@absolutejs/meeting-discord

Discord voice source adapter for @absolutejs/meeting. A bot joins a Discord voice channel and streams each participant's audio into the meeting core, so the @absolutejs/voice scribe can transcribe the conversation live.

Discord gives one audio stream per user, so speakers are known exactly — each transcript turn is labelled with the real Discord user, no diarization needed.

How it works

@discordjs/voice exposes a per-user Opus stream for everyone speaking. This adapter subscribes to each, decodes Opus → PCM (48 kHz stereo) and downmixes to mono pcm_s16le, and emits it as audio tagged with the Discord user id.

Discord voice channel ──(per-user Opus)──▶ adapter (decode + downmix) ──▶ @absolutejs/meeting ──▶ scribe ──▶ per-speaker turns

Install

bun add @absolutejs/meeting-discord @absolutejs/meeting @absolutejs/voice discord.js @discordjs/voice
# Discord voice also needs an Opus codec + an encryption lib at runtime:
bun add @discordjs/opus libsodium-wrappers
# (or the pure-JS / native alternatives: opusscript, sodium-native, tweetnacl)

The bot application needs the Guilds + Guild Voice States gateway intents and the Connect permission in the target channel.

Usage

import { createMeeting } from "@absolutejs/meeting";
import { createDiscordMeetingSource } from "@absolutejs/meeting-discord";
import { deepgram } from "@absolutejs/voice-deepgram";

const source = createDiscordMeetingSource({
  token: process.env.DISCORD_BOT_TOKEN!,
  guildId: "123…", // server id
  channelId: "456…", // voice channel id
});

const meeting = await createMeeting({
  source,
  stt: deepgram({ apiKey: process.env.DEEPGRAM_API_KEY! }),
  sessionId: crypto.randomUUID(),
});
meeting.on("turn", (turn) => console.log(turn.participant?.name, turn.text));

await meeting.start(); // bot joins the channel + starts listening
// …
await meeting.stop(); // bot leaves

Pass a pre-built, logged-in client instead of token to share one Discord client across features.

API

• createDiscordMeetingSource(options) → MeetingSource.
  • start() logs in (if given a token), joins the voice channel, and subscribes to each speaker.
  • stop() leaves the channel + destroys the client it created.
• DISCORD_AUDIO_FORMAT — the mono 48 kHz pcm_s16le format it emits.
• stereoToMono(buf) — the downmix helper (exported for reuse/testing).

License

CC BY-NC 4.0