Acoustic Scene Classification

The goal of acoustic scene classification is to classify a test recording into one of predefined classes that characterizes the environment in which it was recorded — for example “outdoor market”, “busy street”, “office”.

Classification performance is measured using accuracy: the number of correctly classified segments among the total number of test segments.

Metrics

Main functions:

Function sed_eval.scene.SceneClassificationMetrics.evaluate takes as a parameter scene lists, use sed_eval.io.load_scene_list to read them from a file.

Usage example to evaluate files:

Usage example to evaluate results stored in variables:

 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
import sed_eval
import dcase_util

reference = dcase_util.containers.MetaDataContainer([
    {
        'scene_label': 'supermarket',
        'file': 'supermarket09.wav'
    },
    {
        'scene_label': 'tubestation',
        'file': 'tubestation10.wav'
    },
    {
        'scene_label': 'quietstreet',
        'file': 'quietstreet08.wav'
    },
    {
        'scene_label': 'office',
        'file': 'office10.wav'
    },
    {
        'scene_label': 'bus',
        'file': 'bus01.wav'
    },
])

estimated = dcase_util.containers.MetaDataContainer([
    {
        'scene_label': 'supermarket',
        'file': 'supermarket09.wav'
    },
    {
        'scene_label': 'bus',
        'file': 'tubestation10.wav'
    },
    {
        'scene_label': 'quietstreet',
        'file': 'quietstreet08.wav'
    },
    {
        'scene_label': 'park',
        'file': 'office10.wav'
    },
    {
        'scene_label': 'car',
        'file': 'bus01.wav'
    },
])

scene_labels = sed_eval.sound_event.util.unique_scene_labels(reference)

scene_metrics = sed_eval.scene.SceneClassificationMetrics(scene_labels)
scene_metrics.evaluate(
    reference_scene_list=reference,
    estimated_scene_list=estimated
)

print(scene_metrics)
SceneClassificationMetrics([scene_labels])
SceneClassificationMetrics.evaluate(...[, ...]) Evaluate file pair (reference and estimated)
SceneClassificationMetrics.results() All metrics
SceneClassificationMetrics.results_overall_metrics() Overall metrics
SceneClassificationMetrics.results_class_wise_metrics() Class-wise metrics
SceneClassificationMetrics.results_class_wise_average_metrics() Class-wise averaged metrics
SceneClassificationMetrics.result_report_parameters() Report metric parameters
SceneClassificationMetrics.result_report_class_wise() Report class-wise results
SceneClassificationMetrics.result_report_class_wise_average() Report class-wise averages
SceneClassificationMetrics.reset() Reset internal state