Criptosistema esteganográfico de audio, basado en atractores caóticos y compresión de textos por medio de redes neuronales

Abstract

This research proposes an audio steganographic cryptosystem that integrates intelligent text compression through neural networks with encryption based on chaotic attractors. The proposed system leverages LLMLingua to reduce input message size while preserving semantic integrity and applies logistic chaotic maps to generate pseudo-random keys for XOR-based encryption. The compressed and encrypted message is embedded into WAV audio files using least significant bit (LSB) manipulation, via sequential or random insertion guided by chaotic sequences. Performance was evaluated through objective metrics such as PSNR, MSE, entropy, and statistical tests, showing high imperceptibility and resilience against compression, filtering, amplitude scaling, and echo attacks. However, it exhibits weaknesses under Gaussian noise and aggressive resampling. This work contributes to digital security by offering a robust and adaptable model for secure information embedding in audio signals within real-world communication scenarios.