Calculating masking threshold of every critical band by auditory masking model, then quantization steps are controlled by masking threshold and value of audio carrier of the embedding points.
The subtracter (230) calculates the difference between the output of the adder (250) and the quantized value of the output of the adder (250) to determine the quantization error.
The quantization step is adjusted through the neural network according to the statistics of the blocks, and then the watermark is embedded on the singular value of the blocks.