A stochastic, hybrid computational fluid dynamics/computational combustion acoustics approach for combustion noise prediction is applied to the PRECCINSTA laboratory scale combustor (prediction and control of combustion instabilities in industrial gas turbines). The numerical method is validated for its ability to accurately reproduce broadband combustion noise levels from measurements. The approach is based on averaged flow field and turbulence statistics from computational fluid dynamics simulations. The three-dimensional fast random particle method for combustion noise prediction is employed for the modelling of time-resolved dynamics of sound sources and sound propagation via linearised Euler equations. A comprehensive analysis of simulated sound source dynamics is carried out in order to contribute to the understanding of combustion noise formation mechanisms. Therefrom gained knowledge can further on be incorporated for the investigation of onset of thermoacoustic phenomena. The method-inherent stochastic Langevin ansatz for the realisation of turbulence related source decay is analysed in terms of reproduction ability of local one- and two-point statistical input and therefore its applicability to complex test cases. Furthermore, input turbulence statistics are varied, in order to investigate the impact of turbulence on the resulting sound pressure spectra for a swirl stabilised, technically premixed combustor.