Saturday, June 30, 2007

A processing artifact that puzzles me

I was listening to WLYT (Charlotte) HD2 this morning ("The Shuffle Channel", and heard something that truly puzzled me.

If you're familiar with the Bangles song "Eternal Flame", you know it's a lovely ballad in 4/4 time. At the first of the song, and at various points throughout, the fourth beat in each measure is punctuated with a "ding" on a bell. "Close your eyes...'ding'...give me your hand, (ding on the start of the word) Darlin', can you feel my (ding on first of word) heart beating....."

I've heard that song a million times. So imagine my shock and confusion when, listening this morning through headphones to this 48kbps HD2 stream, the "dings" were completely gone. THEY WERE NOT THERE! They were replaced with a very slight high frequency transient, which I wouldn't have noticed if I didn't know what belonged there.

Now I have commented earlier that the processing (audio compression and limiting) are too damn aggressive on "The Shuffle Channel". What I wonder is if this is the work of a Neural pre-processor, or similar device, tweaking the signal to minimize (data) compression artifacts. I know that all limiting softens transients. Cymbal crashes never have the SPLASH after going through the old Optimod (insert favorite processor here). But I have NEVER, in 33 years in radio, heard instruments DISAPPEAR from a mix. Anybody else ever noticed anything like this?


Anonymous said...

I have a theory, do you know what data rate they're using?

A lot of the channels on XM have zero high frequency response. Since the HD codec is based on aacPlus, which XM also uses, it seems possible that the first place frequencies would disappear would be in the higher ranges like that bell.

You said it yourself. You wouldn't have missed it if you didn't know it's supposed to be there. Perhaps the HDC is working as intended. I'm always amazed at how good the HDC sounds at 48k. At that data rate, or lower, you're going to be giving something up. In this case, it's the detail that makes the bell sound like a bell.

Mike Walker said...

Sorry, but nobody who's listened to HD Radio would believe that there's no high frequency information. FULLY EXTENDED highs is one of the most immediately noticable things about HD Radio. With no high frequency pre-emphasis/de-emphasis as on the analog channel, FM radio (AM too) can finally broadcast hf content at FULL modulation. Highs can be just as "loud" as midrange and lows, without sacrificing loudness. This has NEVER been the case before. Traditionally FMs could either be bright, or they could be loud. Because of the EXTREME boost required by the 75us pre-emphasis curve, they couldn't be both. Raising average level ("loudness") REQUIRED curtailing highs. BUT NO MORE!

Besides the bell on the Bangles song is fully audible on AM RADIO...even on a typical radio, with response to only 5khz. People have some strange notions about high frequency response, anyway. Many feel that if frequency response is limited to, say, 8-10khz, INSTRUMENTS will be missing. Dude, there are NO MUSICAL INSTRUMENTS with fundamental notes as high as 8khz. 2-3khz is about it. What's "up there" are multiples of the fundamental "note"...that give individual instruments their particular character, as well as "air" and "crispness".

If you think limiting high frequency response (which isn't even done with HDC) causes instruments to disappear, you should read some more about psychoacoustics, and how lossy compression works. Hint: lossy compression does TWO things. The first is the reason they all have to pay a royalty to Dolby Labs...lossy compression allows noise levels to rise in the presence of strong content of a similar frequency which would mask the noise. Second, lossy compression applies a comb filter...dividing the audible spectrum into multiple bands. The codec then makes choices, moment to moment, about which bands have audible content, and which can be ignored. Even this couldn't cause instruments to disappear, as these decisions are made MANY TIMES PER SECOND. Lossy compression is largely effective because it's a moving target. What SPECIFICALLY it's doing to the signal to reduce necessary bitrate changes at a dizzying rate.

When artifacts are audible, and they certainly can be at low bitrates, they take the form of "swishy", "swirly" forms of distortion, usually on instruments with lots of high frequency content (cymbals, muted trumpets, vocal sibilants). The reason for the "swirly" effect is that the comb-filtering applied by the codec is CHANGING moment-to-moment, and it's audible effect is changing as well.

It's actually possible to show EXACTLY what a lossy codec is removing from a signal, by doing what's known as a "null" test. Take an uncompressed .wav file of a favorite recording. Save the file. Now invert the polarity of that file, and save it in the compressed format you're exploring. Copy and "mix-paste" the inverted (compressed) file on top of the original file in a program like Adobe Audition. If the files are still of exactly the same length (same number of frames and samples), then what's common to the two files will be cancelled, and what's left will be only what the codec removed! You can actually HEAR the difference between compressed, and uncompressed.

What will be left will be a weird, ghostlike "whisper" of the original. And with a good codec, you'll have to turn the volume WAY up to hear it, because it is very, very quiet. It's almost not there. Which proves that good codecs remove almost nothing. Science trumps subjectivist bs. So called "golden ears" can take all the "I hear a slightly warmer bass", and "I hear less "air" and delicacy in the treble", or even "the bell disappeared because of the HDC codec", and shove 'em where the sun don't shine, because SCIENCE proves they're full of it. A null test shows that almost nothing is removed by a good codec, properly applied.

44.1khz linear PCM digital audio (cd format) can accurately capture audio at EVERY frequency between DC and 20khz simultaneously. It requires a bitrate of nearly one and a half megabits per second! And it's completely unnecessary, because energy at all frequencies simultaneously is the definition of WHITE NOISE! Real music has energy at specific fundamentals and harmonics, a few dozen, or in complex music, a few hundred simultaneously. THAT requires FAR less bandwidth to FULLY capture!