I am thinking about how to detect human screams without turning ordinary sounds into alarming false alerts. I want the system to distinguish emergency sounds from ordinary sounds, rather than treating loudness by itself as evidence of danger.
I want positive samples to include Human screaming, distress screams, panic sounds.
I want negative samples to include Laughing, shouting, talking, singing, music, traffic, and children playing.
I do not want the system to treat loudness alone as evidence of danger: loudness alone does not mean danger.
I see a design risk that a joyful shriek could be mistaken for panic.
I want the system to learn the difference between emergency and ordinary sounds.