Hi Daniel, In it's current and future forms, NuPIC could well be used to detect anomalies from the data you mention. Two questions come to mind when reading your email;Choice of PCA over ZCA whitening? and Do you have ECG/EKG data associated with the audio data?
As you may know, Numenta and the NuPIC community have been looking into various medical data anomaly detection. The second question led from some of this work. For example; ECG-derived respiratory signal analysis, tracking of QRS-complex and subsequent R-R peak first/second order derivative anomaly detection (http://www.physionet.org/physiotools/edr/cic85 /). Regards, Richard. On Fri, Jan 1, 2016 at 11:34 AM, Daniel Bell via nupic < [email protected]> wrote: > > > ---------- Forwarded message ---------- > From: Daniel Bell <[email protected]> > To: "NuPIC general mailing list." <[email protected]> > Cc: > Date: Sat, 2 Jan 2016 00:09:53 +1300 > Subject: HTM Applied To Audio > Hello, > > I hope you all had a great Xmas and New Year. > > I am interested in applying the HTM to audio data for sleep studies. > > We have a number of recordings for 'normal' sleep audio and 'abnormal' > recordings (sleep apnea). > > First, we extracted the spectrogram from the training data. The > spectrogram had a 20 ms window size with 10 ms overlaps. The spectrogram > was further processed using PCA whitening (with 80 components) to reduce > the dimensionality. > > Now, what i hope to do is pass the normal recordings into the HTM after > using swarming to get suitable params. Would it be possible to then pass > through the 'abnormal' signals without updating the model, would it detect > the anomalies? > > Regards, > > Daniel > >
