DEV Community

Tejas
Tejas

Posted on

Simple Easy Voice Assistant using Raspberry Pi 1

#ai #raspberrypi #python #devops

Photograph of a Raspberry Pi 1 board connected to a USB webcam, desktop speaker, and laptop on a clean white desk setup

Introduction & Hardware Overview

This is a super easy setup, though your overall speed and performance will depend on your hardware—the better the hardware, the faster the results!

For this project, I used a Raspberry Pi 1 from 2013. It's really old and slow, but it definitely gets the job done! You just need to be a little patient during the initial package installation.

To keep the Pi as fast and lightweight as possible, I installed DietPi as the operating system.

🛠️ What You Need

1. The Client (Raspberry Pi 1)

  • Raspberry Pi 1 (or any newer Pi model)
  • DietPi OS (keeps memory usage super low)
  • USB Microphone: I used an old USB webcam microphone. (Pro Tip: Put a piece of tape over the camera lens if you have privacy concerns!)
  • 3.5mm Speakers / Headphones

2. The Server (Mini PC / Local Host)

  • Mini PC (or any Linux host on your local network)
  • Ollama: Running lightweight models like qwen2.5:1.5b or llama3.2:1b
  • Faster-Whisper: For fast local Speech-to-Text

🌐 Network & Quick Overview

  1. Local Network: Ensure both your Pi and Server are on the same local network. Verify connectivity by running a ping from the server to the Pi and vice versa.
  2. Install & Test: Run the requirement setup and shell scripts on the Pi, and run the server script on the Mini PC.
  3. Verify Logs: Test both scripts running together and check the logs using the provided code commands.
  4. Enable on Boot: Once everything works, convert the scripts into systemd background services and enable them on boot so your AI assistant starts automatically whenever your Pi powers on!

📊 Viewing Logs Live

Checking the logs live is the best way to see your voice queries, transcripts, and AI answers in real-time.

1. View Client Logs (Raspberry Pi 1 / DietPi) 

sudo journalctl -u jarvis-client -f
Enter fullscreen mode Exit fullscreen mode

2. View Server Logs (Mini PC) 

sudo journalctl -u jarvis-server -f
Enter fullscreen mode Exit fullscreen mode

🔧 Troubleshooting Guide & Pro Tips

Here are the most common issues you might encounter when building this project and how to fix them:

1. Error: Device or resource busy

  • Cause: You have multiple scripts or background services trying to access the microphone at the same time.
  • Fix: Kill extra background processes and make sure only one instance of jarvis-client is running: 
  sudo pkill -9 sox
  sudo systemctl restart jarvis-client
Enter fullscreen mode Exit fullscreen mode

2. Sound Device Card Numbers Shift After Rebooting

  • Cause: Linux ALSA card numbers (card 0, card 1) can change randomly every time you restart your Pi.
  • Fix: Use permanent device names instead of numbers:
    • Mic: plughw:Camera,0 (or your webcam name)
    • Speakers: plughw:Headphones,0

3. Audio Crashes with Floating point exception on Pi 1

  • Cause: SoX play or ffplay uses floating-point audio resampling math that crashes on old ARMv6 (Pi 1) processors.
  • Fix: Use native ALSA aplay -D plughw:Headphones,0 for playback, and have the server convert all output audio into standard 16kHz mono WAV format.

4. Audio Cuts Off Before You Finish Speaking

  • Cause: The SoX Voice Activity Detection (VAD) silence pause time is set too short.
  • Fix: Increase the silence pause parameter in jarvis_listener.sh from 0.8s to 1.8s: 
  sox -t alsa "$MIC_DEVICE" -c 1 -r 16000 -b 16 "$TEMP_IN" silence 1 0.1 5% 1 1.8 5%
Enter fullscreen mode Exit fullscreen mode

https://github.com/teejaz/jarvis-edge-ai
Feel free to reach out for any question or additional ideas.

Top comments (0)