A beam forming microphone array, uses the phase difference of sounds to isolate sounds from any direction. This is achieved with two microphones at a particular frequency by combining delayed signals so that sounds from a particular are in phase, while sounds from other direction are out of phase. With multiple microphones multiple frequencies may be cancelled, leaving a direction signal across a range of frequencies.
This is useful for a learning, because it allows isolated sounds to be recorded and recognized later.
Sounds are transformed into the frequency domain, which removes phase information which is irrelevant for recognition purposes. The time series of frequency vectors may be compared with recorded sounds by creating a mapping vector from instances in the recorded sound to instances in the new sound.
The information required to describe the new sound in terms of the old sound is the measure of the similarity of the sounds.