I work with automatic environmental sound recognition. As a limitation of my work, I use only audio without other contextual information to do the recognition.
In nature, many sounds are very alike, for example a bouncing ball or knocking on the door . This similarity between sounds, often confuses us. Thankfully, context can help us to disambiguate.
Many people have worked with automatic sound recognition. Once, I read that in a given experiment, when they asked people to identify sounds, the subjects stated that by listening to only 3 seconds of the sound, it was sometimes was not enough.
In another paper I read, they were stating that in some cases, humans need to listen for approximately 2 seconds to give meaning to the incoming sound. Sadly, I don't recall the reference.
Does anyone know if this is true, and can back it up with a reference?
 Disambiguating sound through context, MA. Niessen.
 Acoustic Environmental recognition. L. Ma