On February 17, we released our improved keyword boosting feature which can be used to boost or suppress recognition of certain keywords. We have determined that some words that are unfamiliar to the speech model may be more sensitive to boosting than a known keyword that receives the same boosting factor. In the example below, “FancyFries” currently could have a higher likelihood of appearing than “burger”, even though the boost factor is the same, “2”.
Known keyword – “keywords=burger:2”
Unfamiliar keyword – “keywords=FancyFries:2”
This is a bug and we are working to fix this issue so the boost factor works similarly for known or unknown keywords. In the meantime, if you notice certain keywords that seem to push other words out of the transcript, please lower the boost factor for that keyword.
To fall back to previous keyword behavior append the query parameter keyword_boost=legacy to your request.
Please contact us at support@deepgram.com if you have any additional questions.