How to Transcribe Discord Calls: Record, Split, Diarise

Discord has quietly become where podcasts get recorded, where communities run AMAs, and where remote teams hold their standups. The problem is that Discord has no built-in recording button and no transcript feature, so the moment the call ends the content is gone. This guide covers how to transcribe Discord calls properly: capturing the audio legally, getting separate tracks per speaker, and turning those tracks into a readable transcript with names attached.

The order matters. A transcript is only as good as the recording underneath it, and a single muddy stereo mix of six people talking over each other will produce a wall of text nobody can use. Get the recording right first.

Step 1: Get consent before you hit record

Before any tooling, settle the consent question, because it decides whether the recording is usable at all.

Many jurisdictions require every participant on a call to agree to being recorded, not just the person pressing the button. Rules vary by country and by state, and Discord's own terms prohibit recording people without their knowledge. If you are recording a public server, this is also a community trust issue, not only a legal one.

Practical norms that work well on real servers:

• Say it out loud at the top of the call: "I am recording this, and it will be published as a podcast." Verbal consent captured in the recording itself is the cleanest evidence you have.
• Pin a note in the voice channel's text chat stating that the channel is recorded, and add it to the channel topic so it shows in the header.
• If you use a recording bot, keep the bot in the channel only while recording. A bot that silently idles in every voice channel is how servers lose members.
• For guest interviews, get a one line confirmation in writing over DM before the call.
• Tell people how to opt out, and honour it by editing them out or not publishing.

Step 2: Record the call with Craig for per speaker tracks

Craig is the standard Discord recording bot and the reason most podcasts are recorded on Discord at all. Its key advantage is multi track output: every participant is recorded to their own separate audio file, so one person's keyboard noise does not sit on top of everyone else's dialogue.

To record with Craig:

1. Invite Craig to your server from craig.chat and grant it access to the voice channels you want to record.
2. Join the voice channel yourself.
3. Type /join in a text channel and select the voice channel. Craig joins and announces that recording has started, which doubles as an audible consent notice.
4. Craig DMs you two links: one to monitor the recording live, and one to delete it. Keep both.
5. When the call is over, type /stop or /leave.
6. Open the download link, choose the multi track option, and pick a lossless format such as FLAC or WAV. Download the zip.

Inside the zip you will find one audio file per speaker, usually named with their Discord username. That naming is what makes speaker labelling almost free later on.

If Craig is unavailable or your server blocks bots, an alternative bot with similar multi track behaviour will work the same way. What matters is that you end up with separate files, not one mix.

Step 3: Record with OBS if you cannot use a bot

Some servers do not allow bots, and some calls are ad hoc. OBS Studio records Discord audio locally and is free on Windows, macOS, and Linux.

The basic single track setup:

1. Open OBS and go to Settings, then Output, and switch Output Mode to Advanced.
2. Under the Recording tab, set the recording format to mkv (it survives crashes) and set your recording path.
3. In the Audio Mixer, add an Audio Output Capture source for your desktop audio. This captures what Discord plays, meaning everyone else on the call.
4. Add an Audio Input Capture source for your microphone. This captures you.
5. Hit Start Recording, take the call, then Stop Recording.

To get closer to per speaker separation, use OBS's audio track routing. In Settings, then Output, then Recording, enable multiple audio tracks. Then in the Advanced Audio Properties panel (the gear icon in the Audio Mixer), assign your microphone to Track 1 and desktop audio to Track 2. You now have your voice isolated from everyone else's, which is enough for a two person interview.

The limitation is real and worth stating plainly: OBS captures Discord as one desktop audio stream, so it cannot separate three guests from each other. For anything beyond a two person call, a multi track bot gives a far better transcript.

An alternative for larger calls is to have each participant record themselves locally, then collect the files afterwards. This gives the highest audio quality but depends on everyone remembering to press record, which is a bigger ask than it sounds.

Step 4: Prepare the audio files before transcription

A few minutes of prep here saves an hour of transcript cleanup.

Craig's multi track files are already aligned to the same start point, so do not trim the beginnings individually or you will lose sync. If you need to cut the first few minutes of small talk, cut the same amount from every track.

Rename each file to the speaker's actual name rather than their Discord handle. sarah-chen.flac is more useful than xX_nightowl_Xx.flac when the transcript comes back.

If you plan to publish the audio, do your noise reduction and levelling now, before transcription. Cleaner audio produces a more accurate transcript, particularly for accents, technical terms, and anyone on a laptop mic.

