Audio coding has been around the professional sound industry since the early 1990s. Codec developers have been and continue to be on a fast track. Whereas the Motion Picture Experts Group (MPEG) once viewed audio quality to be excellent at 256kb/s and 128kb/s, it now offers the same judgment at much lower bit-rates. It's much easier to improve the data payload, as compared to expanding the pipe.

To accomplish this requires using lower bit-rate codecs. Lowering the bit-rate increases potential degradation of audio performance. Advancement of codec design has allowed lower bit-rates to be employed, and most codecs sound decent at these rates, but they are much more fragile with regards to distortion and susceptible to artifacts. Due to the various types of codecs and lower bit-rates, getting a handle on the issues that annoy these functions is a moving target. The goal here is to seek out the gremlins and offer ways and means to avoid them.

All transmission systems suffer from some form of problem. The key to improving audio quality through a coded system is in locating the challenges and avoiding them. Take the FM stereo system — high frequency distortion and peak level overshoots were very common in early FM stereo generators. Both the pre-emphasis boost and sharp cutoff of the required low pass filters caused severe problems within the system. In-depth analysis of the system lead to the discovery of embedded pre-emphasis management and non-overshooting low-pass filters, which dramatically improved FM stereo performance.

While the concern for FM stereo was distortion and overshoot, coded audio suffers from sonic artifacts. These are the perceptible annoyances that bother the listener. Most sound anomalies are categorized as one form of distortion or another. Most common are harmonic distortion (THD) and intermodulation distortion (IMD). Coding artifacts are neither. When they are perceived, they occur due to inadequacies of the coding algorithm. Basically, this is the point where the encoder runs out of capability to reduce the audio data without the process of data reduction being heard. While there have not been specific technical terms assigned to describe these artifacts, they can be referred to as swishy-swirly, underwater-like, gurgle-like and sometimes synthetic-metallic.

Prior attempts

Dedicated audio processors that utilize look-ahead limiting and bandwidth control improve sound performance, but still do not reduce artifacts enough at low bit-rates, especially below 48kb/s. HD Radio, satcasters, podcasters and netcasters employ bit-rates at 24kb/s and lower. Reducing artifacts at these low rates usually requires severe bandwidth reduction, which in turn dulls the sound quality.

Careful listening to lower bit-rate coded audio reveals discoloration — not necessarily artifact-like or distorted, but some type of degrading ghost-like product being carried along with the signal. Attempts to remove it via signal processing seem to increase this characteristic. Listening to the output of the audio processor prior to the encode/decode section sounds very clean. Upon adding a codec to the scenario, the annoyance returns. This problem is observed with use of a common known codec for HD Radio (HDC) and various audio processors of different designers/companies. All produced the same results.

A clue to the problem is revealed when the timing in one of the audio processors is modified to reduce the amount of fast-limiting applied to presence and high frequencies. (This does not remove the limiting in this spectra, but changes the manner in which the limiter's timing responds to transient signals.) The audio immediately opens, along with clarity in the presence and high frequency range.