Separate the reference, transcript, and your recording before identifying a difference.
A concrete example
Illustrative text only: the reference says そういうわけではないんですが, while you may have heard わけは. This is not a recording transcription.
What a good result looks like
Keep three fields: reference transcript, what is actually audible in your recording, and the possible missed では. If the recording is unclear, mark it uncertain. Then repeat listening and shadowing in two rounds.
A prompt you can use
Find the listening question and my latest shadowing recording in JLPT Master. Only if audio:read is authorized and you can process both audio files, compare what you actually hear with the full transcript, identify specific differences, and suggest two listening rounds. Mark any inaudible part as uncertain; do not invent a transcription.
Three practical steps
Confirm the exact question, reference audio, full transcript, and learner recording.
Check audio:read permission and whether your client actually received and processed the audio.
Keep observed sounds separate from guesses, then replay the original question.
Where to check the result
Review the analysis with the recording. The full transcript belongs to the shared audio, while per-question notes explain that question.
Open listening practice →What to watch for
An audio link is not automatically a transcript. An AI that did not hear the recording cannot claim it analyzed your pronunciation.
JLPT Master