From the command line:
git clone --recursive https://github.com/Noura/laughter-properties
cd laughter-properties
touch laughter_detection/__init__.py
touch pywebrtcvad/__init__.py
You need to make sure all the audio files you want to process are in .wav format, mono, sample rate at 8000, 16000, or 32000, and sample width 2. If you need to convert the audio files to this format, you can use VLC, which you can download for free from here.
In VLC, go to File -> Convert/Stream
Under Choose Profile, choose Custom from the dropdown then press the button Customize...
In the popup menu,
under Encapsulation choose WAV
under Audio codec choose Codec: WAV, Channels: 1, Samplerate: 8000
Click Save as new Profile..., I suggest naming it something descriptive like wav mono sr8000.
Now every time you want to convert a file you can choose this profile.
Make sure all the audio files to .wav format. If needed you can conver them with VLC.
Launch VLC, go to File -> Convert/Stream.
Add the audio file you want to convert.
Under Choose Profile, choose wav mono sr8000 or whatever you named your custom profile.
Choose an output destination and name for the .wav file.
Put all the .wav files that you want to have processed together in the same folder. We'll refer to this folder as input_directory
Launch a terminal session in the folder laughter-properties
From the command line:
python process_laughs.py <dir> <threshold> <aggressivness>
Where input_directory is the absolute path of the folder containing the .wav files, and your results will go here too
<threshold> adjusts the minimum probability threshold for classifying a frame as laughter. The default is 0.5, but you can experiment with settings between 0 and 1 to see what works best for your data. Lower threshold values may give more false positives but may also recover a higher percentage of laughs from your file. (quoted from detection/README.md)
<aggressiveness> parameter adjusts the voice extraction. It is an integer between 0 and 3. 0 is the least aggressive about filtering out non-speech, 3 is the most aggressive. (quoted from pywebrtcvad/README.rst)
Navigate to input_directory and open results.txt
"relative loudness" is roughly the loudness of the laughter snippet divided by the average loudness of the whole audio recording. So, this tries to take care of the differences in overall loudness of different audio recordings.