Edge Impulse (Netherlands) has a good track record in making Machine Learning (ML) available and easy to implement on a long list of microprocessors platforms. They span from simple arms based Arduino to advances specialized ML processor like Nvidia's Jetson Nano and Greenwaves GAP7 / 8.
Edge Impulse have just released official support for Espressif ESP32 and in this workshop we will implement a simple keyword detection algorithm on ESP32 a based development board.
We will have Jenny Plunkett from Edge Impulse to introduce the tools and workflow and run the workshop. She will be joining us via a video link and we will have Jørgen Kragh Jakobsen to co-host onsite.
We all know that we can have attention from our smart speakers or phones by calling “Alexa” or “Hey Google”. In this work shop we will create our own list of keywords and create at Machine Learning algorithm that can detect words from this list in a real time audio stream.
We will go through the full process:
1. Selected a worthy machine learning model
2. Design a signal processing pipeline
Train the model with labeled data
4. Apply the model on the ESP32 processor
5. Test the implementation
All steps will be handled using Edge Impulse Studio that links your USB connected device to a web based cloud solution that runs Google's Tensor flow and various data management tools to organize data and generate code that allow your model to run on your local device.
Bring a labtop with a working Linux distribution, native or as Linux for Windows subsystem. Additional you need a working ESP-IDF development system - If you do not have that already, we have created a github repo with a simple installation process.
We will bring ESP32 modules with microphones for keywork recording / training and to test our implementation. The workshop is a physical rerun of the 5. April 2022 webinar held by Danish Sound Cluster and Edge Impulse https://danishsoundcluster.dk/edge-impulse-for-consumer-audio-applications/
There will be served sandwiches, soft drinks and the / coffee during the workshop and there will be plenty of room for networking.