Smart message box

In this tutorial we will look into how turn a Raspberry Pi, a web cam and a speaker into a smart message box.

The concept

That concept might be a little obscure!

We were thinking about a funny smart device that a team / a family / flatmates could use collectively. Then we thought that it would be great if we someone going to the coffee machine could leave a private message (work related or not) to another member of the team who would come to the coffee machine later. Of course, such a function could be filled up with an app running on a tablet. But we wanted something more fun where no login nor typing on a touch screen is required (cause you are holding your coffee, right!).

The control flow is as follows:

  1. The device wait to see a new person in front
  2. If nobody is there, run a small tagline as “come on”
  3. When a face is detected, find the identity of the person
  4. Play all messages for this person (if any)
  5. When no message left, propose to leave a new message
  6. Record the message until “stop” is said
  7. Ask for confirmation
  8. Ask for the recipient’s name
  9. Store the identity, the recipient and the date
  10. Say goodbye
  11. Go back to 1 and repeat.


The hardware indicated below is only a suggestion, it is possible to build you assistant using different options (as long as you have a video/audio input and en audio output).

  • Raspberry Pi Model B/B+/2
  • 4GB SD Card
  • USB web cam (with a microphone)
  • Powered speakers with a 3.5mm jack input



OpenCV will be used to retrieve and format the video stream captured by your web cam. Please follow the steps detailed on opencv installation guides.

Note that on most operating systems, opencv libraries are available as binaries through package managers. On Ubuntu, typing this command is enough to have OpenCV installed:

sudo apt-get install python-opencv


PyAudio will be used to retrieve audio buffer recorded by your web cam. As for OpenCV, you can either follow the instructions found here or look into your operating system package manager.

On Ubuntu, typing this command is enough to have PyAudio installed:

sudo apt-get install python-pyaudio

Angus SDK

We will use Angus API to provide voice and users recognition. Please refers to these steps to install the python SDK.

Main program

Get the script

Retrieve the code of this project on github:

git clone

All the code was commented.

Prepare face directory

The main program looking for a subdirectory “ids” to create the directory for the message box. Please create a new subsubdirectory for each person you want include in the system. And for each, put inside at least one picture (jpg or png) of the face. For example we uploaded in the repository our 3 faces:

├── Aurélien
│   └── face.jpg
├── Gwennael
│   └── face.jpg
└── Sylvain
    └── face.jpg


For run the script just:


After this, the program invite you to select the input and output devices:

2 : HDA Intel PCH: ALC283 Analog (hw:1,0)
3 : HD Pro Webcam C920: USB Audio (hw:2,0)
5 : pulse
6 : default
Select your input: 3
0 : HDA Intel HDMI: 0 (hw:0,3)
1 : HDA Intel HDMI: 1 (hw:0,7)
4 : hdmi
5 : pulse
6 : default
Select your output: 6

You can give the input and output device indices on command line:

python 3 6


  • Sound issues:

    When using PyAudio to play sound directly on the audio output controlled by the bcm2835 on raspberry-pi, you may have some difficulties to get a clean sound. Check this thread for example.

    You can fix this issue by defining a new alsa output by editing a local configuration file .asoundrc (check the doc for more information) in your home directory or a global setting in /etc/asound.conf:

    pcm.convert {
         type plug;
         slave {
               pcm default;
               rate 48000;

    This piece of code creates a new output device that resamples to 48Khz before sending the signal to the standard output (by default the bcm2835 audio jack output). You just have to select “convert” at program startup in output selection.


The codes provided in this project are under an Apache v2.0 license.