The noise generation by turbulent combustion belongs, similarly to the pollutant emissions, to the negative effects of combustion processes (noise pollution). The superior aim of the current work is to develop numerical tools for prediction of noise radiation from combustion. In order to do this, a prerequisite is to rendering the noise generating acoustic sources involved within the turbulent flame as accurately as possible. For this reason, a model for CFD (computational fluid dynamics) simulation of turbulent combustion has been proposed at first, where the entire turbulent flame is viewed as an ensemble of pre-defined reaction zones with different stoichiometries. The interaction between these reaction sheets and the turbulent flow is described by a modeled turbulent flame speed. The combustion model has been validated for a variety of different flames, where the calculated results showed a good agreement with the corresponding experimental data, so that its further use for calculation of combustion noise is reasonable and straightforward.Different methods have been proposed in order to predict combustion generated noise, among them are compressible LES (large eddy simulation) and a number of hybrid CFD/CAA (computational aero-acoustic) methods. Compressible LES, on one hand, provides the necessary acoustic sources for the CAA calculations; on the other hand, it has been used directly to validate the CAA results. By applying the most fundamental but commonly used Lighthill's acoustic analogy for the CAA modeling, a detailed analysis of the individual source terms in the Lighthill equation has been carried out. By doing so, it has been evidenced that the aeroacoustic source due to the flow motion is basically as large as the combustion generated noise source by unsteady heat release in the non-linear flame region. However, the combustion noise dominates the flow noise in the far field, because they exhibit different characteristics, i.e., monopole for combustion noise and quadrupole for flow noise source. In addition, Lighthill's CAA method is able to reproduce the sound level provided by compressible LES accurately.Subsequently, a modified Lighthill equation with the source term essentially built up by the unsteady heat release has been used for CAA simulation of noise propagation. The calculated sound level thereby compares well with results provided by LES and measurement. Furthermore, the analytical solution of the modified equation of Lighthill, where the heat release given by LES exclusively serves as input argument has been compiled. The sound pressure calculated from this approach showed a good agreement with the original CAA solution in the low frequency range, whereas it was over-predicted in the high frequency domain. This is attributed to the retarded times of the discrete cell sources arriving at the receiver location, which have not been considered by the analytical approach and leads to a spurious noise. The major benefit in using the far field analytical solution is attributed to the fact that there is no need to solve the respective acoustic equation numerically as long as the relevant heat release are only available in the flame region. Hence, the computational cost is reduced significantly compared to other direct or hybrid methods where a computational mesh has to be used, which covers all regions of interest in the far field.The spectral model is presented as the last CAA method, which is derived from the analytical solution above and the fact that the unsteady heat release and the turbulent fluctuation can be directly correlated with each other in the spectral domain. It represents the simplest technique among the demonstrated CAA methods in this work, which solely needs input data from a RANS (Reynolds averaged Navier-Stokes) simulation. Application of the approach for prediction of sound field emitted from a non-premixed jet flame showed a quantitatively good agreement with the measurement and LES solution.