radiant mica
Hey,
Anyone else experiencing issues with the google colab script to help make a custom wake word? it mostly seems to be something related to Tensor Flow, but I've been trying to generate a wake work for about a week and it just doesn't seem to go through successfully.
grizzled citrus
Same issue for me
Looking at the Download Data step, it seems as though tensorflow is being uninstalled for whatever reason?
"Could not find a version that satisfies the requirement."
Yea cuz you just uninstalled it! lol
teal dawn
Which model training notebook are you using? It looks like there have been significant updates to the simplified Colab since I last used it in November, the dependencies in the Download section have changed drastically. My experience was plagued by issues of incompatible dependency versions and missing dependency files, but uninstalling the main tensorflow package was still a step in November. I'm going to try this updated version now
So far I've made it through the simplified Colab with no issues. This new version has a few lines in the Download section dependencies list commented out, notably the uninstallation of tensorflow
My instance has made it to the model training section. I see it's only been six days since you tried, but maybe updates were made. I'm using the simplified version straight off of OWW's github
https://github.com/dscripka/openWakeWord?tab=readme-ov-file#training-new-models
jagged dawn
I had this issue, and the updated colab version worked for me. Generated my wake word yesterday successfully.
slim pasture
Which colab script were you using @jagged dawn ? There are two linked on that github (a one called simple and one called more detailed). It's still not working for me (failing at the very last step converting the .ONNX to .TFLITE)
jagged dawn
Which colab script were you using <@641719106592309270> ? There are two linked o...
I followed this guide: https://www.home-assistant.io/voice_control/create_wake_word/
The collab got updated recently and since then it worked for me.
slim pasture
Thank you! It worked for me now too! It must have been fixed yes. 🙏
I wonder if I could also inject my own voice samples instead of only the auto generated ones.. Hmm
But that's a case for another thread 🙂
teal dawn
Keep in mind that these Colabs linked here are for OpenWakeWord, and the Voice PE rune MicroWakeWord, which, in my understanding, has been heavily optimized for ESP32 chipsets. The two models are not interchangeable. I have yet to get the MWW Colab to run successfully, but there seems to be a number of active discussions on the github. For OWW, you'll need a different platform like the ReSpeaker Lite from Seeed
neon abyss
Keep in mind that these Colabs linked here are for OpenWakeWord, and the Voice P...
Everything is right except last sentence. Respeaker Lite works on same software that PE has, and they both actually capable of using OWW. Everything is capable of using OWW. MWW, on the other hand, is dedicated to ESP32.
teal dawn
Thank you for clearing up my comment. I see you all over here, I should have taken more time to maybe mention your posts in other threads explaining that the PE could be reconfigured to use OWW
neon abyss
Thank you for clearing up my comment. I see you all over here, I should have tak...
Sorry for being all over... 🙂
Anyways, everyone just waiting for new approach from Kevin to build MWW models effectively... Probably he will have more time to play with it, since his media_player got merged already. 🙂
hollow venture
Is anyone else still having issues with the simple model training? When I run it I get errors in the last section while trying to convert from onnx to tflite
I get ModuleNotFoundError: No module named 'onnx_tf'
opaque echo
Is anyone else still having issues with the simple model training? When I run it...
I'm also having issues. I just posted my own thread about half an hour ago.