alphacep
How do I get to recognise abbreviations in Vosk ? Does enabling grammar help ?
@sushanthika23 : one way is to create a grammar for single letter recognition. However, I also wonder if there is a better way. E.g., I am non-stop requiring abbreviations like "CPU" and "GPU"; thus, it would be nice if common abbreviations could be recognized directly within the full-language models... :) @nshmyrev, would this be feasible / are there any plans for this?
@omlins I think CPU and GPU should be there and should work fine. Some others are problematic.
We plan to work on post-processing streaming framework some time soon, not there yet.
We plan to work on post-processing streaming framework some time soon, not there yet.
Sounds exciting!
I think CPU and GPU should be there and should work fine.
I just tried with the model vosk-model-en-us-daanzu-20200905 and it does indeed get CPU, but not GPU. "I would like to buy a new GPU today" is understood as: "I would like to buy a new g p you today" "I would like to buy a new cheap you today"
Maybe it would work with vosk-model-en-us-0.22, but there has been this issue that it often adds some "the" in front of word groups. Has this been fixed BTW?
Sorry, if I my question
but there has been this issue that it often adds some "the" in front of word groups. Has this been fixed BTW?
was not clear ;) Searching through VOSK issues, I found that it still seems to be open: https://github.com/alphacep/vosk-api/issues/837
The "the"-issue is not urgent anyways, as long as some other models work fine :)
It is not about clarity, I meant we will need a lot of effort to fix 'the' issue
Nevermind anyway. GPU and CPU should work with recent models. If not, you can add them to LM as described in the docs