Uploading Audio
How do I upload session audio to GM Assistant?
Open New Session in your campaign, pick the upload mode that matches your recording (single file, multiple parts, or multi-track), choose your file, and start the analysis. Most recordings are a single file containing the whole session.
How do I get speaker labels in my transcript?
Upload your session as multi-track audio with one speaker per track, the way recorders like the Craig bot or Zencastr save it. Every transcript line is then labeled with who said it, and the analysis becomes more accurate. If every track captured the whole room, upload a single mixed file instead.
What are the file size and length limits for uploads?
A recording can be up to 4 GB in size and between 1 minute and 10 hours long. If your session is larger or longer, split it and upload each part as a separate session.
There are three ways to upload your session audio to GM Assistant, depending on how you recorded. Pick whichever matches your recording.
Single File
One audio file containing the whole session and all speakers. E.g. a phone in the middle of the table, a single mic capturing everyone, a Discord call recorded as one combined stream.
- Open New Session in your campaign
- Leave the audio toggle on Single Audio File
- Pick your file and continue
Multi-Part Audio
Sometimes a recording gets split mid-session: a Discord call drops, a recorder hits a file size limit, you restart partway through. You don't need to merge the files yourself.
- Switch the audio toggle to Multiple Audio Files
- Add every piece to the same track using Add Part. The parts concatenate in order.
Multi-part also works inside multi-track: each track can have its own parts.
Multi-Track and Speaker Labels
If your session audio has speakers separated across tracks, GM Assistant can label every line in your transcript with who said what and use that information to produce a noticeably more accurate analysis.
Two benefits:
-
Speaker labels in your transcripts. All
transcript formats (plain text, SRT, VTT, JSON) will have
speaker labels.
Here's a plain text example:GM: A goblin steps out from behind the rocks, drawing a rusted shortsword. Gandalf: I light my staff and call out to it. Stand down, creature. Frodo: I keep my hand on the hilt of Sting, watching the treeline. GM: The goblin hisses, glances over its shoulder, and starts backing away. Gandalf: Don't let it run. We need to know who it answers to. - More accurate analysis. Recaps, memorable moments, and other notes all benefit because GM Assistant isn't guessing who said what during overlap or rapid exchanges. Attribution of actions and decisions to the right character is noticeably better.
When multi-track works (and when it doesn't)
- One track per player headset on a Discord call.
- A couple sharing a single mic, with both names assigned to that track.
- The GM voicing every NPC on the GM track.
- Multiple mics in the same room (everyone sitting around a table, two phones recording from different angles, etc.). Every mic picks up every speaker, so there's no segregation for multi-track to attribute.
If your setup falls into the right column, a single audio file is the right choice. Pick the best recording or pre-mix them into a single track before uploading.
About background noise
Kids in the next room, dice rolls, snacks, ambient music, a player briefly chiming in on someone else's mic.... all fine. The system uses your speaker labels as the source of truth for attribution; minor cross-talk or noise on a track won't confuse it.
To capture per-speaker audio in the first place, see the Multi-Track Recording section of the recording guide for setups that work with Craig bot, Zencastr, OBS, and per-player mics.