Hi @aurel
I have sorted out the 100% correct/incorrect problem… recent results attached picture, I am not really sure what do you mean by the “audio augmentation”, found “data augmentation”
have a look 
BR.
Yes I meant Data Augmentation
It is especially useful with small datasets.
Aurelien
Great, I wanted to use this before, however, whatever I try to do with this function, it seems I cannot use it, it is , the text, in grey color, any ideas how to sort this out?
BR.
You need to tick the checkbox first to get access to Data augmentation options. I just enabled it on your project.
Aurelien
Thank you, nice! There was no tick box next to it, looking at this “Warp time axis” … found it, enabled, firefox shows this very distinctively
Apparently, works good, attaching current results…
BR.
Hello @aurel , although the model should perform better, I do not have such a feeling.
Would you please help me out with understanding the whole concept.
Project ID 33093
Indeed, I get a rather good results:the 80% or above , which is very good!
Would that in theory mean that 8 out of 10 commands should work?
The live classification, when loading a sample, in general works ok…
But when testing, namely exporting c code and using it on the Dev. board, the results are not evident to be so correct, for example. A good test is running the inference eg. $ edge-impulse-run-impulse …
Indeed, more data is essential, basic… and I am working on it, but is there anything else that could be done?
Regarding more data, found the corpus of data, in English, French, German, however, can such data (22k words per language) be uploaded with using the .tsv file for classification/labeling. I will never be in a position to label such an amount of data
Any ideas how to get such a pile of data across to the EI?
BR.
Hi @aurel
added different, but more data…
Project ID 33093
please have a look,
Getting very low accuracy, while some of the files have a warning, different freq? Us tgis sample rate question?
Best.
Hi @mu234,
I answered to your email. You would need to crop your data samples to keep the keyword part only, otherwise your keyword class will also contain windows with noise/random data.
Aurelien