System, Methods, and Apparatus for Equalization Preference Learning

Equalization Preference Learning Algorithm NU 2008-087 Equalization Preference Learning Algorithm with Transfer and Active Learning NU 2012-039 Inventors Bryan Pardo Andrew Sabin Darren Gergle David Little Alexander Madjar Abstract Northwestern researchers have developed an algorithm that rapidly determines a person's optimal sound quality, i.e. their desired equalization curves, without direct manipulation of a multitude of equalization controls. The process involves several steps. First, a reference sound is modified by a series of equalization curves. After each modification, the listener indicates how well the filtered sound exemplifies the target sound description (e.g. a "warm" sound). The algorithm generates a weighting function which modifies each channel based upon the user's response. This approach may be used to generate a filter for any particular electronic device, altering the frequency spectrum of a desired sound. The clear benefit is that this technology doesn't require the user to modify technical and complicated audio controls. In addition, some of the inventors also developed an additional ability to estimate the preferences of a particular user's sound quality based on previous user feedback. By comparing those ratings with prior users' preferences, the program can quickly tune to an audio quality that is ideal for the current user. Applications o Phones o Audio devices (e.g. stereos, audio plug-ins, etc) o Hearing aids Advantages o Intuitive user control to manipulate audio output IP Status Issued US Patent No. 8,565,908