Urbansound8K audio classification

Feature requests
#1

Hi,

I ingested part of Urbansound8K audio dataset to Edge Impulse https://studio.edgeimpulse.com/public/7135/latest

I am trying to clasify sound with a Thunderboard Sense 2. Some All .Wav files are normalized in frequency but I still get errors training the model with some files and in the end I don´t get a great accuracy. Could someone have a look to see if a better classifying accuracy is possible?

#2

Hi @iarakis,

I see that you used the EON Tuner which is a great idea! 82% accuracy is actually good for the dataset. With 10 classes, you’d need more data to be able to improve the model. Especially class 6 has very low accuracy but only 4 min of data.

Could you ingest more samples, or try with less classes?

Aurelien

1 Like