Since you dev's said "options we have found so far are financially prohibitive and we do not want to pass those costs on" there are multiple options you guys can use. I would recommend Whisper from Openai, it works very well. However you might have to script it to work live as it mainly used for pre-recorded mp3's etc. Aswell theres a free verison, but the api cost money.
I hope you guys can look into this as well desperately need a better speech rec system for london control.
https://openai.com/index/whisper/
