Docs

Speaker Labels

Can GM Assistant tell who is speaking in a single recording?

Yes. Upload one recording of the whole table, list who was there, and GM Assistant splits the recording into its voices. You name each voice once on the Who's who screen, and the lines of the transcript are then labeled with the player who said them. It works on long sessions and on tables with many speakers.

Does the analysis finish on its own?

Not when you listed who was at the table. The analysis pauses when it is ready for you and waits for you to name each voice on the Who's who screen, or to skip the step. Your notes are written after that. If you skip the Who was at the table step when uploading, there is no pause and no speaker labels.

Do I need a separate microphone for each player?

No. Speaker labels work from a single recording where everyone shares one microphone, like a phone in the middle of the table. If you do have separate recordings per player, upload it as multi-track instead and each track will be precisely labeled.

What does the question mark label mean in my transcript?

A line that starts with a question mark is one where it was too risky to say who was speaking. Either we were not sure, the line was too short to know, or the voice was left without a name. An unlabeled line is safer for your notes than a wrong name.

Can I change the speaker names after I finish?

Not on the Who's who screen. Once you click Finish or Skip, the recording and the voice data used to tell speakers apart are deleted, so the step cannot be reopened. Take a moment to check the clips before you finish. You can still correct who did what in your notes with Smart Edit.

GM Assistant can tell who is speaking in a single recording of your whole table. No separate microphones and no special setup: record the room, tell us who was there, and the lines of your transcript are labeled with the player who said them. It uses state of the art speaker detection, set up for what makes tabletop sessions hard: recordings that run for hours and tables with many voices.

The analysis pauses and waits for you. Your notes are not written until you come back and name the voices, or skip the step.
  1. You list who was at the table when you upload.
  2. We pause when we are ready for you.
  3. You name each voice on the Who's who screen. It takes about a minute.
  4. The analysis continues and your notes are written with the right names.

Before You Upload: Who Was at the Table

When you create a session, you list who played. One row per person, GM included, with the character each one ran. It fills itself in from your last session. Listing people here is what tells the analysis to pause later so you can name their voices.

  • People, not characters. Each player's name becomes their label in the transcript. Click the quoted name on a row to change it.
  • Remove anyone who missed the session. We use the number of people listed to find the voices.
  • Skip means no speaker labels for that session. Everything else works the same.
  • Never want them? Tick Skip speaker labels in the analysis options, then Set as new defaults. The step then stays closed on future uploads. Untick the box any time to bring it back.

Name the Voices: Who's who

When ready, the analysis pauses. Open the session and tell us who is speaking in each clip, then the analysis will resume and extract your notes.

Real tables are messy, so pick a scenario and try it:

  • Drag a clip onto the voice it belongs to, or use its ··· menu. The × marks a clip as unusable.
  • Finish turns on once everyone at the table is on a voice. Someone with no clips of their own goes on the empty row.
  • Skip speaker assignment continues with no speaker labels at all. It asks you to confirm first, since the step cannot be reopened.
Check before you click Finish. This step cannot be reopened afterwards. An analysis that has been waiting for more than 14 days may be resumed for you with no speaker labels.

What You Get

Lines in the Transcript tab start with the player who said them, one color per speaker. With names on the lines, your recaps, character details and memorable moments credit the right people too.

  • Sam: means the line is Sam's.
  • ?: means it was too risky for us to confidently say who it was or the line was too short to know. Unnamed voices also show ?:.
Single recordings only, for now: Voices are detected only in single track audio, where everyone is in the same recording. That is a Single File or Multi-Part upload. If you recorded one track per player, upload it as multi-track instead and each track is labeled directly.

When to Leave a Voice Unnamed

RULE OF THUMB: An unnamed voice costs you nothing. A wrong name puts words in the wrong player's mouth.

If a voice's clips sound like two different people, move the stray clips where they belong or leave the voice unnamed. Your notes come out just as well, and those lines simply carry no speaker label.

Your Audio and Voice Data

Normally your recording is deleted as soon as it has been processed. While Who's who is waiting for you, the recording is kept so you can play the clips. When you click Finish or Skip, the recording is deleted along with the voice data that was used to tell the speakers apart. What stays is the names you chose, which label the lines of your transcript.

See the Privacy Policy for the details.

Something not look right? The Troubleshooting page covers the common speaker label problems.