K. Gopalan (USA)
Audio steganography, psychoacoustical masking, FourierBessel series expansion, phase embedding
A method of embedding covert speech in a cover audio signal for secure storage and transmission is proposed. Covert speech is represented using a truncated set of coefficients in the first order Bessel function expansion of the utterance. These coefficients are quantized to eight bits and the coded values are embedded in the phase spectrum of perceptually masked points in a cover audio signal. The embedded cover audio, or the stego signal, is quantized to 64, 32 or 16 bits for storage and transmission. Initial results show a high rate of embedded data recovery, and hence unaffected covert speech quality, at 64 and 32 bit encoding of the stego signal using a cover speech or music signal. At 16 bits, the message in the recovered covert speech can be perceived with reduced speech quality because of errors in data recovery. Instrumental music signal, with perceptually masked points available in a large number of frames, can embed and recover better quality covert speech. The bit error rate and hence the recovered speech quality can be improved with error detection and correction bits added to the coded covert speech representation.
Important Links:
Go Back
