Build SHEP (Pictures coming soon)

Posted 07/August/2020

In this tutorial we will make our own portable SHEP AI. You can modify the code to work best for your own robotic system. This tutorial will go over the hardware needed, the setting up of libraries and configuring of the Raspberry Pi to make SHEP portable. It is up to you how you want it to look and what interface you use.

Picking parts

Firstly, we will want to pick our hardware. How do we want the AI to look and interact? I have personally used the Pimoroni Unicorn HAT 16x16. I have used th I2S Adafruit Microphone which uses pin 18 on the Pi's GPIO. You will need to consider what pins are being used. If you decided to use the Unicorn HAT 8x8 which also used pin 18, you may want to consider using a USB microphone.
I will be using the Raspberry Pi 3 model A+, as it has a USB where I can insert a WiFi dongel, and it has an Audio jack which saves me needing to purchase an audio HAT or module. I will get a PCB prototyping hat to attach the microphone.

Wiring it up

The HAT for the Pi fits nicely on, but our I2S requires wiring. Firstly we will solder on the I2S microphone onto the PCB. Solder it near one of the sides so your voice will reach it. The Pins connect up as follows:

3V to 3.3V
GND to GND
BCLK to BCM 18
DOUT to BCM20
LRCL to BDM19

Here is a more in depth tutorial for the I2S Mic On top of the Pi you will place the PCB, then on top of that place the Unicorn HAT. I attached a switch to a 4xAA battery pack so that SHEP could be mobile.

Installing software

Once you have picked your hardware you will need to install the libraries. For me that will be the Unicorn Hat 16x16 library. I will boot up my pi, connect it to the internet and open up the terminal (CTRL ALT T).

sudo pip3 install unicornhathd

Then we will will need to configure the rest of our hardware. If you are using your own microphone method the you will need to install that your way. For us we will be using the I2S microphone.

cd ~
wget https://raw.githubusercontent.com/adafruit/ Raspberry-Pi-Installer-Scripts/master/i2smic.sh
chmod +x i2smic.sh
sudo ./i2smic.sh

After this you will follow the prompts. We are using the Pi 3 so will select model option 2. We will then select Y for auto boot as we will want it to load in every time we turn the AI on. After this simply reboot.

sudo reboot

We will then want to make sure that the code is using the audio jack for output (unless you have your own means). This can be done through sudo raspi-config > advanced options > audio >audio jack. This should configure all the audio options that you need.
Next we will actually install SHEP.

sudo wget https://shepai.github.io/code/run.zip

Extract the file which you have downloaded into a folder path of your choosing. Remember the exact pathway as you will be using it later.
You will need to download all the libraries and software needed:

sudo pip3 install wavio
sudo pip3 install sounddevice
sudo pip3 install soundfile
sudo pip3 install wave
sudo apt-get install python3-pyaudio
sudo pip3 install speechrecognition
sudo pip3 install nltk
sudo apt-get install espeak
sudo python3
>>>import nltk
nltk.download()

Follow the prompts on the nltk and click download. This may take a while. What we are doing here is installing all the data needed for SHEP to understand language.
NOTE:
If you have errors later on with espeak then you will need to install nltk like so:

python3
>>>import nltk
nltk.download()

Configuring the Pi

We will want out Python script to load every time we turn the Pi on. This will be done using the following:

sudo nano /etc/profile

At the bottom of this file we will add

sudo python3 /pathway_to_folder/HyperCube.py &

The pathway will be wherever you saved it. If you are having trouble with espeak then you may need to do the following:

sudo adduser pi root

Then in /etc/profile add the same code but without sudo.

python3 /pathway_to_folder/HyperCube.py &

This will then boot your python code automatically. Before we reboot we have one more thing to add. We will not need to use the GUI so can go to:

sudo raspi-config

and change the boot option to boot to console autologin.

Altering the code for your own hardware

Being you set your microphone up as the default Raspberry Pi Mic, the audio should work fine. If you have used other means for audio output rather than the inbuild audio jack then once again, make sure it is the default.
I have added in a green dot into the corner of the eye to signify when the audio is listening. This makes it easier to interact with. Something you may want to consider doing in your own interface.

To alter the eye code for your own blink algorithm you will need to edit 3 different functions.
recordLED(type)
displayEye()
blink()
These are the ones which need editing. The recordLED turns an LED on or off depending if the system is recording. The other two are the ones which read in a text file and display the colours on the map. You can set this in any way you wish, and import new libraries for your own interface.

Below shows a tutorial on youtube of how to set up the voice recognition AI. You can follow this if you wish to start the code from scratch and add your own sensors and outputs.