I'm currently working on an audio classification project using CNNs. The problem is I'm having trouble training my CNN. I doubt if there are outliers in my dataset but I don't know how to detect outliers in an audio dataset. I've searched google and found nothing helpful.
-
1$\begingroup$ Tell us what you have found so far and why it's not helpful. Explain also exactly what you're looking for. $\endgroup$– nbroFeb 13, 2022 at 11:12
-
$\begingroup$ @nbro almost everything I found on google was about finding outliers in tabular data, therefore, they're not applicable to audio or any kind of time series data. $\endgroup$– Sepehr GolestanianFeb 21, 2022 at 19:11
1 Answer
A first think what comes to mind is to train an autoencoder, then identify abnormal data by these heuristics:
- Is the reconstruction error large, for example remove the top 5% of the data?
- Are the codes (outputs of the encoder) within a densely populated region, or are they outliers? You could calculate the distance to Nth nearest neighbor.
-
$\begingroup$ Interesting idea, do you know any articles that has done that on audio or any kind of time series data? $\endgroup$ Feb 21, 2022 at 19:12
-
1$\begingroup$ I haven't read any of these papers, but I found plenty from Google Scolar by searching for "autoencoder outlier detection". $\endgroup$– NikoNyrhFeb 22, 2022 at 12:36