For high-frequency digital signals to be transmitted over long distances, the basic digital signal needs to be modulated with a high-frequency carrier. In this case, the baseband digital signal is called the passband signal. In particular, in wireless communication systems or ultraviolet or infrared communications, transitional band digital modulations are used. The most commonly used transition band modulations are ASK (Amplitude Shift Keying), FSK (Frequency Shift Keying) and PSK (Phase Shift Keying) modulations. In this study, 8 bits of all digital baseband data were obtained from the transition band modulations in ASK, FSK and PSK modulations in MATLAB. Also, the noises ranging from 5 dB to 25 dB were added to these ASK, FSK, and PSK modulations. The originality of this paper is to a single deep learning model to demodulate the ASK, FSK, and PSK modulations by using a data-driven approach. The main aim is to demodulate the baseband numerical data from the transition band noised modulation signals instead of the hardware demodulator circuits. For this aim, the noised modulated signals were applied to deep LSTM (Long short-term memory) model without feature extraction. The performance measures to evaluate the proposed deep learning-based demodulator method have been used, and they are MAPE, MSE, R2, RMSE, and NRMSE. The obtained MAPE demodulation results for the worst case of ASK, FSK, and PSK (added 5 dB to these modulations) are 4.392, 5.60, and 3.166, respectively. The experimental results have demonstrated that the proposed LSTM demodulator model could be used safely in the demodulation of ASK, FSK, and PSK modulations in the real world.