Hello my name is Surk Park from South Korea and I am here to tell you about my project called Narcissus.
As you may of guessed, this project is a smart mirror.
I currently work with Watson API which consists of 8 separate cognitive APIs that focus on
language. I want to create a test application that uses language but also mixes in visual
recognition and voice control.
There are three main technologies used in my project
- Korean speech recognition using Google’s speech API
- Thirty separate voice controlled dialogues. For example “Show me the map”(narcissus pulls up google maps) or “ Show me music, The Beatles” (Narcissus pulls up youtube videos of the Beatles)
- Lastly the personal assistant displays weather, time, schedule and other selected important information
I will tell you the mirror’s architecture: it consists of a frame, 2-way mirror and a monitor you can see the
hardware inside the mirror. This currently consists of a Raspberry pi 3 mini-computer but I have changed to a
Nvidia Jetson TX2 to test deep-learning technology.
In addition, I’ll briefly explain the software architecture of the mirror. The mirror will utilize a Linux OS, Node.js
application framework and Evan Cohan’s open-source smart mirror modules.
On top of the open-source base, the mirror will use four main additional technologies.
- First, it will use SEGAN(Speech Enhancement GAN) as a noise cancellation module.
- Second, the mirror will use google’s STT (Though Watson STT options).
- Once this stage is complete, the mirror will utilize Snowboy’s wakeup function.
- Lastly, the dialogue will be handled by Watson API
The smart-mirror will also display several well-known external APIs such as YouTube.
This is a test application, but if successful, the potential uses are many including:
- Hotel room personal concierge
- personalized home mirrors
- Personal health
- Digital signage
- Home security
- and...home cloud storage