Organizations Seek to Minimize Coding and Transcoding Issues in an Increasingly Digital World
Bit-reduced digital audio. How many times can it be compressed and decompressed to send it where it needs to be, without sacrificing quality?
It’s a question National Public Radio has wrestled with for a while and seems on the verge of improving. Its programming and distribution partner, Sirius, is studying the issue.
Long before word came out in May that the standards-setting NRSC was questioning the performance of Ibiquity Digital’s PAC codec for low bit rates, other radio organizations have been struggling with coding questions – in fact, they’ve done so for the past decade.
“Everybody should get over the fact that digital audio at rates below around 96 kilobits per second is CD-quality. The least objectionable audio is what we are going to have to live with at rates lower than 96 kbps.”
So said Mark Kalman, vice president of Sirius’ national broadcast studio, beginning a discussion about audio coding and transcoding issues for distribution and broadcasting during NPR’s recent Public Radio Engineering Conference.
Kalman shared results obtained by Sirius, which uses the PAC audio encoder developed by Ibiquity Digital and statistical multiplexing, giving Sirius the ability to allocate data capacity intelligently across the range of audio streams, placing more data on channels with higher instantaneous demand. Sirius allocates approximately 60 kbps to a stereo music channel.
“A major source of audio artifacts is noise in the original source material,” Kalman said. He suggested that carefully pre-processing audio to remove all sources of noise, with gating or other techniques, can improve the delivered sound quality, removing much of the “swishy” noise associated with digital audio artifacts.
Sirius is conducting research into transcoding, the concatenation of audio coding in the distribution chain. The effects of multiple passes of audio coding have not yet been well documented, although examples of problems are well known and have been widely discussed in broadcasting.
Privately, some station engineers have complained about the audio quality of NPR programming on Sirius channels. Kalman said Sirius was not satisfied with the initial sound quality on these channels, but said some improvements had recently been completed.
The goal of the company’s research is to identify which types of digital coding cause the most degradation of audio quality when run in tandem, so that these combinations can be avoided. Kalman also suggested that identifiers could be attached to digital audio files to inform subsequent audio processing about previous generations of audio coding, allowing them to adapt.
Finally, Kalman said audio dynamics processing and limiting have an adverse affect on low-bit rate coded audio. “More processing equals worse audio,” he said.
Marty Bloss, director of distribution technology for National Public Radio, is developing the next-generation satellite distribution system that will replace one deployed in 1995. Through the use of non-real time distribution, “the new system may be able to eliminate some of the transcoding that currently exists in the satellite channel,” said Bloss.
To further demonstrate why the network is concerned about audio transcoding, Shawn Fox, engineering manager of NPR West, described how a typical digital audio file moves through the production system. Starting with the reporter, audio is compressed at a 6:1 ratio when recorded on the standard MiniDisc format used by NPR.
In the Washington headquarters, this audio is converted to analog and compressed again as it is dubbed into NPR’s desktop digital editing system. Depending on the number of edits a particular story must undergo, this analog conversion and digital compression can happen several times before the story is assembled for distribution to affiliates.
An additional 6:1 compression then is added via the satellite system. All of this occurs before the audio reaches an affiliate station, which may add further audio processing and digital compression of its own.
Fox said MP3 audio files, popular because they can be sent via e-mail easily, have been a particular source of audio problems.
Reducing the number of generations of analog conversions and digital compression is important, said Fox. But there also is a need to continue to educate program producers on how to avoid this problem at all stages of production.
Because most program producers have no means to test the final sound quality once it has been finally distributed, they often don’t hear the problem until it is too late.
Better things ahead?
Program producers cannot detect the noise in their headphones while they’re editing, said Fox, who described an example. An NPR engineer called Fox from his car on the way to work to complain about an audio file that sounded bad. The engineer had heard it over the air from a Washington NPR affiliate; that last generation of digital compression caused the file to go over the edge. But at the point of final edit, Fox said, the file had sounded okay. The issue is complicated and training is important, said Fox and other engineers at the session.
“IBOC is likely to make the problem worse since it adds another non-linear channel,” said Fox.
Kalman believes the industry is only beginning to explore the limits of digital audio encoding.
“Digital audio encoding below 96 kbps is an art,” said Kalman. As the technology progresses and more experience is gained with low bit rate digital audio, the hope is that improvements will continue.