This research based project aims addresses the issues in several audio processing activities like lost audio transmissions and signals or corrupted audio files. To solve those issues and for possible enhancement of audio reconstruction, machine/deep learning models fusing steganography technique was explored. Experiments were conducted on corrupted audio files with short and long missing gaps which were filled with reconstructed signals with ML/DL models. The results from the experiments concluded that use of AI algorithms has a great impact on audio reconstruction.
The image below shows the reconstruction of lost audio signals with four different techniques: Steganoflage, Random Forest, SVR (Support Vector Regressor) and LSTM (Long Short Term Memory).