Keep permission and source context together
Select an interview recording you are authorized to process. Record any consent or usage restrictions outside the transcript and keep the original file. Browser processing avoids sending media to a transcription service, but it does not decide whether you may record, quote or publish someone. Read the local-processing explanation for the distinction between media and network downloads. There is no live meeting bot or remote recording integration.
Do not infer a speaker from the text
Transcribana does not identify speakers or provide automatic diarization. It mixes the source audio for recognition, so separate stereo channels do not become reliable speaker labels. Listen to the interview and add labels manually in the working text if your project requires them. With overlapping speech, mark uncertainty instead of guessing. For example, the response “not in April” must be attributed to the right person before it becomes a research quotation.
Decide how faithfully to preserve speech
Keep the raw transcript available while editing. A clean reading copy may remove a repeated word, but a research project may need that repetition recorded. Local English/French writing checks can flag spoken constructions that are intentional. Review suggestions individually and preserve the meaning and tone of a quotation. For French interviews, the French review guide covers homophones, agreement and informal speech without treating the checker as an audio verifier.
Export the right record
Copy Text and TXT follow the selected original or working document. SRT, VTT and JSON keep original recognized segments and estimated times, so manual labels or corrections in TXT do not appear in those files. Keep the source with JSON when another workflow needs to trace passages; do not treat estimated times as certified evidence. Check any subtitle export in its destination editor. Download your work before navigation: the application does not maintain an interview archive or project history.