![]() I am using Arduino because I am lazy and they make it easy to record to SD and add in LoRa. This seems to be especially true when then siren is quieter so the background silence gets amplified more during normalization. The MFE based model appears to be better at blocking false positives, but it is not as good at picking out sirens. When I run it against the training data, the accuracy is much less. The accuracy of the MFE based model is only a few percent lower than MFCC. Like I tried switch from MFCC to MFE and I am seeing similar things. It does have occasional false positives, like when trucks are back up, people singing and jazz (saxophone seems similar) ![]() I have it work very reliably and it is able to pick up sirens even when they are very faint. I think it is because most similar sounds do not last as long. ![]() Since this is a long, continuous, I have found that using 2 second samples seems to do a good job at removing false positives. I am trying to build a siren detector that can recognize police and fire engine sirens.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |