I don't know how, But They Found Me (BTFM).
btfm is a Discord bot that listens on a voice channel for key phrases, and plays audio clips into the channel in response.
You'll need to register a bot with Discord. Go to the Developer application page and create an application.
BTFM uses a PostgreSQL database to store audio clip metadata. Install PostgreSQL and create a database. For example:
sudo apt install postgresql postgresql-contrib
sudo systemctl restart postgresql.service
sudo -u postgres createuser btfm
sudo -u postgres createdb btfm
sudo -u postgres psql -c "ALTER USER btfm PASSWORD 'password';"
sudo -u postgres psql -c "ALTER DATABASE btfm OWNER to btfm;"
The btfm-server service will create the database schema when it connects. Any migrations required will also
be run automatically on updates.
Create a user for the bot:
$ sudo useradd --home-dir=/var/lib/btfm --create-home btfm
An example configuration file:
```
data_directory = "/var/lib/btfm/"
database_url = 'postgres://btfm:password@localhost/btfm'
discord_token = 'your discord token here'
channel_id = 0
guild_id = 0
logchannelid = 0
rate_adjuster = 100
randomclipinterval = 900
mimic_endpoint = "http://localhost:8888/api/"
[whisper]
model = "/var/lib/btfm/whisper/base.en.pt"
[http_api]
url = "127.0.0.1:8080"
user = "admin"
password = "admin"
#
tls_certificate = "/var/lib/btfm/fullchain.pem"
tls_key = "/var/lib/btfm/privkey.pem" ```
You can place this, for example, in /var/lib/btfm/btfm.toml.
The Whisper Python API is used to perform transcription, so you need to set up a Python environment and install Whisper. The recommended approach is as follows (assuming you're using Fedora Linux):
sudo dnf install python3-pip
sudo -u btfm bash -c \
'python3 -m venv --upgrade-deps $HOME/.whisper && \
$HOME/.whisper/bin/pip install openai-whisper'
Next, test the installation and download the model to use (replace the model as necessary and be sure to adjust the configured model in btfm.toml to match):
sudo -u btfm bash -c '$HOME/.whisper/bin/whisper --model base.en --model_dir $HOME/whisper/ <an audio file>
An example systemd unit to run BTFM:
``` [Unit] Description=BTFM Discord bot After=network.target
[Service] Type=simple User=btfm Group=btfm Environment="PATH=/var/lib/btfm/.whisper/bin:/usr/bin:/usr/local/bin/" Environment="BTFMCONFIG=/var/lib/btfm/btfm.toml" Environment="RUSTLOG=warn,btfm=info" ExecStart=/usr/local/bin/btfm-server run Restart=always RestartSec=60
[Install] WantedBy=multi-user.target ```
If building from source, install make, autotools, libopus headers, gstreamer headers, libsodium headers, and the openssl headers.
Add clips and phrases with the btfm clip sub-commands:
btfm clip add --phrase "they found me" "I don't know how, but they found me..." run-for-it-marty.mp3
See btfm clip --help for available sub-commands and options.
Start the bot with btfm-server run. See the systemd unit above for details.
See btfm-server run --help for command line arguments and documentation. To
obtain the guild and channel ID, go to your Discord User Settings ->
Appearance, and enable Developer Mode. You can then right-click on the server
for and select "Copy ID" for the guild ID, and then right-click the voice
channel you want the bot to watch and "Copy ID" that as well.
If you are so inclined, there is a Dockerfile and some helper scripts in the devel/ folder
that you may find to be handy for development. The scripts assume you have
podman installed. You can use build.sh to build a development container,
and you can use cargo.sh to run Rust's cargo tool inside
the container. You can probably guess what test.sh does, if you are somebody's kid and are
smart.