We present a parallel, geometric, pattern-matching algorithm that can be used for finding transformed matches of a polyphonic musical pattern, P, in a polyphonic musical dataset, D, where neither P nor D are required to contain any information about the voices or parts to which notes belong. We assume that P and D are represented as sets of k-dimensional points, \(P,D\subset \mathbb {R}^k\) . Given a class, F, of bijections from \(\mathbb {R}^k\) to \(\mathbb {R}^k\) , we define a pattern, Q, to be a maximal transformed match of P in D with respect to F, if there exists a transformation, \(f\in F\) , such that \(Q=D\cap f(P)\) , where \(f(P)=\bigcup _{p\in P}\lbrace f(p)\rbrace \) . Our proposed algorithm returns the maximal transformed matches of P in D with respect to F. If \(|P|=m\) and \(|D|=n\) , then the algorithm does \(O\left( \beta !+\frac{\tau }{(\beta -1)!}(mn)^\beta \log _2m\right) \) work, uses \(O(\tau (|\alpha _F|+k\beta )(mn)^\beta /(\beta !))\) space and has a span of \(O(\beta !+\tau \beta ^2\log _2 m(\log _2 m+\log _2 n - \log _2(\beta !)))\) , where \(\tau \) , \(|\alpha _F|\) and \(\beta \) depend on the transformation class and are typically small natural numbers. We evaluated the algorithm on two musicological tasks: (1) discovering occurrences of the “HAYDN” theme in Ravel’s Menuet sur le nom d’Haydn, on which the algorithm achieved an \(F_1\) score of 0.7; and (2) discovering subject entries in Contrapunctus VI from Bach’s Die Kunst der Fuge, on which the algorithm achieved an \(F_1\) score of 0.93.