Audio Sample Rates: Terminology vs Application

I’ve recently started to develop a grumpy streak with respect to the use of certain terminology with respect to telephony. Maybe telephony isn’t exactly the right word, let’s say that my unease arises from some odd terms surrounding audio quality in the context of communication. I think that some of the language needs to be more application sensitive.

A few years ago the world was a simpler place. The “Plain Old Telephone System” (aka POTS) was definitely a narrowband medium. Where “narrowband” implied digital sampling at 8 KHz and a useful audio channel of 300 Hz – 3,400 Hz.

Some people using more advanced systems enjoyed “wideband” audio. Wideband in that case was defined as a 16 KHz sampling yielding a useful audio channel of 50 Hz – 7 KHz. The TIA-920 standard, which I have referenced previously, spells this out. A variety of audio codecs can deliver this capability including; AMR-WB, EVRC-WB, G.722, G.722.1, CELT & Opus.

From this point onward the terminology gets less comfortable. Skype introduced their SILK codec which can operate at 24 KHz sample rate, providing a useful audio channel of 12 KHz. Polycom has Siren 14 that’s a little brighter yet. Some people, myself included, have used the term “Super-Wideband” to describe such audio paths.

In a 2009 white paper Polycom refers to the G.719 codec created in cooperation with Ericsson as, “G.719: The First ITU-T Standard for Full-Band Audio.” That codec samples at 48 KHz, which is also the sample rate used by essentially all professional audio and video equipment.

Their use of the term “Full-Band” or more formally, “Full Bandwidth” makes perfect sense to me. G.719 allows for the creation of an audio path that completely addresses the practical realities of human hearing.

My growing sense of unease has to do with application context. Narrowband is plainly an artifact of an early 20th century technical reality. Wideband reflects the 1980s when cassette tapes were commonplace and CD digital audio just emerging. Super-wideband suggests something awesome!

Full-bandwidth, as embodied in G.719, CELT or Opus addresses applications well beyond voice. It addresses the need for “production quality audio” over various networks, whether for radio remote broadcast, interactive gaming, streaming music, etc. For such applications “Super Wideband” isn’t really very super at all. In fact, it’s basically lame.

I still think that there’s merit in the term HDVoice. It unambiguously spells out the application context as voice. I’m not certain that anyone will understand what’s meant by “Super Wideband” when it shows up in someone’s marketing. I think it will only confuse and confound an unsuspecting public.

By the way, while I’m taking the semantic deep-dive, Ultra-wideband has nothing to do with audio coding. Let’s not have anyone calling the silly trend in 24 bit 192 KHz sampled music “Ultra-wideband.” UWB is a term used to describe a technique employed in short-range, high-bandwidth radio transmission.

OK. I feel better. Thanks for that.

P.S. – Be on the lookout for an Opus implementation in Blink from AG Projects. Even in the red-headed-stepchild-version that runs on Windows.

  • Lifespeed

    I hope you’re not really disparaging the improved dynamic range and sound quality enabled by 24 bit audio resolution, which is a definite improvement over the 16-bit 44.1 KHz CD standard – largely a result of the limitations of 80’s technology and the desire to fit a full 74 minute album on one disc.

    192 KHz sample rate may indeed be overkill, but there is little downside given the DACs used to reproduce it can easily operate at that sample rate. 24-bit 96 KHz is perfectly reasonable for good music quality, and most Blu-ray discs already contain 24/48 audio. Which is a marked improvement over previously available sound.

    • mjgraves

      Indeed, I am. The science simply doesn’t support all the assertions about high-sample rate and very high bit-depth audio. In fact, I was thinking about this just a few days ago. More to follow in a coming post.

      • Lifespeed

        20*log (2^16) dictates 96 dB signal to noise best case, at maximum volume. And that is with no room left for dynamic range.

        While we can argue about how much bit depth and sample rate is necessary before improvements can’t be heard, I am sure it is not 16/44.1KHz. Both my ears and discrete-time system theory agree. And let’s not forget the improvements in distortion from more-ideal image filters enabled by DACs operating at higher sample rates.

        Have you ever listened to DVD-Audio or blu-ray on a good stereo?

        • mjgraves

          Yes, I’ve heard SACD and Blu-ray on a very good system. I’ve also worked in studios with very high-end monitoring and good isolation. I accept that 20-24 bits may be an appreciable improvement for some kind of recordings, but greater than 48 KHz sampling seems pointless to me.

          http://www.mgraves.org/2013/07/audio-quality-sample-rates-bit-depth-human-physiology-the-golden-ears/

          OTOH, how much merit is there in > 16 bit audio if we always compress the life out of the music? The process of recording music needs to be improved more than the technological standards of it’s delivery.

          • Lifespeed

            Yes the mastering needs to improve.