In my opinion, the problem is running what should be locally based processing over the internet, in the cloud. Most of my system now runs on the Hubitat platform, and no longer encounters the issues you describe.
SHM Delay is stabilized, however feel free to modify the logic and code. A modified BuzzerSwitch module, with some virtual switch devices should easily handle this. Additionally, the SHM Delay Talker module already handles the voice messaging for the conditions defined above.
Regarding the “spotty” TTS messages: Should you be using an Amazon tablet with Fully and LanNouncer, the Amazon OS does not play well with LanNouncer and I suggest using the Fully Kiosk Browser DTH for TTS. Also on my Amazon tablets, TTS improved after totally disabling Alexa.