In addition to the reply by G. Martens, here is an excerpt from Shen et al., 2010 ( http://www.znaturforsch.com/s65b/s65b0067.pdf ):
"Even though Raman spectroscopy is not a quantitative technique, the area integral intensity ratio of D and G peaks (ID/IG) derived by the fitting of separate Gaussian curves usually serves as an indicator of the quality trend of diamond-like material. However, conflicting reports still exist on the diamond-like films analyzed by the fitting method of Raman spectra. In this work we show that the quality trend of the diamond-like/graphitic carbon ratio is critically dependent upon the boundary conditions of the analytic regions. We also examine the possibility of directly applying the peak height ratio HD/HG as an alternative method to analyze Raman spectra. The method based on peak height ratios (HD/HG), and an established method using ID/IG ratios can give similar results depending on the choice, in the latter method, of the boundary conditions used for integration. However, the method based on the determination of peak heights avoids the arbitrary assignment of integral boundary conditions while additionally generating a data set that shows statistically smaller standard deviations than the commonly used integration method".
Indeed in any case, what is required in first instance is an efficient partitioning solution for highly Gaussian mixture modeling. Here is for example a candidate solution by Polanski et al., 2015: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4521892/pdf/pone.0134256.pdf