SOTCD DJ Bot Setup

SOTCD DJ Bot Setup Documentation

Overview

The DJ Bot is a headless Puppeteer-based automation that runs Strudel (live coding music environment) to generate Acid House audio for the Suns of the Cosmic Dust Twitch stream.

Infrastructure

Host Machine

• IP: 192.168.1.16 (Proxmox LXC container)
• Hostname: SOTCD streaming container
• SSH Access: ssh -i ~/.ssh/github_ed25519 root@192.168.1.16

Directory Structure

/root/sunsofthecosmicdust/
├── .env                    # Twitch credentials (TWITCH_STREAM_KEY, TWITCH_INGEST_URL)
├── go-live.sh              # Main stream launcher script
├── dj-bot/
│   ├── index.js            # Puppeteer script that runs Strudel
│   ├── package.json        # Node.js dependencies
│   └── node_modules/       # Installed dependencies (puppeteer)
└── website/                # GitHub Pages site (red40mademedoit/sunsofthecosmicdust)

Config Backup Location

• Local workspace: /home/shdwdev/clawd/sunsofthecosmicdust/
• Remote: /root/sunsofthecosmicdust/ on 192.168.1.16

Dependencies Installed on 192.168.1.16

System Packages

apt-get install -y xvfb fluxbox xterm cmatrix pulseaudio \
  libnss3 libnspr4 libatk1.0-0 libatk-bridge2.0-0 libcups2 \
  libdrm2 libxkbcommon0 libxcomposite1 libxdamage1 libxfixes3 \
  libxrandr2 libgbm1 libasound2

Node.js (via NVM)

# NVM installed at ~/.nvm
source ~/.nvm/nvm.sh
nvm install --lts  # v24.13.0

Puppeteer Chrome

cd ~/sunsofthecosmicdust/dj-bot
npm install
npx puppeteer browsers install chrome
# Chrome installed at: ~/.cache/puppeteer/chrome/linux-*/chrome-linux64/chrome

Stream Components

1. Xvfb (Virtual Display)

• Display: :99
• Resolution: 1920x1080x24

2. Fluxbox (Window Manager)

• Runs on display :99
• Provides window management for xterm

3. Visual Layer (Pink & Green Matrix)

• Two xterm instances running cmatrix
• Left side: Green (-C green)
• Right side: Magenta (-C magenta)

4. DJ Bot (Strudel Audio)

• Headless Chrome via Puppeteer (headless: true)
• Navigates to https://strudel.cc/?panel=repl
• Injects Acid House code into the REPL
• Key Chrome args for audio:
  • --autoplay-policy=no-user-gesture-required
  • --no-sandbox
  • --disable-setuid-sandbox

5. PulseAudio

• Virtual sink: virtual_speaker
• Monitor source: virtual_speaker.monitor
• Start with: pulseaudio --start --exit-idle-time=-1
• Create sink: pactl load-module module-null-sink sink_name=virtual_speaker

6. FFmpeg (Stream Encoder)

• Video: x11grab from :99
• Audio: PulseAudio monitor OR lavfi sine wave (drone fallback)
• Output: RTMP to Twitch

go-live.sh Script Flow

1. Load environment variables from .env
2. Start Xvfb on :99
3. Start Fluxbox window manager
4. Launch dual xterm windows with cmatrix (Pink & Green)
5. Start DJ Bot (node index.js)
6. Start FFmpeg to capture and stream

Troubleshooting

DJ Bot Crashes

• Check /tmp/dj-bot.log for errors
• Common issue: Missing X server → ensure headless: true in index.js
• Missing Chrome libs → reinstall dependencies

No Audio

• Verify PulseAudio is running: pulseaudio --check
• Check virtual sink exists: pactl list sinks short
• Verify Chrome is outputting to sink: pactl list sink-inputs

Stream Not Starting

• Check for stale lock file: rm -f /tmp/.X99-lock
• Kill all processes: pkill -9 ffmpeg xterm fluxbox Xvfb node

Quick Commands

Start Stream (Matrix + Drone Audio)

ssh -i ~/.ssh/github_ed25519 root@192.168.1.16 \
  "cd ~/sunsofthecosmicdust && nohup bash go-live.sh > /tmp/sotcd.log 2>&1 &"

Stop Stream

ssh -i ~/.ssh/github_ed25519 root@192.168.1.16 \
  "pkill -9 ffmpeg xterm fluxbox Xvfb node; rm -f /tmp/.X99-lock"

Check Stream Status

ssh -i ~/.ssh/github_ed25519 root@192.168.1.16 "pgrep -a ffmpeg && tail -20 /tmp/sotcd.log"

Current Status (2026-02-02)

• ✅ Matrix visuals working (Pink & Green)
• ✅ Drone audio working (55Hz sine wave)
• ❌ DJ Bot Strudel audio: ABANDONED (WebAudio issues - see below)
• ✅ SuperCollider acid house engine: WORKING
• ✅ Multi-genre audio switching: FIXED (v3 with crossfades)
• ✅ Stream bitrates: 6500k for both Twitch & YouTube

Milestone: 15+ Hour Livestream POC (2026-02-02)

Completed 15+ hour continuous Twitch stream documenting full proof-of-concept build. Video cached for archival. Demonstrated working SuperCollider audio pipeline with VB-Cable routing to FFmpeg.

