Recordings, Transcripts, and recaps
What happens between `/recording start` in Discord and a written recap on the Session page. Stems, segments, and the order things arrive in.
Ultimo aggiornamento 19/08/26
Wyrdli · Aiuto e documentazioneWhat happens between `/recording start` in Discord and a written recap on the Session page. Stems, segments, and the order things arrive in.
You wrapped at midnight, two scenes deeper than you expected and with the harbour master in custody. By Sunday afternoon, when you sit down to prep next week, you want the night's audio, a Transcript that lets you search for "burned letter", and a recap that tells you what to follow up on. The pipeline from /recording start in Discord to that recap on the Session page is the chain this article walks.
A Recording begins in Discord. You're in the play text channel of the Campaign, the players are in voice, and you run /recording start. The bot joins voice and starts capturing audio, one separate track per speaker who's connected.
Nothing else changes in Discord while the Recording is running. The bot just sits in voice, recording. You play the Session as normal.
When the table's done, you run /recording stop in the same text channel. The bot leaves voice and uploads the audio. Stopping is also what closes the file. Until you stop, the Recording is still growing.
A moment after /recording stop finishes, the Recording shows up on the Session's page on the web, with its length and the date it was captured. The Session and the Recording are now linked: open either one and you can reach the other.
The Recording's own detail page shows everything Wyrdli captured: the individual audio tracks (one per speaker) and the Transcript as it builds.
A Recording in Wyrdli isn't one big mono file, it's a set of stems, one per speaker, plus a mixed copy. A stem is the audio of a single Discord voice participant. If four players and a GM were in voice, you get five stems and a mix.
The Transcript builds from the stems, segment by segment. A segment is one utterance, one stretch of one speaker talking, with a timestamp. The Transcript is the ordered list of segments across all stems, so you read the conversation in the order it actually happened. While the Transcript is being built, segments appear progressively, and refreshing the Recording's detail page lets you watch new ones land.
If a stem looks broken (silence, a glitchy section, the wrong speaker labelled), the Recording page lets you retry that one stem on its own. The other stems and the mix don't need to be touched.
The Recording becomes a recap in two stages. First the Transcript builds in the background; then you ask for a recap once the Transcript is ready.
The "Generate Recap" button on the Session page is only useful once the Transcript exists. Click it and the recap runs against the Transcript, producing three things side by side with the Recording itself: a narrative summary of the night, a list of events that happened during the Session, and a structured debrief.
You can re-run the recap. The Transcript stays the same, the recap is regenerated from it.
The chain is forgiving. If the bot drops out, you record a second segment. If a stem glitches, you retry the stem. If the recap reads strangely, you re-run it. None of this requires touching the audio file itself.