I am trying to use the Whisper API to transcribe recordings of word pronunciations with phonetic spellings and/or misspellings included verbatim. Often the API to “too intelligent” and it tries to correct misspellings with other, properly spelled words.
Has anyone successfully used the prompt parameter to feed phonetic information or a list of known misspellings into the API and have it transcribe the misspelled results?
Has anyone used any publicly available word lists or datasets to do something similar?
Thank you.