How to generate "YOUR FEATURES HERE" for MFCC model deployed to the browser?

#1

Are there any sample code on how to generate “YOUR FEATURES HERE” for MFCC model deployed to the browser? There are several related posts but there isn’t any that gives a direct answer with a clear example. For image recognition, it is more straightforward but for audio it seems very complicated. Thanks!

EI_Capture
EI_Capture756×384 32.3 KB

#2

You copy the Features in the EI Studio. Look for the Raw Features copy icon.

This link show an example of Raw Features.

#3

Copying and pasting from EI Studio works fine, no problem there. However, I want to classify from my web application. If I open a clip from my web application, how do I extract the features from the clip and pass these on as inputs to the model for classification? I would like to do this from within the web application (in javascript) .

Thanks again.

#4

I believe this is what your seek.

If not, here is a CPP file that opens the audio card and reads the PCM data into a buffer for inferencing. It also prepends the WAV header and writes the sample to a file, that may be useful when false positives are identified allowing you to direct a re-training mission.

1 Like