search

bpwhelan / GameSentenceMiner

This is (was) a simple project, that watchdogs a folder for a recording (I use OBS with "Replay Buffer"), converts it to audio, trims it according to clipboard and a Local Language Model before adding it to a flashcard.
7 stars 1 forks source link

Improve audio matching #1

Closed bpwhelan closed 1 week ago

bpwhelan commented 1 week ago

Azure matches different than i expected, so sometimes it gets the line before or after, but the match is still above my threshhold, one thing is i could just make sure it contains the term instead of doing arbitrary comparison.

bpwhelan commented 1 week ago

Another idea would be to monitor clipboard for time, and based on the time of the audio's creation trim everything before, and then just grab all the audio from that, should be easy

bpwhelan commented 1 week ago

Seems to be done and working much better than before, checks the last clipboard for the start, and then azure for the end of "voice"