Custom wake word with few samples

[roshan2024nar]

I have sample of a wake word its like 100 files of 1-2 sec of that word how do i train a model can i combine this with automatic one for better performance its an Indian word so which model should i prefer ??

[roshan2024nar]

Sorry for your loss, Take care.

[roshan2024nar]

hey is everything alright, can we discuss this if possible

[muhammadabidPro]

Okay, so you still haven't been able to build a good wake word model. You can try DNN or RNN architecture. DNN is very fast and RNN is better than DNN in terms of accuracy, but DNN model training is fast. You need to increase your positive data. And you should take a good look at this adversial_texts file. There is a generator below that is not being used. I will update it later. You can try it yourself. Feel free to ask anything. You can also join our discord server.

[roshan2024nar]

so i have tried the automatic training and it has a lot of false detects what i need is i have 400 samples of 1-2 sec in mp3 send on WhatsApp do i need to do some pre-processing in this for training should i add syntactic plus this can you guide me