Why can speech recognition fail in a quiet room?
Quiet surroundings do not guarantee a good input. Microphone selection, capture level, pronunciation, language settings and the recognition system can still affect the result. Start by establishing what audio actually reached the application.
Separate the recording from the transcript
We would first check whether the intended microphone was active and whether the captured speech was complete. A missing first word may come from starting capture late. A clear recording with a wrong transcript points toward a different stage. Without that distinction, changing recognition settings can hide the original problem.
Use a short phrase containing a made-up name and a number, then compare the input with the transcript if the app provides access to both. Repeat with one variable changed. This is a diagnostic method, not a guarantee that a single phrase represents every speaker or accent.