onnxruntime issue on raspberry pi zero 2w

[rpatapa]

Hi,

I was trying to install EfficientWord-Net on raspberry pi zero 2W.

I am having trouble installing the prerequisite - onnxruntime !

$ python test.py

Traceback (most recent call last):
File "/home/aneka/test.py", line 2, in
from eff_word_net.streams import SimpleMicStream
File "/home/aneka/ewn/lib/python3.9/site-packages/eff_word_net/streams.py", line 4, in
from eff_word_net.engine import HotwordDetector
File "/home/aneka/ewn/lib/python3.9/site-packages/eff_word_net/engine.py", line 8, in
from eff_word_net.audio_processing import First_Iteration_Siamese, ModelRawBackend, Resnet50_Arc_loss
File "/home/aneka/ewn/lib/python3.9/site-packages/eff_word_net/audio_processing.py", line 8, in
import onnxruntime as rt

ModuleNotFoundError: No module named 'onnxruntime'

$ pip install onnxruntime

Looking in indexes: https://pypi.org/simple, https://www.piwheels.org/simple

ERROR:` Could not find a version that satisfies the requirement onnxruntime

ERROR: No matching distribution found for onnxruntime

Any suggestions?

[TheSeriousProgrammer]

Soon we will have a tflite model with much lower sizes, once that is created the current models would be deprecated, stay tuned

[rpatapa]

Thank you,

I am not sure if you plan to continue resnet based detection or not, if yes, in the current version, its functionality does not work for new wakeword training. If the resent is going to exist in tflite based approach, please take a look at the associated issues so that it works too..! Thanks for noting!

[TheSeriousProgrammer]

let me do one thing, will try to incorporate tiny classifiers on top of the embedding model to increase the robustness of the prediction, this should be fairly simple to make , will update soon

[Jotakak-yu]

let me do one thing, will try to incorporate tiny classifiers on top of the embedding model to increase the robustness of the prediction, this should be fairly simple to make , will update soon

Hi, is there any progress?