Audio restoration method and system based on low consistency dictionary and sparse expression

The invention discloses an audio restoration technical scheme based on low consistency dictionary learning and orthogonal matching sparse coding. The method comprises the steps: carrying out the overlapped framing of an original audio signal through a window function; carrying out the multiplying of a mask matrix and the original audio signal, and dividing the original audio signal into a reliable audio part and a damage audio part; enabling the reliable audio part to serve as a training set, and learning one dictionary through a dictionary learning method based on singular value decomposition; carrying out the clustering of atoms with the bigger correlation in the dictionary, carrying out the decoupling of each two clustered atoms through proper rotating conversion, and finally obtaining a low-consistency dictionary; solving a sparse decomposition coefficient through an orthogonal matching algorithm, carrying out the repairing and restoration of a damaged part frame through sparse reconstruction, obtaining a