Hi All, I’m following Jan’s steps and able to build the project cloned from GitHub. This code works but when I tried to collect my own audio keyword dataset (5 minutes), create the model and then rebuild the whole project, it doesn’t work, the keyword (microbit) is not recognized, in the best case I get around 0.4 of confidence ! Someone has any idea to help me ? I already spent 2 weeks on that ! Have a nice day ! Julien
EI_IMPULSE_ERROR r = run_classifier_continuous(&signal, &result, false, false);
Alternatively: if you see 0.4 max confidence, but don’t see many false positives you can lower the threshold here (change
Thanks so much @janjongboom for your help! It works now with lower confidence threshold (0.4). I can now go to the weekend with good mood !
It is already ‘false’ originally in your code.
I think there is something wrong with the microbit microphone level, how to increase the microphone sensitivity, I have to be closed enough to make it work (approx 30 cm from the microphone). Is there anyway to increase the microphone sensitivity allowing to keep higher confidence level to avoid false detection ? Have a nice week-end. Julien
@Juph There’s one
false already but you want another one (first false is for debugging, second for moving average filter).
Is there anyway to increase the microphone sensitivity allowing to keep higher confidence level to avoid false detection
Good question, I don’t have the code in front of me, but I’d search the repository for ‘gain’. That’s typically how you control this…