Separate the layers
If text also fails, return to AI diagnosis. If text works, the model connection is alive and the remaining problem is input capture, speech recognition, TTS, volume, or an optional voice add-on.
Language boundary
Do not interpret 'AI supports all languages' as proof that voice does. Test English with a simple phrase before debugging another language, and report the exact speech direction that fails.
Privacy
Microphone and third-party voice providers may process audio. Review their policies and avoid uploading recordings or keys to MMB Guide; this site does not request microphone access.