This is still early-stage demo software, that does not necessarily reflect the performance of the final software.
The demos you will see are videos that contain lots of siren sounds (or none at all) and have been recorded in various environments, with camera/audio equipment unknown to us. The sonoware detection algorithm is designed to run in real-time, but was processed offline on the audio that was extracted from the video. The algorithm output is used to create a video-overlay, that shows if there is siren activity.
The algorithm is currently configured to
All of these demos have been processed with the same set of parameters for the detection algorithm. There was no fine-tuning done to produce better results for different scenarios, i.e. the presented videos were only used for validation and testing. Also it is unknown if the camera-equipment did some (adaptive) audio processing.