Auro-3D hides high-frequency sound in the lower bits of PCM—and is sensitive to processing

In Auro-3D, the height channels are not contained in a separate “meta-track” alongside the standard 5.1 or 7.1 channels. The format hides their descriptions and decoding data in the least significant bits of 24-bit PCM, so without a special decoder, the track sounds like regular surround sound, but with one, it can be expanded to 9.1, 11.1, or 13.1.

This is the main trade-off of Auro-3D: compatibility is achieved not through some free magic, but by utilizing the precision headroom of PCM.

The author of the article first discovered the metadata in the silence: even there, characteristic blocks of bits were repeated. It then became clear that the data contained not only layout flags, but also predictor parameters, scales, restoration settings, and residuals—the residual data used by the decoder to reconstruct the height channels.

The scheme is not simply a matter of “adding the top columns to the file,” but rather a controlled mixing process. The original bed and height channels are first combined into a compatible carrier, and then instructions are written to the least significant bits to enable the reverse dematrix/unmix operation. Depending on the mode, two or even three output channels can be reconstructed from a single carrier.

Therefore, Auro-3D differs from Dolby Atmos and DTS:X in more than just its marketing name. Classic Auro is a channel-based format with low and height layers, rather than an object-based system where sound moves freely through space. AuroCX already incorporates an object-based approach, but the old packaging method remains an interesting engineering solution for delivering a greater number of channels over a limited transmission medium.

The practical conclusion is simple: an Auro file cannot be considered ordinary PCM, even if it looks exactly like it. It can be stored in WAV, FLAC, DTS-HD MA, or Dolby TrueHD, but any manipulation of the signal poses a potential threat to the hidden service stream.

The limitations are also significant. The least significant bits are selected in blocks and depend on the signal level: at low volume, the codec uses fewer bits; at high volume, it may use more. Normalization, resampling, volume adjustment, dithering, and improper remuxing can corrupt the metadata. After decoding, the audio is not guaranteed to be bit-for-bit identical to the original: there are quantization errors, rounding errors, and LSB replacement. The author himself refers to the decoder he developed as a beta version, and there is noticeably less native Auro-3D content than there is for Atmos.

In other words, Auro-3D is not a way to generate additional channels “out of thin air,” but rather a precise, reversible—yet fragile—container for mixed audio. The format particularly well illustrates why an audio file sometimes cannot be mindlessly run through a standard processing pipeline: on the surface, it remains compatible, but internally it already contains a second layer of data.

Would you trust such a track to a standard audio editor, or would you first check each processing step to ensure the hidden stream remains intact? Source: All Articles in a Row / Artificial Intelligence / Habr