-
|
Trying to generate a super accurate transcription of exactly what's being said and noticed that if the speaker repeats a word it only shows up as one instance. Is it possible to bypass whatever filtering is happening there? |
Beta Was this translation helpful? Give feedback.
Replies: 1 comment
-
|
The decoding heuristics does not appear to suppress repeated words. So the "filtering" is performed by the model. Typical transcripts done by human omit repeated words. As a result, it has likely learned this from its training data. One way to "bypass this filtering" is to fine-tune the model on data that does not omit repeating words. |
Beta Was this translation helpful? Give feedback.
The decoding heuristics does not appear to suppress repeated words. So the "filtering" is performed by the model. Typical transcripts done by human omit repeated words. As a result, it has likely learned this from its training data. One way to "bypass this filtering" is to fine-tune the model on data that does not omit repeating words.