Techniques for automatically controlling the expressive behavior of a virtual musical instrument by analyzing an audio recording of a live musician are provided. In some embodiments, an audio recording may be analyzed at various points along the timeline of the recording to derive corresponding values of a parameter that is in some way representative of the musical expression of the live musician. Values of control parameters that control one or more aspects of the audio playback of a virtual instrument may then be generated based on the determined values of the expression parameter. Values of control parameters may be provided to a sample library to control how a digital score selects and/or plays back samples from the library, and/or values of the control parameters may be stored with the digital score for subsequent playback.