We’ve been building towards this project in the previous set of videos. And we’re now ready to build our very own DIY Alexa!

All the code for this project is on GitHub - https://github.com/atomic14/diy-alexa

What are we building - 1:15 Wake Word Detection - 2:27 Command Recognition - 11:47 Digging into the code - 16:30 What’s life all about Marvin? - 21:52

To detect the wake words we stream audio from either an I2S microphone or from the ADC. The wake word detector looks at a 1-second window of audio. The spectrogram of the audio is calculated and fed into a TensorFlow Lite model.

Once we detect the wake word we stream the audio up to wit.ai to recognise the user’s intent.

It works surprisingly well for such a small model, there are improvements that could be made with more training data.

I’ll leave the access token for wit.ai live for as long as I can, but at some point, you will need to generate your own wit.ai application.

Let me know how you get on in the comments!

Related Videos:

Audio Input

https://youtu.be/VVGln_-XEgI https://youtu.be/3g7l5bm7fZ8 https://youtu.be/pPh3_ciEmzs

Audio Output


And TensorFlow Lite for machine learning


Components you could use:

MAX98357 - https://amzn.to/3cg88Z5 TinyPico - https://amzn.to/3vVoONp INMP441 I2S Microphone: https://amzn.to/3cicuiv ICS-43434 I2S Microphone: https://www.tindie.com/products/21519/ ESP32 Dev board: https://amzn.to/3gb6fyc Analogue Audio Amplifier: https://amzn.to/3pxkEJr Speakers: https://amzn.to/3pjWFgq

Want to help support the channel? I’m accepting coffee on https://ko-fi.com/atomic14