Copy the directory to your configuration folder and restart HA.

Add ASR proxy integration.

Enter the data for the primary and fallback STT servers.

The first server can be Speech-to-Phrase, which returns an empty value on a miss, serving as a trigger to retry speech recognition on the backup server.