Getting the best speaker labels

Speaker labels are only as good as the audio they're built from. A few habits at recording time make a bigger difference than anything you can fix afterward — and knowing what to clean up after the fact covers the rest.

Last updated 2026-08-31

While you're recording

Put the mic in the middlePhone or recorder in the center of the table, roughly equidistant from everyone. A mic that's close to one person and far from another hears them at very different volumes, which makes voices harder to tell apart.
One voice at a timeSpeaker labeling works best when people mostly take turns. When two people talk over each other, the tool has to guess which voice owns which words in the overlap — and it often gets it wrong. Ask the room to avoid interrupting where you can.
Watch the room, not just the micA hard-surfaced room — glass, tile, bare walls — bounces sound around and blurs the line between voices. A carpeted room with soft furniture records noticeably cleaner than a conference room with a glass table.
Distance hurts everyone equallySomeone leaning back from the table, or a large table with the mic at one end, pushes a voice below the noise floor. If your room is bigger than a normal table, that's when a second mic or a dedicated kit starts to pay off.

Why "phone in the middle" works

A phone's built-in microphone is small and unremarkable, but it's good enough for this — the thing that actually matters is position, not the hardware. Centering it means every voice reaches it at roughly the same level, which is exactly what a speaker-labeling model needs to tell voices apart. For a table of 2–6 people in a normal room, phone-in-the-middle is the whole trick.

When dedicated mics help

Once a room gets big, loud, or echo-y — or you're recording something more like a podcast than a meeting — a single centered phone starts to strain. Wireless podcast-style kits (the small clip-on mic + receiver setups you see on YouTube, from DJI, Rode, and similar) give each speaker their own close mic, which sidesteps distance and room-echo problems entirely and makes overlapping speech far easier to separate. They're overkill for a two-person call at a kitchen table, but worth it for a panel, an interview in an unfamiliar space, or anywhere crosstalk is unavoidable.

Overlapping speech is hard for any tool

Be honest with yourself about this one: no speaker-labeling system, including LymeScribe's, reliably untangles multiple people talking at once. When two voices genuinely overlap, the transcript may attribute the overlap to the wrong speaker or merge it into one line. The fix is upstream — turn-taking habits and mic placement — not downstream. A transcript from a well-behaved conversation will always beat post-processing tricks applied to a messy recording.

Cleaning up after transcription

  1. Name speakers. The labels come out as Speaker 1, Speaker 2, and so on — LymeScribe plays a short clip of each voice, you type the name, and the whole transcript updates. Do this once per person and it applies everywhere they speak.
  2. Corrections dictionary. If a name, term, or phrase keeps coming out wrong across your transcripts — a colleague's name, a product name, an acronym your team uses — add it to your corrections dictionary once and it's applied automatically on future transcripts, no repeated manual fixing.

Between good recording habits and these two cleanup steps, most transcripts need little to no manual correction by the time you're ready to hand them to an AI action — see prompt recipes for transcripts for what to do with the result.

Ready to try it? Download LymeScribe — speaker labels are part of the paid hosting unlock, running entirely on your own machine. Questions? Email us — a real person answers.