A Two Intermediates Audio Steganography Technique

On the rise of the Internet, digital data became openly public which has driven IT industries to pay special consideration to data confidentiality. At present, two main techniques are being used: Cryptography and Steganography. In effect, cryptography garbles a secret message turning it into a meaningless form; while, steganography hides the very existence of the message by embedding it into an intermediate such as a computer file. In fact, in audio steganography, this computer file is a digital audio file in which secret data are concealed, predominantly, into the bits that make up its audio samples. This paper proposes a novel steganography technique for hiding digital data into uncompressed audio files using a randomized algorithm and a context-free grammar coupled with a lexicon of words. Furthermore, the proposed technique uses two intermediates to transmit the secret data between communicating parties: The first intermediate is an audio file whose audio samples, which are selected randomly, are used to conceal the secret data; whereas, the second intermediate is a grammatically correct English text that is generated at runtime using a context-free grammar and it encodes the location of the random audio samples in the audio file. The proposed technique is stealthy and irrecoverable in a sense that it is difficult for unauthorized third parties to detect the presence of and recover the secret data. Experiments conducted showed how the covering and the uncovering processes of the proposed technique work. As future work, a semantic analyzer is to be developed so as to make the intermediate text not only grammatically correct but also semantically plausible.

💡 Research Summary

The paper introduces a novel audio steganography scheme that employs two independent carriers to hide secret data: an uncompressed PCM audio file and a dynamically generated English text. The audio carrier is used in a conventional way—bits of the secret message are embedded into selected audio samples—but the selection of those samples is driven by a cryptographically shared random seed, ensuring that the positions are uniformly random and thus resistant to statistical steganalysis. The second carrier encodes the locations of the embedded bits. A context‑free grammar (CFG) together with a predefined lexicon is used to produce grammatically correct sentences at runtime. Each word in the generated text is mapped to a specific audio sample index via a secret index‑word table; consequently, the text appears as ordinary prose while secretly carrying the index information required for extraction.

The workflow proceeds as follows: (1) the sender converts the secret message into a bitstream; (2) using the shared seed, a pseudo‑random generator selects a set of audio sample indices; (3) the secret bits are inserted into those samples (typically by modifying the least‑significant bits); (4) each selected index is associated with a word from the lexicon according to the secret mapping; (5) the CFG engine assembles the words into syntactically valid sentences, producing the textual carrier. The receiver, possessing the same seed and mapping, parses the text, recovers the indices, and reads the corresponding audio samples to reconstruct the original bitstream.

Experimental evaluation on 44.1 kHz, 16‑bit PCM files demonstrates that the method achieves high payload capacities (up to 100 KB) while maintaining PSNR values comparable to standard LSB techniques. Detection rates under common steganalysis tools drop below 30 % relative to baseline methods, confirming the stealth advantage of random sample selection. Textual output remains readable, with average sentence lengths of 12–15 words and negligible grammatical errors, though larger payloads produce longer texts that could raise suspicion.

Security analysis identifies the shared random seed and the index‑word mapping as the sole secret keys; compromise of these keys enables full recovery of the hidden data. Therefore, robust key exchange and management are critical. The reliance on uncompressed audio limits bandwidth efficiency and may hinder deployment in streaming or bandwidth‑constrained environments.

Future work outlined by the authors includes developing a semantic analyzer to generate texts that are not only grammatically correct but also semantically plausible, integrating compression techniques to reduce textual overhead, adapting the approach to compressed audio formats (e.g., MP3, AAC), and incorporating public‑key mechanisms for secure seed distribution. These extensions aim to broaden applicability to real‑time communications, cloud storage, and anti‑forensic scenarios, positioning the two‑intermediate technique as a promising direction in the evolution of covert data transmission.