calculate and
|
|
c@9
|
12 % evaluate the ideal masks for the purposes of comparison.
|
|
c@0
|
13 %
|
|
c@0
|
14 % Sources may have any number of channels; the framework evaluates each
|
|
c@0
|
15 % channel. The use of iosr.bss.mixture objects facilitate the evaluation of
|
|
c@0
|
16 % spatialised mixtures (e.g. binaural).
|
|
c@0
|
17 %
|
|
c@0
|
18 % The framework comprises two main classes:
|
|
c@0
|
19 %
|
|
c@0
|
20 % * <help_MASSEF.html MASSEF>—multichannel audio source separation
|
|
c@0
|
21 % evaluation framework class.
|
|
c@0
|
22 % * <help_MASSEFresults.html MASSEFresults>—the results class used by
|
|
c@0
|
23 % MASSEF.
|
|
c@0
|
24 %
|
|
c@0
|
25 % <help_Example.html View an example of the use of these classes.>
|
|
c@0
|
26 %
|
|
c@0
|
27 % MASSEF can be used to evaluate and compare binaural separation algorithms,
|
|
c@0
|
28 % provided that the algorithms conform to the <help_Separators.html
|
|
c@0
|
29 % required format>.
|
|
c@0
|
30 %
|
|
c@0
|
31 %% Installation
|
|
c@0
|
32 %
|
|
c@0
|
33 % Before using the framework, you need to download and install its
|
|
c@0
|
34 % dependencies. This can be done automatically using the
|
|
c@9
|
35 % <help_MASSEF_install.html MASSEF.install()> method. Before each session,
|
|
c@0
|
36 % MASSEF needs to be started using the <help_MASSEF_start.html
|
|
c@9
|
37 % MASSEF.start()> method.
|
|
c@0
|
38 %
|
|
c@0
|
39 %% Usage
|
|
c@0
|
40 %
|
|
c@0
|
41 % The framework can be run in two ways:
|
|
c@0
|
42 %
|
|
c@0
|
43 % # by providing |iosr.bss.mixture| objects and separation algorithms, or
|
|
c@0
|
44 % # by providing estimate and true source wav files.
|
|
c@0
|
45 %
|
|
c@9
|
46 % If 1), the framework operates as described above. In addition, the
|
|
c@9
|
47 % framework can:
|
|
c@9
|
48 %
|
|
c@9
|
49 % * evaluate localisation accuracy (if the algorithm performs localisation)
|
|
c@9
|
50 % of any azimuth/elevation estimates returned by the algorithm, and
|
|
c@9
|
51 % * evaluate time-frequency mask accuracy (if the algorithm calculates
|
|
c@9
|
52 % one).
|
|
c@9
|
53 %
|
|
c@9
|
54 % Use the <help_MASSEF_execute.html MASSEF.execute()> method to operate in
|
|
c@9
|
55 % this mode.
|
|
c@0
|
56 %
|
|
c@0
|
57 % If 2), the framework evaluates only the supplied estimate(s) using
|
|
c@9
|
58 % signal-related metrics. Use the <help_MASSEF_evaluate.html
|
|
c@9
|
59 % MASSEF.evaluate()> method to operate in this mode.
|
|
c@0
|
60 %
|
|
c@0
|
61 % <html>
|
|
c@0
|
62 % <hr>
|
|
c@0
|
63 % <p>Copyright © 2016 <a href="http://www.surrey.ac.uk">University of Surrey</a><br>
|
|
c@0
|
64 % <a href="http://iosr.uk">Institute of Sound Recording, University of Surrey, UK</a></p>
|
|
c@0
|
65 % </html>
|
