How a Developer Used a Raspberry Pi and AI to Jam a Neighbor’s Bluetooth Speaker

A Raspberry Pi‑based device named Reggaeton Be Gone monitors ambient audio, classifies Reggaeton music with a TinyML model, and when detected sends crafted Bluetooth packets that disrupt the speaker’s audio stream.

Hardware

Raspberry Pi 3B+ (1.4 GHz quad‑core, Bluetooth 4.2, dual‑band Wi‑Fi)

DFRobot 0.91‑inch OLED display (128×32 px, I²C)

DFRobot Gravity digital large‑button module (5 V/3.3 V compatible)

USB microphone or sound‑card (e.g., Behringer Xenyx302)

microSD card, 5 V 3 A power supply, jumper wires, 3D‑printed enclosure

Optional cheap Bluetooth audio receiver board for testing

Hardware Assembly

Button → GND and GPIO 26

OLED → VCC, GND, SDA (GPIO 2), SCL (GPIO 3)

USB mic or sound‑card → USB port

Software Setup

Install Raspberry Pi OS, enable SSH and I²C via sudo raspi-config, then install the following packages:

pip install RPi.GPIO</code>
<code>pip3 install Adafruit_GPIO</code>
<code>sudo apt install git</code>
<code>pip install picamera</code>
<code>sudo apt install python3-opencv</code>
<code>sudo apt-get install libatlas-base-dev libportaudio0 libportaudio2 libportaudiocpp0 portaudio19-dev</code>
<code>git clone https://github.com/edgeimpulse/linux-sdk-python</code>
<code>sudo python3 -m pip install edge_impulse_linux -i https://pypi.python.org/simple</code>
<code>sudo python3 -m pip install numpy</code>
<code>sudo python3 -m pip install pyaudio

Verify OLED detection with i2cdetect -y 1 and configure audio input with alsamixer (select the USB sound card with F6 , adjust input volume with F4 ).

Machine‑Learning Model

Because the GTZAN dataset lacks a Reggaeton class, several Reggaeton tracks and other genres were recorded, converted to 16 kHz WAV files, and uploaded to the Edge Impulse platform. A TinyML model was trained for Linux ARM, exported as an .eim file, and copied to the Pi with permissions set to 744.

Python Logic

The script continuously captures audio, sends each sample to the Edge Impulse model, and if the Reggaeton confidence exceeds 0.95 it triggers a Bluetooth attack method (either rfconn or l2ping). Logs are written to log.txt and status is shown on the OLED.

myPath=""               # project folder
selectedDeviceId = 1    # USB mic ID
method = 1               # 1‑3, BT attack method
targetAddr = ""          # MAC of BT speaker
packagesSize = 0         # try 600‑1000 bytes
threadsCount = 0         # number of threads
threshold = 0.95         # Reggaeton confidence threshold
myDelay = 0.1            # delay between connections (seconds)
forceFire = 1            # for testing
model = ".eim"           # model file name

Obtaining the Speaker’s MAC Address

Two methods are provided:

Pair the speaker with a smartphone and read the MAC address from the connection details.

Use bluetoothctl to scan, then filter results with RSSI using sudo btmon | grep RSSI & sudo hcitool scan to infer the MAC.

sudo bluetoothctl</code>
<code>scan on

sudo btmon | grep RSSI & sudo hcitool scan

Final Result

The assembled Reggaeton Be Gone device successfully identifies Reggaeton tracks and, upon detection, sends Bluetooth packets that corrupt the audio stream of nearby speakers, forcing them to stop playback. Effectiveness depends on proximity to the speaker and varies across Bluetooth speaker models.

Reference: https://www.hackster.io/roni-bandini/reggaeton-be-gone-e5b6e2