Laugh Finder

A machine learning program for human laughter detection.

Setup

1. Install Python 3.4.3 or later
2. Install Python's SciPy package and related NumPy package

• If using Windows, access NumPy and SciPy wheels here
  • Note your version of Python and architecture (i.e., 32-bit v. 64-bit)

3. Install WEKA

Using the Programs

Training

Generate .arff File for WEKA

1. Run main.py under the python-training folder
2. In the GUI, click "Browse" next to "File with laughter audio file names"
3. Select the file containing a series of paths to laughter training files
4. Click "Browse" next to "File with non-laughter audio file names"
5. Select the file containing a series of paths to non-laughter training files
6. Click "Prepare" to generate a wekaFile.arff file within the same directory as main.py

Using WEKA

1. Run WEKA using the following command:

• java -Xmx512m -classpath /location/of/weka/jar/weka.jar weka.gui.GUIChooser

2. In the GUI, click the "Explorer" button
3. Under the "Preprocess" tab, click the "Open file..." button
4. Select the wekaFile.arff file generated earlier
5. Under the "Classify" tab, select "Choose" under "Classifier"
6. To use K-nearest neigbors algorithm: under "lazy", select "IBk"
7. Press "Start"
8. After the classification is finished, right-click the output in "Result List" and click "Save Model"

Testing

1. Run main.py under the python-testing folder

• Include the following params:
  • --phase 0; for testing algorithm
  • --audio /path/to/audio.wav; the audio file you want to examine
    • .wav files only, not 24 bit depth (limitation of SciPy)
  • --arff /path/to/arff/result.arff; the result .arff file from training

Re-Training

1. Change file paths in Constants.java accordingly, relative to the values of your machine
2. Compile and run the WavReader main class (include the weka.jar to your classpath)
3. Run the main Java file
4. In the GUI, select the audio file containig the laughter segments to be identified
5. Click the "Get Laughter Segments" button
6. You will be presented with a series of snippets identified as laughter, select each one that is correctly identified
7. Click the retrain model process
8. The re-trained model will be placed in the same directory as the old model (see Constants.java)

Useful Commands

ffmpeg

If you are using a Macintosh, you can install ffmpeg using Homebrew

• brew install ffmpeg

Using that application, you can do the following:

• Split a wav file into segments

  • ffmpeg -i audio.wav -f segment -segment_time 0.5 -c copy parts/out%03d.wav

• Find the exact duration of a wav file

  • ffmpeg -i audio.wav 2>&1 | grep Duration | awk '{print $2}' | tr -d ,

• Cut for a given duration of a wav file

  • ffmpeg -i audio.wav -ss 00:00:07.100 -t 00:00:01.1 audio.wav

• Convert .mp4 to .wav

  • ffmpeg -i sourcefile.mp4 destination.wav

Reading

• WEKA .arff filetype
• ffmpeg documentation