Longer Phrases and Custom Personal Data #208
Unanswered
gordon1213
asked this question in
Q&A
Replies: 0 comments
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
-
Hello!
I'm currently trying to create a model that is able to address multiple different accents and potential different pronunciations for my given phrase. There are some issues that I need to address to increase True Positive activation such as duration that it takes for the phrase to be spoken and not activating based on certain assumed pronunciations.
I'm worried about a target audience that takes longer to say the phrase and have addressed part of this issue by adding ellipses or underscores (mainly underscores) between syllables/phonemes of the phrase and have had some success, but I noticed that adding more underscores or ellipses consecutively does not extend the synthetic generation clip (made this conclusion using your simpler notebook that generates an example clip of the phrase). Is there a way to extend the phrase further? I've also noticed a post in the past where you adjusted clip size in the past and was wondering if extending it was still a possibility.
I noticed that there is a section in the yaml file that has supplementation for negative files that reduce false positives, but is there an area that I can add positive files to address false negatives? My first assumption is to add files to the positive training and test folders in the my_custom_model folder for a given phrase.
Thanks in advance!
Beta Was this translation helpful? Give feedback.
All reactions