Piano Evaluation for Level Normalisation » History » Version 38

Chris Cannam, 2014-07-23 02:07 PM

1 1 Chris Cannam
h1. Piano Evaluation for Level Normalisation
2 1 Chris Cannam
3 1 Chris Cannam
Lack of normalisation for Vamp plugin inputs is a problem when analysing quiet recordings (see #1028).
4 1 Chris Cannam
5 1 Chris Cannam
Testing using a small set of piano recordings, quickly evaluating performance across the first 30 seconds under a number of different normalisation / level management regimes.
6 1 Chris Cannam
7 3 Chris Cannam
h3. Input files
8 1 Chris Cannam
9 1 Chris Cannam
|Filename|Signal max approx|
10 4 Chris Cannam
|@31.wav@|0.57|
11 4 Chris Cannam
|@MAPS_MUS-bach_846_AkPnBcht.wav@|0.12|
12 4 Chris Cannam
|@MAPS_MUS-chpn_op7_1_ENSTDkAm.wav@|0.33|
13 4 Chris Cannam
|@MAPS_MUS-scn15_7_SptkBGAm.wav@|0.13|
14 4 Chris Cannam
|@mz_333_1MINp_align.wav@|0.10|
15 2 Chris Cannam
16 2 Chris Cannam
The plugin has one internal threshold parameter, which can be lowered to find quieter notes (at the expense of course of more false positives). We don't really want to expose this (or any continuous controls) as a parameter. But we need to have approximately predictable input levels, for this threshold to be meaningful.
17 2 Chris Cannam
18 3 Chris Cannam
h3. Methods
19 2 Chris Cannam
20 2 Chris Cannam
|Name|Hg revision|Description|
21 25 Chris Cannam
|@norm@|commit:d721a17f3e14|Normalise to 0.50 max before running plugin (can't do this in plugin: it's here as the reference case)|
22 4 Chris Cannam
|@as-is@|commit:d721a17f3e14|No normalisation|
23 4 Chris Cannam
|@to-date@|commit:d9b688700819|Track max signal level _so far_, adjust each sample so that max is at 0.50|
24 28 Chris Cannam
|@r2@,@r3@,@r4@,@r5@,@r6@|commit:b5a8836dd2a4|Preprocess with "Flatten Dynamics":/projects/flattendynamics at 0.02, 0.03, 0.04, 0.05, 0.06 target RMS levels respectively|
25 29 Chris Cannam
|@s8@|commit:4ac067799e0b|With "Flatten Dynamics second attempt":/projects/flattendynamics/wiki/Wiki with max RMS targeted to 0.08|
26 32 Chris Cannam
|@t4@|commit:d67fae2bb29e|With Flatten Dynamics attempt 2a with max RMS targeted to 0.04|
27 37 Chris Cannam
|@u4@|commit:70773820e719|With Flatten Dynamics attempt 2b with max RMS targeted to 0.04|
28 32 Chris Cannam
29 3 Chris Cannam
h3. Results
30 3 Chris Cannam
31 3 Chris Cannam
Reporting only the note onset F-measure for the first 30 seconds of each piece.
32 1 Chris Cannam
33 34 Chris Cannam
|Filename|@norm@|@as-is@|@to-date@|@r2@|@r3@|@r4@|@r5@|@r6@|@s8@|@t4@|@u4@|
34 34 Chris Cannam
|@31.wav@|50|33|40|45|47|48|45|43|42|49|45|
35 34 Chris Cannam
|@MAPS_MUS-bach_846_AkPnBcht.wav@|87|15|62|64|85|87|87|86|81|86|87|
36 34 Chris Cannam
|@MAPS_MUS-chpn_op7_1_ENSTDkAm.wav@|33|31|31|11|25|31|32|31|32|34|35|
37 34 Chris Cannam
|@MAPS_MUS-scn15_7_SptkBGAm.wav@|73|16|61|50|57|67|74|75|70|69|68|
38 34 Chris Cannam
|@mz_333_1MINp_align.wav@|66|3|58|42|60|64|66|63|66|63|65|
39 7 Chris Cannam
40 10 Chris Cannam
The precision (_proportion of correct onsets among detected onsets, or 1 minus the false-positive rate_) and recall (_proportion of correctly-detected onsets among all ground-truth onsets, or true-positive rate_) vary as you would hope:
41 10 Chris Cannam
42 10 Chris Cannam
 * when the resulting audio level is quieter than the @norm@ case, precision is high and recall is low but the F-measure is worse than the @norm@ case
43 10 Chris Cannam
 * when the resulting audio level is louder than the @norm@ case, precision is low and recall is high and the F-measure is still worse than the @norm@ case
44 10 Chris Cannam
45 12 Chris Cannam
This suggests that our threshold (which happens to be 6) is moderately well-suited to the @norm@ case, at least to optimise F-measure (this might not be the most perceptually useful measure though).
