Voice In My Head

eSIM setup

1. Remove any previous eSIM.
2. Sign up for a new eSIM using Airalo or Nomad.
3. Install the eSIM, and turn on Data Roaming.
4. Check the latency to cloud servers.

Setup

Create a cloud server. If installing on Digital Ocean, make sure to enable the agent with advanced metrics.

For 4 users, 8 CPUs and 16 GB RAM is recommended. After creating the machine, add the IP address to the appropriate DNS record.

Prep the packages:

sudo apt update
sudo apt upgrade -y
sudo apt install -y build-essential # needed for streamp3 package
sudo apt install -y libmp3lame-dev # needed for elevenlabs
sudo apt install -y ffmpeg # for processing elevenlabs input

Install Anaconda:

wget https://repo.anaconda.com/archive/Anaconda3-2024.02-1-Linux-x86_64.sh
bash Anaconda3-2024.02-1-Linux-x86_64.sh -b
$HOME/anaconda3/bin/conda init
source ~/.bashrc
rm Anaconda3-2024.02-1-Linux-x86_64.sh

Clone the repo:

git clone https://github.com/kylemcdonald/voice-in-my-head.git
cd voice-in-my-head

Create the environment:

conda create -y -n vimh python=3.9
conda activate vimh
conda install -y -c conda-forge libstdcxx-ng # needed for daily-python
conda install -y pytorch torchvision torchaudio cpuonly -c pytorch
pip install -r requirements.txt

Setup nginx:

sudo apt install -y nginx
sudo ufw allow 'Nginx Full'
sudo cp .nginx /etc/nginx/sites-available/vimh.iyoiyo.studio
sudo ln -s /etc/nginx/sites-available/vimh.iyoiyo.studio /etc/nginx/sites-enabled/

Setup certbot:

sudo snap install --classic certbot
sudo ln -s /snap/bin/certbot /usr/bin/certbot
sudo certbot --nginx

Install nvm, Node, and Tailwind:

curl -o- https://raw.githubusercontent.com/nvm-sh/nvm/v0.39.5/install.sh | bash
source ~/.bashrc
nvm install 16
npm install -D tailwindcss
npx tailwindcss init
npm run buildcss

Fill out the .env file with the appropriate keys.

ELEVENLABS_API_KEY=...
OPENAI_API_KEY==...
DAILY_API_KEY==...
DEEPGRAM_API_KEY==...
TURN_TIME_SECONDS=50
TOTAL_TIME_MINUTES=25
ROOM_EXPIRE_MINUTES=35

Install the service:

bash install-service.sh

Setting up the iPhones

• Setup the iPhones without any Apple account
• Go to Settings > Notifications
  • Display As: Count
  • Siri Suggestions: Off
  • App-by-app: Off
  • Emergency Alerts: Off
• Get Airalo e-SIM for region
• Settings > Cellular Data Options > Data Roaming: On
• Home screen: remove Calendar stack, move bottom bar to lower row
• Enable AirDrop and share phone/background.png picture
• Under photo sharing at bottom left, set as wallpaper and pinch to resize
• Safari > Microphone > Allow
• Lock to portrait mode
• After connecting the AirPods, turn Automatic Ear Detection: Off
• Disable NameDrop

Setting up Guided Access

Apple Reference

• Open iOS page and "Add Shortcut to Home Screen"
• Go to Settings > Accessibility, then turn on Guided Access.
• Tap Passcode Settings, then tap Set Guided Access Passcode.
• Open the app from home screen.
• Triple-click the Home button.
• Set options:
  • Side Button ON
  • Volume Buttons ON
  • Motion OFF
  • Software Keyboards ON
  • Touch ON
  • Dictionary OFF
  • Time Limit OFF
• Tap Guided Access, then tap Start.

Testing

Run server with Flask autoreloading:

flask --app server.py --debug run

Run the server with gunicorn:

gunicorn -w 4 server:app

Shortcut for running gunicorn:

./run.sh

Setting the Duration

The duration of the experience is controlled in the .env file.

Notes on sound design

Sounds should match the audio stream:

• 1 channels (mono)
• 44.1kHz
• 16-bit depth

Note they might get slightly glitched by the compression and streaming algorithms.

They should also always fade out quickly, or sometimes they can create a lingering noise.

helpers/prepare-sound.sh will help prepare sounds for this format.

Notes on the script

Each row of the script has a function, input and output.

The function is the name of a function instead the VoiceInMyHead class. The input is the input to that function, and the output is where the output is saved.

When you save output to a variable, you can reference that as an input in later rows.

If a variable is referenced in a speak line, it should be surrounded by {curly braces}. (This is because the speak lines get preprocessed and combined.)