Precise: multiple wake word detection with probabilities

I need a keyword spotting with multiple possible keywords / wake words / hotwords. For all possible keywords in a list (e.g. 5) I need the probability that they were uttered in a given audio sequence. This should work with a neural network as it is used in mycroft because we can put a softmax on the last layer and get probabilities. But I could also find the functionality for training your own model for one wake word. Is there the possibility to work with a list of possible keywords, getting their probabilities?

This one is probably a great question for @Wolfgange or @Mn0491.

Hi, Precise currently doesn’t support classification between multiple wake words. At the moment you can either train multiple different models and run the engine once for each wakeword. If you would like to use less CPU and build a classification model, it is possible to modify Precise to work that way, but it will take a little bit of effort. If your interested in doing the changes to make that work, we can talk on mattermost (@matthew-mycroft) about it.

Let me know if you have any other questions :slight_smile:

1 Like