46 13 Chris Cannam
47 38 Chris Cannam
The best results (apart from @norm@) above seem to be @r5@ and @u4@. Let's try to refine the parameters for each of those and see if any patterns emerge.
48 38 Chris Cannam
49 38 Chris Cannam
h4. Flatten Dynamics fine-tuning
50 38 Chris Cannam
51 14 Chris Cannam
h4. For different piano template sets
52 14 Chris Cannam
53 17 Chris Cannam
The above results are all generated using four piano templates, numbered 1-3 plus @pianorwc@.
54 17 Chris Cannam
55 17 Chris Cannam
Here are results using the @norm@ and @as-is@ methods, but with different sets of piano templates: first with three templates (1-3) and then with each template in turn as the only one.
56 17 Chris Cannam
57 19 Chris Cannam
The template turns out not to make an enormous difference -- perhaps because these recordings contain nothing but piano?
58 13 Chris Cannam
59 13 Chris Cannam
|Filename|@norm@/all|@as-is@/all|@norm@/3of4|@as-is@/3of4|@norm@/1|@as-is@/1|@norm@/2|@as-is@/2|@norm@/3|@as-is@/3|@norm@/rwc|@as-is@/rwc|
60 22 Chris Cannam
|@31.wav@|50|33|51|30|50|34|44|42|50|32|56|36|
61 22 Chris Cannam
|@MAPS_MUS-bach_846_AkPnBcht.wav@|87|15|86|16|86|24|75|20|73|10|71|18|
62 22 Chris Cannam
|@MAPS_MUS-chpn_op7_1_ENSTDkAm.wav@|33|31|32|32|31|22|29|31|35|34|32|28|
63 22 Chris Cannam
|@MAPS_MUS-scn15_7_SptkBGAm.wav@|73|16|71|19|71|12|68|14|72|17|70|15|
64 22 Chris Cannam
|@mz_333_1MINp_align.wav@|66|3|68|1|63|4|67|2|67|1|63|3|
65 20 Chris Cannam
66 20 Chris Cannam
h4. For "generic" template set
67 20 Chris Cannam
68 20 Chris Cannam
The above results all use template sets with only piano templates in them.
69 20 Chris Cannam
70 20 Chris Cannam
Here are results using the @norm@ and @as-is@ methods, but with the full set of instrument templates (four pianos plus all the rest).
71 21 Chris Cannam
72 1 Chris Cannam
|Filename|@norm@|@as-is@|
73 1 Chris Cannam
|@31.wav@|49|37|
74 1 Chris Cannam
|@MAPS_MUS-bach_846_AkPnBcht.wav@|79|34|
75 1 Chris Cannam
|@MAPS_MUS-chpn_op7_1_ENSTDkAm.wav@|31|28|
76 1 Chris Cannam
|@MAPS_MUS-scn15_7_SptkBGAm.wav@|67|16|
77 1 Chris Cannam
|@mz_333_1MINp_align.wav@|63|5|
78 34 Chris Cannam
79 34 Chris Cannam
h4. Cross-checking with non-piano test data
80 34 Chris Cannam
81 34 Chris Cannam
The results need to be roughly comparable with those obtained from pre-normalised data using other datasets as well as the piano one. Here is a subset of the TRIOS dataset. The @norm@ result is that obtained from the plugin prior to doing this work, using pre-normalised data.
82 34 Chris Cannam
83 36 Chris Cannam
The @mirex@ result is that from the MIREX 2012 submission in MATLAB, but note that this always uses all instrument templates while the plugin results are based on selecting the "right" instrument for the piece (which is assumed to be the best, though we aren't actually testing that here).
84 35 Chris Cannam
85 35 Chris Cannam
|File|@mirex@|@norm@|@u4@|
86 35 Chris Cannam
|mozart/piano|60|64|56|
87 35 Chris Cannam
|mozart/viola|33|37|35|
88 35 Chris Cannam
|mozart/mix|51|58|55|
89 35 Chris Cannam
|mozart/clarinet|74|80|86|
90 35 Chris Cannam
|lussier/piano|45|52|63|
91 35 Chris Cannam
|lussier/mix|36|43|40|
92 35 Chris Cannam
|lussier/bassoon|43|75|80|
93 35 Chris Cannam
|lussier/trumpet|43|46|51|
94 35 Chris Cannam
|take_five/piano|61|46|69|
95 35 Chris Cannam
|take_five/mix|62|73|69|
96 35 Chris Cannam
|take_five/saxophone|78|80|84|