The World Health Organization (WHO) claims that 0.8% to 1.2% of newborns worldwide are affected by congenital heart diseases (CHDs). There are many methods for CHD identification, and the most prevalent is phonocardiography (PCG). It is a non-invasive method that offers crucial knowledge about the sounds (S1, S2, S3, and S4) and beats of the heart. This research study aims to train a binary categorization system using a deep neural network for CHDs by using a combination of local and public datasets. The local dataset (LD) had 583 signals (normal and abnormal PCG), while the public dataset (PD) taken from Michigan University had 23 PCG recordings. Both datasets were down-sampled to 8 kHz. The band pass filter was designed such that it ensured that any signals outside of the 20–650 Hz range were filtered out, allowing only the desired frequencies to be processed. All signals were chunked at a signal duration of 4 seconds. For data augmentation, pitch-shifting was applied and passed to a 1D convolutional neural network (CNN). The best results were achieved for case C, with an accuracy of 98.56 %, precision of 98.57 %, F1 score of 98.56 %, specificity of 98.0 %, and sensitivity of 99.0%.