File: Tracksplit2.txt

package info (click to toggle)
gramofile 1.6-4
  • links: PTS
  • area: main
  • in suites: woody
  • size: 820 kB
  • ctags: 536
  • sloc: ansic: 10,238; makefile: 55
file content (139 lines) | stat: -rw-r--r-- 6,711 bytes parent folder | download | duplicates (7)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139

In version 1.4 of GramoFile, a entirely new track location algorithm is
implemented, which will be briefly discussed here. The various values
mentioned are the defaults offered in GramoFile 1.4 (and higher). Also, a
procedure is described that you may follow to get more or less optimal
track location.


TRACK LOCATION

First of all, a `power profile' of the sound file is created, by taking
the Root-Mean-Square (`RMS') of 4410 samples of the signal. We get
something like: 


                                            |     |
             |     ||                   ||  ||  |||             ||  |  ||
       |  | || |||||||                 ||||||||||||||           ||| ||||||
      ||||| ||||||||||||              |||||||||||||||||         ||||||||||
      ||||||||||||||||||              |||||||||||||||||        |||||||||||
      |||||||||||||||||||            ||||||||||||||||||       ||||||||||||
     |||||||||||||||||||||          ||||||||||||||||||||      |||||||||||||
     |||||||||||||||||||||||     |||||||||||||||||||||||     ||||||||||||||
 |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||

  ^            ^              ^             ^             ^        ^        ^
 silence     track         silence        track        silence   track silence


The RMS-computation takes quite some time. Therefore the (binary)RMS-data
may/will be saved to a `.rms' file. This .rms file contains all
information necessary to locate tracks (the .wav file is not needed any
more). 

To beautify the data a little, a running median (length 3) is applied. The
above graph gets a little smoother by doing this. The result of this
operation may be saved to a .med `graph file', that shows a rotated (and
much longer) version of the graph above.

Then, a first coarse track location will be applied. This is simply
walking through the `.med data' and finding sequences that are either
above (= possible track) or below (= possible silence) a certain
threshold, the `global silence threshold'. 

This global silence threshold is different for each sound file. To
`automatically' find a reasonable threshold, first a minimum threshold and
a maximum threshold are computed. The minimum threshold is (about) the
lowest `.med value' in the data set. The maximum threshold is the middle
`.med value' (that's usually very close to the mean `.med value'). To find
these, the `.med data' is sorted, and we get something like this: 


 min. thresh.                     max. thresh.                               |
    v                                  v                                    ||
                                                                          ||||
                                                                        ||||||
                                                                    ||||||||||
                                                ||||||||||||||||||||||||||||||
                            ||||||||||||||||||||||||||||||||||||||||||||||||||
               |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
           |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
 |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||

      ^                                        ^
 collected silence                           music


The minimum silence now is the 10th element in the sorted data, and the
maximum threshold is the middle element. The sorted data may also be saved
to a .sor `graph file'.

When these min and max thresholds are found, the real global silence
threshold is found by taking a value between the min anc max threshold. 
Default, that value lies on 15% (= 150 * 0.1%, which seems to work well in
most cases), as shown below:


                        value `Global silence factor'
                            in Parameter-screen
                                     |
                                     V

            maximum threshold      1000     100% --
                                                 --
                                                 --
                                                 --
                                                 --
                                                 --
                                                 --
     ->  global_silence_threshold   150      15% --
                                                 --
            minimum threshold         0       0% --


When the possible tracks and silences have been located, they are
`transformed' into certain tracks / silences by applying a few `rules'. 
The most important one is that possible tracks longer than 50 blocks (= 5
seconds when using 4410-sample blocks) are certain tracks, and possible
silences longer than 20 blocks (= 2 seconds) are certain silences. For
more information, see the funtion tracksplit_findtracks() in the file
tracksplit.c. 

To detect the track starts/ends more correctly, new local thresholds are
computed, by taking the mean of the .med's within the individual silences
and adding 5% on top of that (the `Local silence factor'). The tracks
adjecent to a silence are now extended using the local threshold.

Finally the detected `certain tracks' are extended a bit by adding 3
blocks (0.3 secs) before and 6 blocks (0.6 secs) after the detected track.
This is to account for fade-in/fade-out effects, that are hard to detect.

When all this is done, track location is completed. The detected location
of the tracks is written to the .tracks file. 


PROCEDURE

The best way to use track location is this:

 - First, try track location with the default parameters.
 - If too few tracks are detected, check if the `minimal length of inter-
    track silences' is correct. Some records require 10 (blocks).
 - If inter-track silence was not the problem, try to increase the `global
    silence factor' to for example 250 or 350. Really weird recordings may
    need even higher values, but 1000 should really be enough.
 - It is generally _no_ problem if 2 or 3 tracks too much are detected.
 - When you're satisfied with the automatic detection, you may fine-tune/
    adjust the starts/ends manually, using a text editor and the `Play'-
    `Track' option. You may have to join some tracks, but that won't be
    too difficult.


If you encounter a situation in which the track splitting (still) doesn't
function properly, you can contact me about it. You may include the .rms
file of the .wav you're having problems with, and describe your problem as
clearly and completely as possible. Also, corrections, additions,
suggestions and compliments are heartily welcomed.  Contact information is
in the general `README' file.