AI-Gan final repo

AI-Gan is a cool demo application on HiKey 970. The flow is as follows:

Also please see Block diagram

Uses MTCNN for Face detection. Uses DLib for face alignment. Uses FaceNet for generating embeddings. Uses pre-trained TF graph for speech recognition.

• code to train speech model for your words can be provided if needed
• no training of models required, only inference performed

This code is to be run on HiKey970 only.

To run on x86_64 computer, modifications are required (should be only done if you know what you are doing)

To run this demo you need to connect 2 Arduino Uno boards to HiKey.

2 bit signals are passed to Arduinos for 4 different statuses

Arduino no. 1: Servo motor for dispenser

461  A0
487  A1 

Arduino no. 2: For LED indications

337  A0
501  A1 

Kernel modules inside sound_drivers folder have to be insmod for microphone to work

Place them in /lib/modules/... as directory structure present and then insmod them in order chnage name of kernel as per actual kernel name (found using uname -a)

1)midi
2)hdwep
3)usb-audio

Check if modules have been successfully installed using lsmod

sudo apt-get install portaudio19-dev
pip install pyaudio; wave; pydub

Check which card mic is using in /proc/asound/

find sound card number being used and change accordingly device index in /utils/sound_record.py

In /etc/modprobe.d/alsa-base.conf change the index values on these lines:

options snd-usb-audio index=0		to  (whatever index you find on your device)
options snd-bcm2835 index=1		

paste .asoundrc in home directory and reboot insmod modules again after reboot

models directory contains all files that are imported/called by the actual code.

main_in.py and main_out.py have to be run simultaneously in separate terminals

main_in.py has code for input

1)Code to check if a person is present in front of camera or not
2)If person detected, capture face data 
3)Capture voice input, recognise the word and assign it as label 
3)Save data

main_out.py has code for recognition

1)Code to check if a person is present in front of camera or not
2)If person detected, recognise him/her and find label
3)Send command to robotic arm only once for each person

****Some model files are too big for uploading on git. Contact for details.