Improving Speed for Realtime Applications (On CPU) load, since it seems vosk-api is single threaded?

Hey folks :)

I'm writing a program using the python version of the vosk-api and I have a few questions.

I'm wondering if there's any way to make the recogniser faster? I'm using the large 1.8G model en-us (latest) but sometimes it lags behind the realtime stream from the microphone.

I was thinking I could just throw more CPU cores at it, but it appears that the recogniser is limited to a single thread, or am I mistaken there?

Are there any settings or config tweaks I can use to get a bit more performance out of it? I'm running it on an intel i7 4790k at 4GHz (but do have the potential to run it on a 24 core Xeon system at 2.4Ghz per core, but given it's single threaded I don't think that'll give me any faster recognitions)

I started using the test_microphone.py script, is there anything I can change about it to get just a bit more speed from the recogniser, so it can keep up with real time audio? It's okay if there's only a few words spoken but kinda slows down slower than realtime if there's a long sentence spoken.

I read in another issue that changing the sampling rate to only 16,000Hz should improve speed but isn't test_microphone.py already doing that?

Can anyone point me in the right direction? I've read all the documentation and there's not a decent list of any advanced options or tweaks to use to just give it that bit of a performance boost.

I appreciate any pointers you can give me in the right direction, because I'm a bit lost. Thanks in advanced :)

Kind regards,
Jessica.