Open JRMeyer opened 2 years ago
This feature would be great! My application already has an "expected" transcript and merely needs to verify it using SST. Currently, there seems to be no way for Coqui to benefit from this expected transcript. I tried boosting all individual words from the expected transcript, but that didn't give any useful results.
The ability to boost entire phrases is exactly what I need. Especially if this mechanism still works if the actual phrase is similar to, but not identical with, the boosted one. For instance, boosting "don't smear the colors" should prevent Coqui from recognizing "sneer" instead of "smear" not only in recordings of this exact phrase, but also in recordings of similar phrases like "don't smear these colors".
currently you can't boost phrases, only words split on whitespace. We should be able to boost phrases, too