The long-term goal we are striving for is to create an optical quantum computer. This task consists of three sub-stages: encoding information in light quanta (photons), causing interactions between photons and reading the encoded information. In our project, we try to implement the first stage by modifying the temporal-spectral properties of one-photon pulses. Unlike traditional pulse shaping based on filtering part of the power, we focus on modifying only the signal phase, which generates significantly lower losses. This is a significant improvement for systems operating with classical light and crucial for quantum systems, which are particularly sensitive to signal decoherence. There is no universal algorithm enabling the calculation of the phase used for any modification of a one-photon pulse due to the non-linearity of this operation and physical limitations. We expect that deep neural networks will be up to the task of emulating this transformation.