This is the presentation from the demonstration event in 2023.
[Overview]
I developed a program that reports the presence of cough sounds in audio files within a folder on Google Colabration.
[Background and Objectives]
Currently, due to the impact of the novel coronavirus, online consultations using smartphones and PCs have become widespread. Additionally, research and development of cough sound detection systems for monitoring the health of livestock such as pigs are underway. Therefore, I thought it might be possible to create a program to detect cough sounds on the PC and decided to work on the program for the demonstration event.
[System Description]
To perform cough sound detection, we utilized the ‘Short-Time Fourier Transform’.
Computers calculate the difference and similarity of spectrograms, and determine whether it is a cough sound using a threshold.
【Experiment at the Demo Event】
We conducted experiments at the demo event using 11 audio files. The breakdown of the audio files includes 8 cough sounds, 1 voice saying “Hello,” 2 voices uttering “Ah,” and 1 silent audio file.
【Experimental Results】
The figure below shows the output results on the console.
Out of the 11 audio files, 10 were correctly identified. The misclassification occurred with the audio saying ‘Konnichiwa.’
【Development Environment】
Programming Language : Python
Environment : Google Colabration
【Audio Sources Used】
I recorded cough sounds by coughing into a smartphone. In addition to cough sounds, I recorded other sounds using my own voice.
<Sample of the cough sound>
<Sample of my voice>
【References】
Understanding Fourier Transform and Spectrogram in Audio Analysis
Qihttps://qiita.com/shirowanisan/items/6af2cc4c4be0c57bef06ta
The initiation of the ‘Japan’s First’ demonstration experiment on health management of raised pigs utilizing the ‘Cough Sound Detection Technology/SoundTalks TM.