Revised Jupyter Notebook for Course 3, Section 1.5, TinyML Custom Keyword Spotting (KWS)

Education
1

Hello fellow TinyMLers,

Although the vast majority of the notebooks ran without issue, the Keyword Spotting Notebooks in Courses 2 & 3 of the TinyML Certificate program did not run. From what I can see, my fellow students had the same issues. I was revised the code and now can train and deploy a custom KWS exercise in Section 1.5 of the “Deploying TinyML”.

I have just finished putting together a repo on GitHub with a readme discussion and two jupyter notebooks, one to be run on Google Colab and one to be run on a local install of Anaconda.

I am requesting that the community review the work. I was able to run the notebooks and deploy the code onto the TinyML kit h/w. Please keep in mind that I have a background in h/w and not s/w engineering. I have been learning Python for a little while and this is the first time I have posted work of this level on GitHub. I am sure that there is room for improvement and would value your thoughts and suggestions.

The repo is at:

Thanks,
John

2

Thanks John! We’re actively trying to hire some software developers to fix / update the Colabs and so I’ll make sure to pass this along to them as well!

3

Hello Brian,

Thanks for the reply, and you’re welcome.

Take care,
John

4

Hi,

One observation: Now, in February 2026, the speech_commands train.py sample code used in the 3.5.18 notebook is so obsolete that even the migration tutorials to some less obsolete API variants can no longer be found in the tensorflow online documentation. Most documentation links no longer work. I stumbled upon some difficulties with making it work on a local Apple M2 hardware, so I thought I would take a look if I can upgrade the sample code, fix the tensorflow-metal issues on the way and learn a lot about the inner workings on the way. Unfortunately, this is a major task now since the API changes since 2018 or so are no longer easily traceable. So it would probably better to redesign the code from scratch. I have quit the course session now to see if something changes with the new session starting end of february, but since I until now got zero feedback to any questions on this course I am not quite sure where the whole course series stands now. The traffic profile on this site does not make me too optimistic either. So where do we stand with the topic? Is there a new course I should take instead to get up to speed with the topic and get a different professional certificate that is not partly obsolete because this one is based on APIs no longer very useful? As far as I can see, there have been a couple of transitions that are right now not covered in the courses: 1.The transition to Tensorflow beyond 2.14, 2. the transition to keras 3, 3. the transition from Tensorflow Lite to LiteRT. What else is missing?

Regards

Peter