Intro to TensorFlow Lite Part 1: Wake Word Feature Extraction - Maker.io | Digi-Key Electronics

[@JS19190]

Thank you for all the hard work you put into the tutorial! Know that it is recognized and greatly appreciated ^^

[@dps9988]

Nice! thanks for this! I wish there were more major players like you that made these kinds of videos. It really helps to motivate and teach us that are very new to all of this, and I am sure the legends and pros appreciate it too!

[@elvenkim]

Most excellent video and authentic application. I like the way you presented very technically yet very systematic

[@kubal6822]

Hi. Amazing video ;) I am not sure if you mentioned one thing. Which is pretty important if you want to use your own recordings: THE WAV FILES MUST BE EQUAL LENGTH. So the python script cutting silence etc. is necessary ;)

[@CULTURE_dz]

MERCI , THANK YOU , شكرا لك

[@Misterbee0407]

Solid!

[@egoNikitsolo]

Hi, thanks for the great Tutorial. I have a Question, what parameters do I change if my sample is 5 Sec long?

[@eldhokurian4245]

I have the same doubt, did you find any luck ?

[@gaboceron100]

Well done, good project.

[@paintballbob21]

So which mechwarrior 5 mechs are the best?

[@OranCollins]

Tip: id give python pathlib a try! If you haven't already, Its really cool and I find easier to use than os.path. example pathlib.Path().cwd() for the current dir, Neat! Also you can use syntax like pathlib.Path().cwd() / "mydocks" / "file.txt"

[@ShawnHymel]

Good to know, thanks! I've used cwd() before, but I find I just end up using os.path for some reason :-/

[@OranCollins]

​@@ShawnHymel Oh, ok. Interesting video! I really enjoyed your other basics electronics videos good stuff. I was surprised that there is a embedded ml models! I'm really interested in including the speech recognition in my next hardware project! Question: Did you fallow an example for the code you wrote or started from scratch?

[@ShawnHymel]

@@OranCollins Yes please! I'm really hoping that these videos can help people get started making their own ML models to do cool, fun stuff in projects. I think embedded ML is still a very new and growing field. Most of this I wrote myself, pulling from examples here and there. Once I had a basic understanding of MFCCs, I just used a library to compute them (rather than trying to recreate all that math in Python on my own). The only real big example I used (which you'll see in the next episode) is the convolutional neural network that I found here: www.geeksforgeeks.org/python-image-classification-using-keras/

[@OranCollins]

@@ShawnHymel cool. Good work, man. I'm surprised I dont see more embedded ml tutorials. It's cool to be on the cutting edge of stuff like this.

[@ShawnHymel]

@@OranCollins Thanks! Agreed...I think we're going to start seeing more and more. This book just came out at the end of December: www.amazon.com/TinyML-Learning-TensorFlow-Ultra-Low-Power-Microcontrollers/dp/1492052043/ref=sr_1_1. It was written by the authors of the TensorFlow Lite for Microcontrollers library, so you know it's new stuff.

[@nasrkruithof6372]

Hi, can you explain why if I use 3s. length wav files, it print out mfcc.shape 56 while for 1s you get 16. How is that calculated?

[@palliyil]

I think that is a function of your winlen and winstep too. I think its the number of winsteps of sampling each winlen, that would cover your entire sample length

[@nicolasugrinovic2886]

Can you do this with the arduino nano 33?

[@kubal6822]

Watch out! listdir() in Python does not have to put files in alphanumerical order - indexes would be random then and you can accidentally use not intended wakeword. Consider using sorted(all_targets)

[@kubal6822]

or all_targets.sort()

[@austinbernard5241]

how would you go about creating your own dataset and making sure all samples are exactly one second long without manually listening and cropping?

[@Gavy093]

Not sure if you still need it, but the best way would be using a python script. You can set sampling rate, duration and where the output should be saved.

[@IT-jq6lr]

What vesion keras and tensorflow?

[@drbizzarow]

I don't know why are you in so hurry ???

[@carthic4u]

Please change your intro music. Its annoying.