Batch crystallization is an important process in many industries, for example, fine chemicals, foods, and pharmaceuticals. On-line measurement of process parameters, such as crystal size distribution (CSD), crystal shape, solid concentration, nucleation, and growth rates, is important for process understanding and for product quality control. In this paper, L-glutamic acid crystallization process parameters (CSD, concentration, and the onset of crystallization) are monitored in-situ by ultrasonic attenuation spectroscopy, showing that ultrasonic attenuation spectroscopy has significant advantage over existing particle sizing methods such as laser diffraction when measuring concentrated suspensions. However, as currently implemented, the CSD and concentration analyses are conducted off-line because of the complexity of the analysis procedure, which requires lengthy calculation and a large set of physical parameters describing solid and liquid phases. A novel inversion method based on two neural networks working together is developed to determine CSD and concentration directly from ultrasonic attenuation measurements during crystallization. The neural network to recover CSD has 50 hidden neurons and the other with 20 hidden neurons approximates the reference intrinsic attenuation. This approach has been used to obtain parameters of the crystallization process in simulations and experiments. Compared to current analysis methods, neural networks supply solutions essentially instantaneously and without the need to know physical parameters of the solid and liquid phases.