Autism is a neurobiological development disorder experienced by a person from birth or toddlers. Autistic sufferers have difficulty in forming social interaction, communicating, emotional, sensory and motor disturbances as well as slow or abnormal mental development. Therefore, people with autism will be isolated from normal human beings and enter the world of repetitive, activities and obsessive interests. Autism therapy using dolphins that produce ultrasonic waves with a certain frequency can align the function of motor and sensory nerves of patients with autistic. A direct therapy using dolphins requires a special time and a cost is quite expensive for once therapy. This research proposes autism therapy with artificial dolphin ultrasonic waves based on signal characterization issued by dolphins. The main difficulty in this characterization is to eliminate the unwanted signals (noise) from the desired signal when recording the signal issued by the dolphins. Noise filtering (denoising) is done using wavelet transform. This research proposes a model of a dolphin ultrasonic wave characterization using wavelet transformation by determining decomposition and threshold levels to obtain the best signal quality for autism therapy. The effectiveness of the proposed method is measured using the SNR parameter. The results showed that the level of decomposition 5 and hard threshold gave the highest SNR value compared to the others.