Psychoacoustic model refers to the information that an audio encoder decides to keep vs. the information it throws away when it encodes an audio waveform using a lossy compression algorithm. The models make assumptions about what human listeners will perceive. Audio encoding programs often differentiate themselves based on their psychoacoustic models and much research is focused on obtaining optimal models for many different types of audio.
Implementations so far
Both of the above were rated highly in recent listening tests.