Interaction with Face & Voice
Ankit Shukla, Feb 2
Agenda
 Introduction
 Why Now ?
 Why This ?
 Facial Recognition
 Voice Recognition
 A Small Demo
 Q & A
INTRODUCTION
Why Now ?
Why These ?
 Maturity of models
 Rapid Research - openCV to tensorflow
 Interaction Easiness
 Hardware Support
 No need of physical devices – in case of authentication
Facial Recognition – Step by Step
1. Find a face - HOG
Picture
Grayscale
Gradient
Gradients
Gradients – contd …
Gradients – contd …
Finally found a face 
It looks like our face
2. Pose & Project
3. Encode Face
4. Training Model
Voice Recognition – Step by Step
1. Analog To Digital
2. Acoustic Modelling
3. Language Modelling
Overall
What to Test ?
 Accuracy
 Precision
 Other Tests
 Load
 Security, etc.
Confusion Matrix
Accuracy
Recall (Sensitivity)
Precision
Prepare your test data
 Right mix
 Non – uniform distribution
 Foreign Element Introduction
 Leveraging Preprocessing
Demo
Q & A
Thank You

Face & Voice Recognition

Editor's Notes

  • #8 Histogram of Oriented Gradients