Audio Watermarking using plaintext watermark

This Python script implements a watermarking technique to embed and detect a plaintext watermark in an audio file. The watermark can be any text string, which is converted into a binary format and embedded into a host audio signal. The script uses a pseudo-random sequence for watermark embedding and includes functionality to detect the watermark after it has been embedded.

Table of Contents

• Requirements
• Installation
• Usage
• Configuration
• Functions
• Results
• License

Requirements

• Python 3.x
• Jupyter Notebook
• NumPy
• SciPy (for WAV file handling)
• Pydub (for audio processing)

You can install the required packages using pip:

pip install numpy scipy pydub

Installation

1. Clone the repository or download the Jupyter Notebook file.

2. Ensure you have the required .mp3 file at the root directory, which will be preproccessed to get a .files/preproccessed.wav file for watermarking. Alternatively: place a mono channel .wav audio file in the ./files directory and name it preprocessed.wav, this won't require preprocessing.

3. Open a terminal and navigate to the directory containing the notebook.

4. Launch Jupyter Notebook:

   jupyter notebook
   

Usage

To run the script, simply execute it in your terminal or command prompt:

To run the watermarking process:

1. Make sure that the audio file (raw.mp3 or preprocessed.wav) is available in the specified directory. Update the WATERMARK_TEXT variable in the notebook with your desired watermark.
2. Run the cells sequentially.

Configuration

You can customize the watermarking process by adjusting the following configuration variables in the notebook:

• HOST_SIGNAL_FILE: Path to the host audio file (default: ./files/preprocessed.wav).
• WATERMARK_TEXT: The text to be embedded as a watermark (default: 'This is your watermark').
• WATERMARK_SIGNAL_FILE: Output path for the watermarked audio (default: wmark_signal.wav).
• PSEUDO_RAND_FILE: File to store the generated pseudo-random sequence (default: pseudo_rand.dat).
• WATERMARK_ORIGINAL_FILE: File to store the original watermark data (default: watermark_binary.dat).
• WATERMARK_EXTENDED_FILE: File to store the extended watermark data (default: watermark_extended.dat).
• REP_CODE: Boolean to enable/disable repetition coding for watermark robustness (default: True).
• FRAME_LENGTH: Length of each frame for analysis (default: 1024).
• CONTROL_STRENGTH: Embedding strength controlling watermark intensity (default: 0.03).
• OVERLAP: Overlap percentage between frames (default: 0.0, no overlap).
• NUM_REPS: Number of repetitions per watermark bit if REP_CODE is enabled (default: 3).

Functions

fix(xs) Emulates MATLAB's fix function, which rounds towards zero.

embed() Embeds the watermark into the host audio signal. Generates a pseudo-random sequence, saves the original and extended watermark, and modifies the host signal accordingly.

detect() Detects the watermark from the watermarked audio signal. It uses correlation with the pseudo-random sequence to recover the watermark and calculates the Bit Error Rate (BER) and Signal-to-Noise Ratio (SNR).

main() The main function that orchestrates the embedding and detection processes.

Results

After running the notebook, the original and recovered watermark texts will be printed, along with the Bit Error Rate (BER) and Signal-to-Noise Ratio (SNR) of the watermarked audio.

Example Output:

Copy code
ORIGINAL WATERMARK:  This is your watermark  
RECOVERED WATERMARK: This is your watermark  
bit error rate = 0.0 %  
SNR = 30.5dB

License

This project is licensed under the Do What The **** You Want To Public License (WTFPL).

References

• Twenty years of digital audio watermarking
• Detecting Hidden Information from a Spread Spectrum Watermarked Signal by Genetic Algorithm