Fixed-size-segment Segmentation & Classification

Fixed-size-segment segmentation & Classification is applied on fix-sized segments of an audio recording, and yields a sequence of class labels.

Fixed-size-segment segmentation refers to mtFileClassification() function in audioSegmentation.py

mtFileClassification()

1. mtFileClassification() splits an audio signal to successive mid-term segments (fixed-sized segment) and extracts mid-term feature statistics from each segments, using mtFeatureExtraction() in audioFeatureExtraction.py
2. mtFileClassification() classifies each mid-term segment using a pre-trained supervised model.
3. mtFileClassification() merges successive fix-sized segments, which share the same class label, into larger segments.
4. The statistics of the result of segmentation-classification process can be visualized.

Usage example$\colon$

from pyAudioAnalysis import audioSegmention as aS
[flagsInd, classesAll, acc, CM] = as.mtFileClassification("data/scottish.wav", "data/svmSM", "svm", True, 'data/scottish.segments')

data/scottish.segments (last argument) is used as ground-truth (if available) to estimate overall performance of classification-segmentation method. If data/scottish.segments does not exist, the performance measure is not calculated.

The .segments files are comma-separated format$\colon$ <segment start (seconds)>, <segment end (seconds)>, <segment label>. For example

0.01, 9.90, speech
9.90, 10.70, silence
10.70, 23.50, speech
23.50, 184.30, music
184.30, 185.10, silence
185.10, 200.75, speech
...

Function plotSegmentationResults() plots segmentation-classification result and evaluate the performance of this result (if ground-truth file is available).

Command-line usage$\colon$

python audioAnalysis.py segmentClassifyFile -i <inputFile> --model <model type svm or knn)> --modelName <path to classifier model>

Example$\colon$

python audioAnalysis.py segmentClassifyFile -i data/scottish.wav --model svm --modelName data/svmSM

Figure 1 Segmentation-Classification Result of data/scottish.wav

Figure 1 shows response of commanding code python audioAnalysis.py segmentClassifyFile -i data/scottish.wav --model svm --modelName data/svmSM result.

Figure 1(紅)中 紅色虛線是ground-truth答案，藍色實線Fixed-size-segment Segmentation & Classification的結果。

The overall accuracy of 99.6% is calculated from ground-truth data/scottish.segments.

Figure 1(橙)顯示每個Class在audio duration的比例。

Figure 1(綠)顯示每個Class平均的時間長度。