At the monthly meeting on January 13th, 2004, the Boston Chapter of the AES was treated to an insightful presentation by Bill Allen, head of Dolby Laboratories. New York office. Bill discussed the principals of the various audio coders/decoders (codecs) that are in common use. Bill described codecs as trying to reduce the redundant and irrelevant data to more non-redundant and more relevant data.

Codecs can be divided into two classes, lossless and lossy codecs. Lossless codecs can provide a compression ration (the ratio of coded to original data rates) of 2 or 3 to 1, whereas the lossy codecs can provide ratios of 10 or even 20 to 1. Lossy codecs are almost universally based upon a model of the human perception of sound and each codec has its own proprietary model. Mathematical techniques are then applied to distinguish the audible from the inaudible sounds, and eliminate the latter. Different codecs may use more or fewer criteria. Finally, proprietary interactive loops are applied to the data to optimize the bit allocation and minimize the final data rate.

Bill then played samples of different codecs, from the same original recording and with the same final bit rate. In order to more clearly identify the differences, Bill played not only the codec output, but also the difference between the codec output and the original signal. The results were very instructive. Bill also played examples of transcoded music, which were consecutively coded/decoded by different codecs. There was substantial degradation.

Bill then described the listening tests used to evaluate different codecs. His example was based upon the Canadian CRC listening tests. He played both the original and codec outputs for a number of different samples and asked the audience to rate the differences as imperceptible, barely perceptible, slightly annoying, annoying or extremely annoying. Only in one case did the audience fail to distinguish the codec output from the original. Bill cited listening tests of a few years ago, which showed that, at a high enough data rate, the degradation from certain codecs was almost imperceptible.

Finally, Bill described a codec for broadcast and recording use, which allows for a number of consecutive coding/decoding cycles without any significant further audio degradation. This is needed whenever the audio signal has to be processed, for example through a mixing console. So nowadays, eight audio channels (5.1 surround plus stereo SAP) can be passed through the same data path that only two uncompressed audio channels used to pass.

The presentation concluded with a lively question and answer session, where many elucidating comments were made.

--Michael Rudd