Overview / Usage

This project features both the hardware implementation of a retrofit voice-controlled home automation device using Raspberry Pi.
The voice-control will be based on translated text from a pre-trained neural-network speech to text model. The other major aspect of this project will be training of a speech to text engine using natural language processing and also a text to speech engine.

Methodology / Approach

1. The hardware is built. This consists of a digital uni-directional USB microphone, a raspberry pi, and an Intel Movidius Neural Compute Stick, we may need the Intel Up2 board.
2. A corpus of home automation commands is gathered from different speakers. This corpus will be the dataset that will be used to train the model
3. The model for the speech to text engine is designed utilizing deep learning and natural language processing.
4. The model is trained and evaluated
5. A control program is written for the hardware system, this control program preprocesses audio commands, submits the processed audio to the speech to text engine and awaits inference from the engine. It analyzes the translation returned from the speech to text engine and correspondingly sends the required commands to the actuators.
6. A text to speech engine will provide corresponding audio feedback for confirmation and engagement with the user.

Technologies Used

Deep Learning,
Natural Language Processing,
Python
Intel Movidius Neural Compute Stick
Raspberry Pi
Intel UP2 boards