A Matlab project uses Signal Processing Technique — Digital Watermark to hide data in an audio track.

Digital Watermark is a technique that embed specific information into digital signals like audio tracks, video films, or images. As to hiding information in music, it is called Audio Watermarking, using Masking Effect to make audience difficult to hear the information under the music. Moreover, there are two kinds of typical watermarking schemes: DC Watermarking and Frequency Watermarking, which one is based on Temporal Masking and while another is based on Frequency Masking.

A “weak” sound is difficult to be heard while a “strong” sound is also being played.

• Temporal Masking - Amplitude (Volume)
• Frequency Masking - Amplitude in Frequency

DC Watermarking puts the information into the part of music, where the magnitude is lower than the perceptual threshold of human ears.

For Frequency Watermarking, since human ear has limit on distinguishing audio frequency, it is possible to find out a frequency band loud enough while the frequency near it are rather quiet. Therefore, we replace the frequency band with low magnitude to our information to make the watermark.

We implemented Frequency Watermarking in this project to achieve our goal.

The original soundtrack we used in this project is:

• Length: 30 seconds
• Sampling Rate: 44100 (1/s)
• Frame Length: 512 samples
• Original Sound Track

The image that we put in:

• Size: 150*150 pixels
  

Frame the audio track and do FFT, and find out the frequency threshold.

Original Frequency Spectrum

Frequency Spectrum after In-audible Removal

Watermarked Sound Track

Extracted Information (Image):

Store more bits in one “slot” of frequency:

The reconstructed image would inevitably follow with some noise due to the translation loss of doing FFT and IFFT. Moreover, the number of bit in frequency replacement will influence the result.