Keep the files lossless until the transcription step. Heavy compression damages exactly the frequency detail that speech recognition relies on.

Step 5: Turn the recording into a diarised transcript

You now have either a set of per speaker tracks or a single mixed file. The path differs slightly.

If you have per speaker tracks, you have the best possible starting point. Transcribe each track separately, then merge by timestamp. Because each file contains exactly one voice, attribution is guaranteed correct rather than guessed. This is the approach that produces publication ready podcast transcripts.

If you have one mixed file, you need speaker diarisation, which is the process of detecting voice changes and splitting the transcript into labelled turns. Modern transcription tools handle this automatically, though they will label speakers generically until you rename them.

Either way, upload the audio and let the tool do the work. Tapescribe handles both cases: drop in a file or paste a link and you get back a timestamped transcript plus SRT and VTT subtitle files, which is what you want if the call is going on YouTube as well as into a blog post. The same workflow applies whether you are handling a recorded community call or a webinar recording from a different platform.

To merge multi track transcripts manually:

1. Transcribe each speaker's file, exporting with timestamps.
2. Open all transcripts side by side.
3. Sort every line from every file by its start timestamp into one document.
4. Prefix each line with the speaker's name.
5. Read through and delete the empty segments where a speaker was silent but the tool caught background noise.

A quick note on formatting: if the call is going out as a podcast episode, keep the timestamps in. Listeners use them to jump, and search engines index them. The same logic applies to YouTube videos with timestamps, where timestamped text feeds directly into chapters.

Common problems and how to fix them

Craig's tracks are out of sync after editing. You almost certainly trimmed one file and not the others. Re-download the original zip and trim all tracks by the identical amount, or use your editor's group function so cuts apply to every track at once.

One speaker is much quieter than everyone else. Normalise each track individually before transcribing. Speech recognition struggles with very low input levels and will silently drop words rather than flag them.

Bleed between tracks. If a participant used speakers instead of headphones, their track contains everyone else's voices too. The transcript will then duplicate lines across speakers. Fix it by applying a noise gate to the offending track, and ask participants to wear headphones next time.

The transcript mislabels speakers on a mixed recording. Diarisation guesses when voices are similar or when people talk over each other. If accuracy matters, re-record with multi track. If you are stuck with the mix, correct the labels once at the top of each section rather than line by line, since errors usually cluster.

Names, jargon, and Discord handles come out wrong. Every transcription tool struggles with invented words. Keep a short glossary of recurring names and terms and do a find and replace pass after export. It takes two minutes and it is the single highest value cleanup step.

The recording link expired. Craig deletes recordings after a retention window. Download the zip immediately after the call and store it somewhere permanent rather than relying on the link.

Frequently Asked Questions

Can you record a Discord call without a bot?

Yes. OBS Studio, or any screen recorder that captures desktop audio, will record Discord calls locally without any bot in the server. The trade off is that desktop audio arrives as one mixed stream, so you cannot separate multiple guests from each other, only yourself from everyone else. For solo interviews that is fine; for panel calls a multi track bot produces a far cleaner transcript.

Is it legal to record Discord calls?

It depends on where you and the participants are located. Some jurisdictions require consent from every party on the call, others only from one, and Discord's own terms of service require that people know they are being recorded. The safe approach everywhere is to announce the recording at the start of the call and get a verbal yes, which also captures the consent in the recording itself.

How do I get separate audio tracks for each Discord speaker?

Use a multi track recording bot such as Craig. It records each participant into their own file, so you receive a zip containing one audio file per speaker, typically named by Discord username. Download in a lossless format like FLAC, and do not trim individual tracks or you will break sync between them.

What is the best format to transcribe a Discord recording?

Keep the audio lossless, so WAV or FLAC, right up until the transcription step. Compressed formats strip out detail that speech recognition uses to distinguish similar sounding words, and the damage is worst on quiet speakers and strong accents. If file size forces a compressed format, use a high bitrate rather than a default one.

Wrapping up

The single decision that determines transcript quality is made before anyone speaks: record multi track, with consent stated out loud, and everything downstream gets easier. Per speaker files give you accurate attribution without guesswork, clean audio gives you accurate words, and a labelled transcript turns one call into show notes, blog posts, clips, and searchable archives. If you are transcribing calls regularly, the workflow described in this guide to transcribing audio to text online will apply to just about any recording you make, not just Discord.