The vocal_numeric project is a hardware/software emulation of 10 digit numeric keypad functionality using voice commands. The project has been created to participate in the TinyML edX class project competition.
UPDATE 09/29/2021: Though the class project has ended, I am continuing to develop this system. Please be aware that the current version of vocal_numeric no longer matches the version submitted for the class project.
Goals:
- To maximize learning by using concepts and procedures from the TinyML EdX course in practical application
- To develop a utility device which combines hardware and ML voice command application software to emulate the function of a 10 digit keypad robustly and inexpensively
- To win a prize in the competition
To achieve these goals, this project will:
- Utilize readily available resources combined in creative, practical and novel ways
- Follow development best practices outlined in the TinyML course including those for dataset development, responsible AI practices, and suggested workflows
- Be open to collaborations which will expand and enrich this project
I am hopeful to also add examples of additional functionality which may be possible with this device by the competition deadline of August 31, 2021.
Thanks go to Vijay Reddi and the entire staff of the TinyML Edx course for providing the opportunity to participate in this competition.
A video walk through of the code and a look at the circuit in action can be found here:
A video of some potential use case ideas can be found here:
An additional thought about multiple devices:
The code in this github is intended for use on an Arduino Nano 33 BLE Sense as we used in the class, but the model code is available for many other platforms via my public Edge Impulse project here:
https://studio.edgeimpulse.com/public/31878/latest
Please note that I am continuing to experiment with models and DSP on my Edge Impulse project so the latest version may not be the best performing version. Check the version number of the source code in this repository to see what I am currently using as my on-device standard.
Many thanks to Edge Impulse for supporting this project with additional processing time on their servers.