Audio Data Hiding That Is Robust with Respect to Aerial Transmission and Speech Codecs

A technique for audio data hiding by using subband amplitude modulation was evaluated by computer simulations in terms of robustness with respect to the cumulative effects of reverberations, background noise, and encoding and decoding with a speech codec. Speech signals from 22 speakers and signals from 100 pieces of music were used as the host audio data. Computer simulations revealed that speech and music signals with background noise and reverberations were able to transmit at least 80% of the embedded data at 8 bps after encoding and decoding using the Adaptive Multi-Rate (AMR) speech codec at a bitrate of 12.2 kbps. In actual measurements in a reverberant room using a cell phone terminal, data hidden in a 12.2-kbps AMR recording having a signal-to-noise ratio of 20 dB was detected at a mean bit error rate of 15 %. The sound quality degradation induced by data hiding was objectively measured using the perceptual evaluation of speech quality (PESQ) and perceptual evaluation of audio quality (PEAQ) algorithms. The average PESQ score of the speech signals approximately corresponded to a subjective evaluation of ‘fair’, while the average PEAQ score of the music signals corresponded to a somewhat lower subjective evaluation of ‘slightly annoying’.