Loading...

Robust Speech Steganography Using Deep Learning

Raeisian Dashtaki, Ebrahim | 2024

0 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 57502 (05)
  4. University: Sharif University of Technology
  5. Department: Electrical Engineering
  6. Advisor(s): Ghaemmaghami, Shahrokh; Gholampour, Iman
  7. Abstract:
  8. With the expansion of digital communications, the need for secure methods to transmit information confidentially has increased. Steganography, as one of the effective solutions, faces challenges in terms of quality, capacity, and security. This research presents novel schemes for steganography and steganalysis that utilize deep learning for embedding images into audio and detecting the audio carrier of the message. In the proposed steganography method, the image is embedded at the bit level in the frequency domain of the audio file. For this purpose, a structure with multiple paths between the layers of the network, utilizing the Inception-A network, is proposed, which improves processing and facilitates gradient flow. To enhance the embedding rate, image compression in the YUV color model is employed, reducing image data by up to 50. The proposed steganalysis network, also using new blocks similar to the Inception structure, analyzes the audio features in the frequency and cepstrum domains. These features are processed in stages, allowing for deeper analysis. The results indicate that the proposed method achieved the best signal-to-noise ratio for the carrier audio at 39.6 dB and a peak signal-to-noise ratio for the image at 35.5 dB. The highest data transmission rate reached 19 kB, and the highest accuracy in detecting the message-carrying signal was 59.5. The signal-to-noise ratio for the carrier audio ranged from 26.9 to 39.6 dB, and the peak signal-to-noise ratio for the image varied between 17.22 and 35.5 dB. Compared to other methods, the proposed scheme demonstrates better performance in terms of carrier audio quality, extracted image quality, and security against steganalysis
  9. Keywords:
  10. Information Hiding ; Signal Processing ; Steganalysis ; Audio Watermarking ; Steganography ; Robust Speech Steganography

 Digital Object List

 Bookmark

...see more