I have tried to train the model on my custom dataset. I have done different tries.
1- I make 500 audios of 10 sec with background noise on which positive audios and negative are inserted and trained for 2000 epochs.
2- Then I make 2000 audios and train it for 800 epochs.
I also tried some other things like this but could not get any results. Not even 1%. Either it detects every thing the trigger word or does not detect even the trigger word.
Can you help me with this?