Great, thanks guys! Is there any way to know if a new sdk has been released without redeploying the model?
SDK releases end up here: https://github.com/edgeimpulse/inferencing-sdk-cpp but it’s not an automated process so we do a sync every two weeks or so to the public repo (export in the Studio always gives you the latest, so this is trailing a bit) - not really a better way of signaling changes. But I’ll let you know when this is released!
1 Like
This has now been updated in the SDK and will be released to production today or tomorrow. Will do a sync to the public repo somewhere this week, but until then just export as C++ Library and you have the latest!
It definitely seems better but I’m not sure the performance is on par with the online version. I’ll do some more testing later today though. Thanks for letting me know!
Hi, it’s not on production yet 
lol, my bad. I take back what I said about it seeming better
Now this has been released 
Hi @jefffhaynes, updates in the MFCC block are live as Jan mentioned. Sample data is now aligned to an even number of cepstrals to produce the same output as in normal classification.
In the Pre-emphasis step we use a sample at the end of the sample buffer as an input coefficient. Since continuous classification uses smaller buffer sizes, different coefficients are used for the pre-emphasis step. Resulting in a small shift in mfcc output, depending on the input signal. In the tests I run this had no effect on the classification results (although the numbers were slightly different). If wanted, this effect can be reduced even more by lowering the pre-emphasis coefficient.
Thanks for all your hard work on this. I need to do some more testing but so far it still seems to be performing behind what is possible online. I’ll try to get some concrete results when I get a minute.
Ok I actually think everything is working. Unfortunately I switched to a new model so I can’t easily go back and compare the previous results. However, I think the only issue is possibly my model being too narrow at the moment. It is definitely performing much better now. Thanks!
2 Likes
Not sure if you made some changes (looks like maybe?) but with averaging turned on performance is now fantastic! Thanks so much for your hard work, this is amazing stuff.
3 Likes
Great! Happy to hear every is working now for you.