search

rhasspy / wyoming-vosk

Wyoming protocol server for the vosk speech to text system
MIT License
10 stars 5 forks source link

Please add an extended description for some sections #7

Open mitrokun opened 7 months ago

mitrokun commented 7 months ago

From the position of an ordinary user🙄

After reading the description, a few questions remain.

What is the range of values for the "correct-sentences" parameter?

And the main misunderstanding is caused by the "No Correct Patterns" section. The meaning of the parameter is well explained, but it is completely unclear how to use it. It would be great to add an example of the actual use of this option. I tried to search on Google, looked at the available examples and problems related to vosk, вut I didn't find an answer anywhere.

//Looks like I finally figured it out. The first parameter is linked to error estimation.

And in the second case, it is enough to specify something like this if we want to have a request to Wikipedia no_correct_patterns:

roschi02 commented 3 weeks ago

What is the range of values for the "correct-sentences" parameter?

I have the exact same question. Did you ever figure it out?

synesthesiam commented 3 weeks ago

I've added some more to the README, but I'll reproduce here:

See description the score_cutoff parameter in the rapidfuzz docs for more details (weights=(1, 1, 3)).

So the range of the "correct-sentences" parameter depends on the size of the strings that are being compared. It's effectively the maximum number of "changes" that can be made try and turn the text that vosk gives back into one of the pre-defined sentences. I set the weights to (1, 1, 3) so that adding/removing a character is less expensive than changing an existing one.

roschi02 commented 3 weeks ago

Do I really need to run the script, or can I just change the values in the Vosk configuration? And what exactly am I changing there? I can only enter one number, not a series of numbers. image