Let’s look at the conclusion first

The page shows a sound change, which only proves that the microphone is receiving audio. The AI also needs to match the recognition results with the currently visible document before it moves. First read the entire original sentence near the focus line, and then adjust it item by item. Do not change all settings to the maximum at the beginning.

Troubleshooting steps

1. Confirm that the AI is ready and listening, rather than still preparing, paused, or reporting an error.
2. Start at the beginning of the sentence near the focus line and read two or three sentences in a row; do not start with later paragraphs, bracket prompts, or off-screen content.
3. Try to read aloud according to the wording of the original script. A large number of impromptu rewrites, continuous omission of sentences, or only reciting key words will reduce the reliability of matching.
4. When names, brands or professional words are often misidentified, add a few hotwords and confirm that standard writing exists in the text.
5. Choose "Tolerant" if you often make small rewrites; choose "Strict" if similar sentences lead to misjudgements.
6. Confirm that the target text is located within the current screen or a natural segment that intersects with the screen; for further locations, you need to manually scroll to the vicinity of the screen first.
7. Stay close to the microphone in noisy environments and avoid simultaneous recording from other applications.

still unresolved

Use a short draft with simple content and no repeated sentences to test. If the short draft is normal, the problem most likely comes from the structure or terminology of the script; if the short draft is also abnormal, submit a recent diagnostic session for easy location.