In this project, I was tasked with cleaning hundreds of spoken excerpts of Dementia Patients' speech data.

The structure of the recordings was a sequence of words spoken with large gaps between each word, with the occasional remark from the interviewer. For such a seemingly simplistic problem, many existing silence detecting solutions I tried did not work. Hence, I decided to build my own, and also explore how silence detection works.

I used the python package librosa to conduct the audio analysis. Through my experimentations with this Jupyter Notebook, I fixed DC offsets and used the RMS of the audio signal to calculate when noise occured to detect the silences.