If you record a two-person interview and run it through a free transcription tool, here's what you'll actually get back: one continuous block of text, no speaker labels, no "Q:" and "A:" — just everything anyone said, in order, unattributed.
This trips people up because it feels like it should be solved. Diarization — automatically tagging "who's speaking" — is a real, mature area of speech AI. But most free and low-cost transcription tools, including Longscribe, don't run it by default, and it's worth understanding why before you're staring at an unlabeled wall of text on deadline.
Why it's not just switched on
Speaker diarization is a separate model pass on top of transcription, not a free byproduct of it — it adds real processing time and compute cost, and it's meaningfully less reliable on phone-call audio, overlapping speech, or a mono recording where both voices are on the same channel (all common in real interview setups). Tools that offer it well usually put it behind a paid tier for a reason: it's the expensive part, not the transcription itself.
The actual workaround
For a two-person interview, this is more tedious than hard: the transcript comes with timestamps, so you scrub through once, add "Q:" / "A:" (or names) at each turn, and cross-check against the audio where the wording matters for a direct quote. It's manual, but for a single interview it's a few minutes of editing, not a blocker.
Where this genuinely gets painful is a panel or roundtable with three or more speakers — at that point, manually attributing every line is real work, and that's when paying for proper diarization (or recording each speaker on a separate track to begin with) starts to make sense.
What I'd actually check before you record
If speaker labels matter for your workflow, the highest-leverage fix happens before transcription: record each participant on a separate channel if your setup allows it (most conferencing tools support this), since a transcription tool can then process each track cleanly instead of guessing from a mixed recording.
Longscribe's interview transcription page has the specifics on formats and export options — I'd rather say plainly that speaker labeling isn't automatic than have someone find out mid-deadline.
Top comments (0)