To compute the acoustic emission of combustion processes, it is necessary to capture the details of the flow field and the flame dynamics in the computation. In previous studies, the sound emission of various premixed flame configurations was investigated by solving the compressible Navier-Stokes equations with a finite-volume large-eddy simulation method. The finite-volume solver was coupled to a level-set solver to model the flame with a combined G-equation progress variable approach. Using this numerical approach, a good agreement with experimental results was obtained for turbulent premixed jet flames and swirl flames in confined and unconfined burner configurations [1, 2]. In general, this approach is more efficient than, e.g., considering detailed chemisty. However, a reinitialization method needs to be applied to the solution of the G-equation, to maintain its signed-distance property, which conventionally was achieved by a computationally expensive iterative constrained method [3]. In the current study, this reinitialization method is replaced by a machine learning model. The study addresses the following two questions: 1. Which kind of machine learning model is suitable to achieve a high quality of the computed acoustic emission of the flame? 2. Is a coupling with this machine learning model more efficient than the original iterative reinitialization approach? The training data for the models are generated by the conventional approach for a 2D configuration of an acoustically excited laminar flame. The results of the acoustic emission of a lean premixed flame using the machine learning approach is juxtaposed to the results of a computation using the conventional approach.