There are many corrected versions of AIC which aim at reducing bias. They typically depend on the dimension, the sample size, or the number of covariates in linear regression models. If you want to find a corrected AIC version in your context, you may need to specify more details on your model.
The use of AIC is not recommended to select the number of components in mixture models. See for instance:
Choice of the Number of Component Clusters in Mixture Models by Information Criteria.
Generating Gaussian Mixture Models by Model Selection For Speech Recognition
