Snore pre-detection: processing window size effects

Snoring is a problem that affects a large part of the population and is usually associated with or indicative of other diseases, several of them are linked to sleep disorders such as apnea. Tools that automatically identify snoring, providing information to the patient and to the medical professional are very useful since it is possible to identify the problem and follow the progress of someone undergoing anti-snoring treatment. One of the tasks that this tool must perform is the identification of sound events. One technique that can be used for this involves the use of an adaptive threshold over a predefined length of the audio itself. This paper aimed to identify the behavior of the sound events identification (snoring and non-snoring) by varying the length of the audio chosen to calculate the threshold. When using small windows, a large number of events are detected. This value drops exponentially by increasing the size of the window until it reaches a plateau. Furthermore, not only the number of events that are not snoring are impact, but even the number of real snoring, which leads to the loss of the solution’s sensitivity.