Hi
i would like to "concentrate" the recognition on certain words which are already part of the english/german models.
E.g the word "Whatsapp" is often recognized as "what's up". Its better for me to get nothing returned than "what's up".
I know that I can replace 'what's up' with 'Whatsapp' in postprocessing, but this defies the purpose.
I found in previous issues the following suggestions e.g. :
KaldiRecognizer(model, 16000, "zero oh one two three four five six seven eight nine whatsapp")
A. to get a word e.g. "whatsapp" more "frequent" in my recognition result as it's specifically specified without the need to finetune the model
B. never the word "what's up" in my results as i didn't specify it in the set of words
C. better performance (?)
D. load the same model only once and only pass the words "to be concentrated on" to the KaldiRecognizer function with every request
But in my case i get all possible words of the model recognized instead of the provided only.
It's pretty much the same as if i didn't provide any words at all. What am i doing wrong here?
Hi i would like to "concentrate" the recognition on certain words which are already part of the english/german models. E.g the word "Whatsapp" is often recognized as "what's up". Its better for me to get nothing returned than "what's up". I know that I can replace 'what's up' with 'Whatsapp' in postprocessing, but this defies the purpose.
I found in previous issues the following suggestions e.g. :
KaldiRecognizer(model, 16000, "zero oh one two three four five six seven eight nine whatsapp")
and
KaldiRecognizer(model, wf.getframerate(), '[ "whatsapp", "[unk]" ]')
So how i understand, this way you only get one of the words provided or nothing: https://github.com/alphacep/vosk-api/issues/107#issuecomment-756640282
This way i was hoping
A. to get a word e.g. "whatsapp" more "frequent" in my recognition result as it's specifically specified without the need to finetune the model B. never the word "what's up" in my results as i didn't specify it in the set of words C. better performance (?) D. load the same model only once and only pass the words "to be concentrated on" to the
KaldiRecognizer
function with every requestBut in my case i get all possible words of the model recognized instead of the provided only. It's pretty much the same as if i didn't provide any words at all. What am i doing wrong here?