Building a personalized audio equalizer interface with transfer learning and active learning

Potential users of audio production software, such as audio equalizers, may be discouraged by the complexity of the interface and a lack of clear affordances in typical interfaces. In this work, we create a personalized on-screen slider that lets the user manipulate the audio with an equalizer in terms of a descriptive term (e.g. "warm"). The system learns mappings by presenting a sequence of sounds to the user and correlating the gain in each frequency band with the user's preference rating. This method is extended and improved on by incorporating knowledge from a database of prior concepts taught to the system by prior users. This is done with a combination of active learning and simple transfer learning. Results on a study of 35 participants show personalized audio manipulation tool can be built with 10 times fewer interactions than is possible with the baseline approach.