Written September 2026.
Five voices, one laptop mic in the corner, and a transcript with the same problems the audio had. If you've ever tried to record an in-person D&D session off a laptop sitting on the table, you already know it: one mic, picking up whoever's closest to it, at whatever volume they happen to be talking. The GM narrating from three feet away comes through clean. The player at the far end mumbling their plan to the person next to them doesn't come through at all. Crosstalk turns into mush.
None of this is really about the microphone being bad. A laptop's built-in mic is designed to pick up one person talking directly at the screen, not five people scattered around a table, some of them turned sideways to look at a map.
Checks are remembered in this browser. Clear all checks
A good default: figure out which of the three problems you actually have before you buy anything, because only one of them is fixed by picking a mic shape.
People shopping for a “D&D microphone” usually think they're solving one problem. They're solving three, and only the middle one is fixed by picking a mic shape.
Distance is physics, not gear. A voice six inches from a mic arrives dramatically louder than a voice six feet away, on any mic, expensive or free. Nothing you buy changes that. The single biggest fix available before you spend a cent is moving whatever mic you already have closer to the middle of the table.
Pattern is what a mic shape actually buys you. A laptop or headset mic favors whoever's directly in front of it, so the player turned sideways to squint at the map gets quieter for a reason that has nothing to do with distance. An omnidirectional mic doesn't play favorites by angle. Put one in the center and that penalty disappears. Distance is still whatever it is for the seat six feet out; the pattern didn't fix that, placement did.
Turn-taking is free, and it breaks the most transcripts. Two people talking over each other into one mic is the hardest case for any speech-separation engine, on any mic at any price. It gets its own section below, because it matters more than either of the first two.
What tends to work: start with the phone or mic you already own and place it well, before spending on anything new.
Four options, in order of what they cost versus what they fix. We're not putting dollar figures on these, prices move and we haven't run a comparison test across models, but the ranking reflects what each option actually buys.
| Option | Cost | Effort | What it actually fixes |
|---|---|---|---|
| Phone in the middle of the table | Free | Set it down, hit record | Closer to every seat than a laptop mic; turn off any voice-isolation or noise-suppression the recording app defaults to first |
| Small wired or USB mic that sits on the table | Cheap, one-time | Buy once, plug in | An omnidirectional capsule you place yourself, instead of wherever your phone happens to sit |
| USB conference-room mic, disc or puck-shaped | Moderate, one-time | Buy once, plug in | Built for a room of people talking from different directions into one table-top device, this table's exact shape of problem |
| Lav or headset mic per player | Adds up, one per person | Buy several, manage cables or dongles | Solves distance and pattern per seat instead of for the room, at the cost of a bigger setup than most home games want |
One scoping note on the phone row: this means the device actually running your in-person recording mode, in your hand or on the table, not a second phone recording separately elsewhere in the room.
What tends to work: the mic is a one-time decision, but what you do with it every session matters more than which one you bought.
The mic is a one-time decision. What you do with it every session matters more than which one you bought.
Before the first die rolls: everyone, one at a time, say a full sentence into the mic in your own voice, on purpose. Just your name and one thing your character is doing tonight is plenty.
Right before you start: whoever talks quietest reads a line. Whoever talks loudest interrupts them. Then everyone talks over each other on purpose for five seconds. Play it back and listen for hiss, distortion, or mush.
A good default: agree on a turn-taking signal, because crosstalk confuses diarization more than distance or mic shape ever will.
Crosstalk, not distance and not mic shape, is the main reason diarization gets confused. Every habit above helps around the edges of that; this is the one that attacks it directly, and it costs nothing.
Agree on a table habit: let the last word land before the next one starts. For a table that talks over itself a lot, a physical signal beats a rule nobody remembers mid-scene, a raised hand or a token passed to whoever wants in next. It feels artificial for a session or two, then it's just how your table talks, and the transcript is better for it.
A good default: center of the table, away from dice and drinks, wired over Bluetooth.
A good default: expect letters, not names, from a single shared mic, and plan to assign names once yourself.
Here's the honest version, not the hopeful one.
At a physical table with one shared mic, the transcript separates voices, it doesn't identify them by name automatically. The transcription engine listens for distinct voices and tags each with a letter, A, B, C, and so on, as it hears them speak, up to ten voices, a fixed streaming ceiling that doesn't grow with your table size. A normal table of four to six never gets close to it; a one-shot with guests, or players voicing extra NPCs, is the case worth knowing the number exists. It never guesses that Speaker A is a specific player based on who usually talks first or what their character sounds like; guessing wrong and looking confident about it is worse than an honest letter, so it doesn't try.
Turning Speaker A into an actual name is a step you, or anyone else in the campaign, do: name a voice once and the later lines from it carry the name; fix a wrongly tagged line afterward and choose whether the fix applies forward. How that looks in any given tool is in the box below.
A crosstalk-heavy session needs more of that manual reassigning afterward than one that used the turn-taking habit above; that's a table habit to fix, not a mic to buy.
The record screen has a free Mic Test button: a short test take that's saved so you can read the transcript back, never counted against anything and never part of a recap. Run the one-minute check through it.
In-person mode records through whatever device you run it on and warns you off Bluetooth mics in the setup screen, for the same narrowband reason above. Audio streams live for transcription as you play; the audio itself is discarded on your device, not kept as a file, once the transcript has saved.
This is a real difference from Threadfall's Discord bot, which records each player from their own voice-channel track and knows the name automatically, because Discord already knows who's talking. A single shared mic doesn't have that information to start with; there's no account attached to a voice coming through one microphone. Diarization gets you separated voices. Only a person turns that into names, once, and it's remembered from then on. See recording a session on Discord for that setup.
What tends to work: a cleaner setup means a cleaner transcript, and that's the whole payoff.
However the audio gets captured, the payoff is the same: a cleaner setup means a cleaner transcript, which is what you're actually keeping.
Run the one-minute check above before your first real session so you find out whether your placement and gain actually worked before it's a session you can't redo.
A cleaner setup means a cleaner transcript, which means a recap that needed less manual correction afterward. That's the payoff, and it doesn't require spending anything: a phone in the middle of the table, used with the habits above, gets most tables most of the way there. The rest is placement and habit, not price.
An omnidirectional mic placed in the center of the table beats a directional mic or a laptop's built-in mic almost every time, because it removes the angle penalty a directional mic imposes on anyone not facing it straight on. Conference-style USB mics are built for exactly this shape of problem.
Yes, that's exactly the category built for this problem: a device meant to sit on a table and hear a room of people, not one person talking into a headset.
No. A phone placed in the middle of the table, away from the dice and drinks, with the roll call and the one-minute check run before you start, gets most tables most of the way there for free.
Up to ten, a fixed ceiling on the streaming transcription that doesn't scale with headcount. A normal table of four to six is nowhere near it; guests or extra-NPC voicing is the case worth keeping in mind.
It will tell voices apart and label them by letter as it hears them. It won't know real names on its own from a single shared mic, since there's no account attached to that audio the way there is on Discord. You assign the name once, live or afterward, and it's remembered for the rest of the session.
More: record a session on Discord · the safety tools page · what makes a good session recap · what happens to my recording · What is Threadfall? · all Threadfall guides