Mercurial > hg > camir-aes2014

comparison toolboxes/mp3readwrite/demo_mp3readwrite.m @ 0:e9a9cd732c1e tip

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
first hg version after svn
author wolffd
date Tue, 10 Feb 2015 15:05:51 +0000
parents
children
comparison
equal deleted inserted replaced
-1:000000000000 0:e9a9cd732c1e
1 %% MP3 reading and writing
2 %
3 % These function, mp3read and mp3write, aim to exactly duplicate
4 % the operation of wavread and wavwrite for accessing soundfiles,
5 % except the soundfiles are in Mpeg-Audio layer 3 (MP3) compressed
6 % format. All the hard work is done by external binaries written
7 % by others: mp3info to query the format of existing mp3 files,
8 % mpg123 to decode mp3 files, and lame to encode audio files.
9 % Binaries for these files are widely available (and may be
10 % included in this distribution).
11 %
12 % These functions were originally developed for access to very
13 % large mp3 files (i.e. many hours long), and so avoid creating
14 % the entire uncompressed audio stream if possible. mp3read
15 % allows you to specify the range of frames you want to read
16 % (as a second argument), and mp3read will construct an mpg123
17 % command that skips blocks to decode only the part of the file
18 % that is required. This can be much quicker (and require less
19 % memory/temporary disk) than decoding the whole file.
20 %
21 % mpg123 also provides for "on the fly" downsampling at conversion
22 % to mono, which are supported as extra options in mp3read.
23 %
24 % mpg123 can read MP3s across the network. This is supported
25 % if the FILE argument is a URL (e.g. beginning 'http://...').
26 %
27 % mp3info sometimes gets the file size wrong (as returned by the
28 % mp3read(...'size') syntax). I'm not sure when this happens
29 % exactly, but it's probably a result of VBR files. In the worst
30 % case, figuring the number of samples in such a file requires
31 % scanning through the whole file, and mp3info doesn't usually do
32 % this.
33 %
34 % For more information, including advice on handling MP4 files,
35 % see http://labrosa.ee.columbia.edu/matlab/mp3read.html
36
37 %% Example usage
38 % Here, we read a wav file in, then write it out as an MP3, then
39 % read the resulting MP3 back in, and compare it to the original
40 % file.
41
42 % Read an audio waveform
43 [d,sr] = wavread('piano.wav');
44 % Save to mp3 (default settings)
45 mp3write(d,sr,'piano.mp3');
46 % Read it back again
47 [d2,sr] = mp3read('piano.mp3');
48 % mp3 encoding involves some extra padding at each end; we attempt
49 % to cut it off at the start, but can't do that at the end, because
50 % mp3read doesn't know how long the original was. But we do, so..
51 % Chop it down to be the same length as the original
52 d2 = d2(1:length(d),:);
53 % What is the SNR (distortion)?
54 ddiff = d - d2;
55 disp(['SNR is ',num2str(10*log10(sum(d(:).^2)/sum(ddiff(:).^2))),' dB']);
56 % Do they look similar?
57 subplot(211)
58 specgram(d(:,1),1024,sr);
59 subplot(212)
60 plot(1:5000,d(10000+(1:5000),1),1:5000,d2(10000+(1:5000)));
61 % Yes, pretty close
62 %
63 % NB: lame followed by mpg123 causes a little attenuation; you
64 % can get a better match by scaling up the read-back waveform:
65 ddiff = d - 1.052*d2;
66 disp(['SNR is ',num2str(10*log10(sum(d(:).^2)/sum(ddiff(:).^2))),' dB']);
67
68 %% Delay, size, and alignment
69 %
70 % In mid-2006 I noticed that mp3read followed by mp3write followed by
71 % mp3read effectively delayed the waveform by 2257 samples (at 44
72 % kHz). So I introduced code to discard the first 2257 samples to ensure
73 % that the waveforms remained time aligned. As best I could understand,
74 % mpg123 (v 0.5.9) was including the "warm-up" samples from the
75 % synthesis filterbank which are more properly discarded.
76 %
77 % Then in late 2009 I noticed that some chord recognition code, which
78 % used mp3read to read files which were then segmented on the basis of
79 % some hand-marked timings, suddenly started getting much poorer
80 % results. It turned out that I had upgraded my version of mpg123 to v
81 % 1.9.0, and the warm-up samples had been fixed in this version. So my
82 % code was discarding 2257 *good* samples, and the data was skewed 51ms
83 % early relative to the hand labels.
84 %
85 % Hence, the current version of mp3read does not
86 % discard any samples by default -- appropriate for the recent versions
87 % of mpg123 included here. But if you know you're running an old, v
88 % 0.5.9, mpg123, you should edit the mp3read.m source to set the flag
89 % MPG123059 = 1.
90 %
91 % Note also that the 'size' function relies on the number of
92 % blocks reported by mp3info. However, many mp3 files include
93 % additional information about the size of the file in the
94 % so-called Xing header, embedded in the first frame, which can
95 % specify that a certain number of samples from start and end
96 % should additionally be dropped. mp3info doesn't read that,
97 % and there's no way for my code to probe it except by running
98 % mpg123. Hence, the results of mp3read(fn,'size') may sometimes
99 % overestimate the length of the actual vector you'll get if
100 % you read the whole file.
101
102 %% External binaries
103 % The m files rely on three external binaries, each of which is
104 % available for Linux, Mac OS X, or Windows:
105 %
106 % *mpg123* is a high-performance mp3 decoder. Its home page is
107 % http://www.mpg123.de/ .
108 %
109 % *mp3info* is a utility to read technical information on an mp3
110 % file. Its home page is http://www.ibiblio.org/mp3info/ .
111 %
112 % *lame* is an open-source MP3 encoder. Its homepage is
113 % http://lame.sourceforge.net/ .
114 %
115 % The various authors of these packages are gratefully acknowledged
116 % for doing all the hard work to make these Matlab functions possible.
117
118 %% Installation
119 % The two routines, mp3read.m and mp3write.m, will look for their
120 % binaries (mpg123 and mp3info for mp3read; lame for mp3write) in
121 % the same directory where they are installed. Binaries for
122 % different architectures are distinguished by their extension,
123 % which is the standard Matlab computer code e.g. ".mac" for Mac
124 % PPC OS X, ".glnx86" for i386-linux. The exception is Windows,
125 % where the binaries have the extension ".exe".
126 %
127 % Temporary files
128 % will be written to (a) a directory taken from the environment
129 % variable TMPDIR (b) /tmp if it exists, or (c) the current
130 % directory. This can easily be changed by editing the m files.
131
132 % Last updated: $Date: 2009/03/15 18:29:58 $
133 % Dan Ellis <dpwe@ee.columbia.edu>