In this article, we describe the Variable Markov Oracle and how it can be used in stylistic machine music improvisation scenarios. A Variable Markov Oracle is a data structure capable of identifying repeated subsequences within a multivariate time series. A Variable Markov Oracle symbolizes a time series by maximizing an information theoretic measure. After symbolizing the time series, repetitive structures can be extracted and used for music improvisation. We present a machine improvisation framework, using a Variable Markov Oracle, that is capable of generating novel audio content for either real-time or stored audio input. This work focuses on guided improvisation and structured improvisation.