Hi, In my project “noise 2s dataset”, it is obviously to see that I use the provided neural network training in the “NN classifier” interface, and the effect of training and testing is not optimistic. How to improve the performance of the model in model design through edge impulse? Any better suggestion? thanks!
Trying increasing the number of epochs you train. See if the accuracy increases gradually if the number of epochs increases.
Another suggestion would be try running EON tuner which will provide you best models for your dataset which you can use as the primary block in your project.
From my personal experience, For initial stage of tuning these would help. Hope the experts might like to add a few more.
Thanks and Regards,
Ramson Jehu K
I took a look at your project, and it seems that your audio files are pretty quiet. I recommend using an MFE block (for non-voice audio data processing) and setting the noise floor to -100 (instead of the default -52). The noise floor sets any audio values to 0 that are below a certain volume threshold. Setting it to something like -100 should keep all the information in quiet samples.
Hope that helps!
Your dataset is quite big so it reaches our 20min compute time limit. I just increased it to 60min on your project, let us know how it goes.
And the project ID 45905 also needs to increase it !Thank you so much !
We have unlimited compute time as part of our enterprise subscription. We can increase it for some free specific projects but not all. Can you tell us more about your use case, which company your work for? Feel free also to email us at firstname.lastname@example.org and we could plan a call with our sales team.
Thank you for your reply
I just want to find a suitable model in the EON TUNER in the noise 2s data project, but it always shows the error of exceeding the running time, even if it is adjusted to 60 min, it is still the same, what should I do?