I want to save wakewords on a satellite so i can use them for retraining a model and gathering false positives. I have seen a python script from Daenara which works fine if i am using mqtt for all wakewords but not when it is doing UDP streaming as the wakeword is detected locally.
Is there a way to access these sound files? or some way to save them ? i had planned to mount a folder on my pi in the profiles.json folder so it is easily accessible from the docker container.
Next step would be to move them into a specific folder based on if i say “Stop | Cancel” due to a false positive. which would help with the organisation.
Not sure what exists but there is a certain logic to ASR where you can pair KW and command sentence.
Often you don’t get a “Stop” or “Cancel” as the command is not recognised or processed and its likely a secondary kw+command is given.
So you can pretty much assume that a KW and command with a stop or cancel is false and also ones that don’t complete a command via ASR.
Likely completed ASR commands that don’t have instantaneous follow up can definately be considered good.
Also !KW are equally as important and the same goes as good completed commands authenticate the words and if not then to save with meta-data or naming of the words has much value.