We created an original synthetic dialogue with alternating voices and a known turn order. In the real browser tool, with English selected, the result contained the spoken sentences in one paragraph, without Voice A or Voice B labels. It also ended with a blank audio marker that was not part of the authored dialogue. A completed transcript did not perform speaker attribution for us.

We preserved the original generated text before editing. Using the known authored turn sequence, we added neutral Voice A and Voice B labels and separate lines in the editable transcript. We removed the blank audio marker. These were editorial changes, not automatic speaker identification. Copy text returned the edited dialogue. The raw and reviewed files remain separate so readers can inspect what changed.

For your own saved conversation, choose the spoken language and generate a draft, then review each voice change against the recording before adding labels. Do not infer a real name from a voice, a line break, or the wording of a sentence. If attribution remains uncertain, leave it unknown. This tool does not generate speaker labels. Our controlled synthetic example is not a test of every interview or overlapping conversation. The field note includes the source audio and both text versions. Original diagrams and synthetic narration are used here.