SuperCollider Setup (ShadowMaster - Windows)

Host Machine

• IP: 192.168.1.214
• Hostname: ShadowMaster
• User: shdwadmin
• SSH Access: ssh shdwadmin@192.168.1.214 -i ~/.ssh/id_ed25519

Directory Structure

C:\Users\shdwadmin\sotcd\
├── START_CONDUIT.bat       # Main launcher (starts SC + Bridge + Hydra)
├── cosmic_genres.scd       # Multi-genre engine with crossfades (v3)
├── headless_boot.scd       # Single-genre adaptive acid
├── signal_bridge_visual.js # Cosmic data → OSC + WebSocket bridge
├── hydra_cosmic.js         # Hydra visual code
├── stream_twitch.bat       # FFmpeg → Twitch (6500k bitrate)
├── stream_youtube.bat      # FFmpeg → YouTube (6500k bitrate)
├── kill_stream.bat         # Stop all streams
└── node_modules/           # Node dependencies (osc, ws)

Audio Routing

SuperCollider → VB-Audio Virtual Cable → FFmpeg capture → RTMP

• Server port: 58110 (avoiding Windows Hyper-V reserved 57xxx range)
• Audio device: “CABLE Input (VB-Audio Virtual Cable)“

The Conduit System

START_CONDUIT.bat orchestrates the full stream:

1. Launches SuperCollider IDE with cosmic_genres.scd
2. Starts Signal Bridge (signal_bridge_visual.js)
3. Opens Hydra code + Hydra website for visuals
4. User manually runs stream_twitch.bat / stream_youtube.bat

Signal Bridge pulls real-time cosmic data:

• 🌍 USGS Earthquake feed
• ☀️ NOAA Solar X-ray flux
• 🌬️ Solar wind plasma (density, speed, temp)
• 🌙 Planetary positions & moon phase

Data maps to audio parameters via OSC on port 57120.

Working SynthDefs

• \acidBass - 303-style acid bass with filter envelope
• \acid303 - Classic squelchy acid line
• \kick - 4/4 kick drum with pitch sweep
• \hihat - Hi-hat (white noise + HPF)
• \dronePad - Ambient pad with reverb
• \darkDrone - Low rumbling drone
• \chaosPad - Chaos-modulated pad

Multi-Genre Switching (FIXED 2026-02-02)

v3 now working with smooth transitions:

• 4-second crossfades between genres (patterns overlap)
• Gradual BPM interpolation (5% per tick, no jarring tempo jumps)
• Transition lock prevents rapid genre thrashing
• Signal-based auto-switching:
  • < 0.33 → Ambient (75 BPM)
  • 0.33-0.66 → Acid House (125 BPM)
  • > 0.66 → Dark (85 BPM)
• Manual override: send /genre ambient|acid|dark via OSC

Fixes applied 2026-02-02:

• START_CONDUIT.bat now loads cosmic_genres.scd (was incorrectly loading headless_boot.scd)
• Genre switcher rewritten with crossfade overlap instead of hard stops
• BPM smoother routine for gradual tempo changes
• Both stream batch files set to 6500k bitrate

OSC Endpoints (port 57120)

Address	Range	Description
/signal	0-1	Master signal (triggers genre switch)
/bpm	60-200	Target tempo
/filter	100-8000	Filter cutoff frequency
/chaos	0-1	Chaos/modulation amount
/genre	symbol	Manual genre override (ambient/acid/dark)

Stream Encoding (FFmpeg)

Both Twitch and YouTube use identical settings:

ffmpeg -f gdigrab -framerate 30 -i desktop \
  -f dshow -i audio="CABLE Output (VB-Audio Virtual Cable)" \
  -c:v libx264 -preset veryfast -b:v 6500k \
  -pix_fmt yuv420p -g 60 \
  -c:a aac -b:a 128k -ar 44100 -ac 2 \
  -f flv "rtmp://..."

WebAudio Suspension Problem

The Issue

Browser WebAudio (used by Strudel) is designed to prevent autoplay abuse. Even with all the “right” flags, the AudioContext suspends without real user interaction:

• --autoplay-policy=no-user-gesture-required doesn’t fully work
• AudioContext gets suspended randomly
• Audio plays for a few seconds/bars, then stops
• Not viable for 24/7 unattended streaming

What We Tried

• Puppeteer with headless Chrome
• xdotool to simulate clicks
• Various Chrome flags for autoplay
• PulseAudio virtual sink routing

Result

Occasional 2-bar bursts of audio, then silence.

Alternative Approaches (If Revisiting)

Approach	Pros	Cons
SuperCollider	Headless-native, pro audio	Learning curve
SonicPi	Ruby-based, no browser	Requires display for GUI
Node + tone.js	JavaScript, server-side	node-web-audio-api limitations
Pre-rendered tracks	Simple, reliable	No live generation
Icecast + liquidsoap	Proper radio stack	Complex setup

Lessons Learned

• Browser WebAudio is hostile to automation by design (anti-autoplay)
• For reliable 24/7 headless audio: use native audio tools, not browsers

Related Resources

• Twitch: https://twitch.tv/sunsofthecosmicdust
• Website: https://red40mademedoit.github.io/sunsofthecosmicdust/
• Strudel: https://strudel.cc/

See Also

• Docker-Services-Deployment