Hacker News Reader

Comment by martin-adams | Hacker News Reader

I worked with an audio engineer on real-time video encoding hardware and he said that in audio compression, you can throw away audio data immediately after a loud noise because the ear can't pick it up. This is probably an oversimplification, but sounds like a similar technique in this safety feature. I like it.

View on HN

This is actually a big part of how MP3 compression works. Rather than using straight up data compression, it uses a psychoacoustic model to simplify different components of the signal that are affected by masking.

There's a nice high-level (human readable) technical overview here if you'd like to learn more: http://arstechnica.com/features/2007/10/the-audiofile-unders...

I should add, as another side note / rabbit hole to those new to the art of 'fucking with brains through audio' — masking is also frequently put to use in mixing, particularly EDM and more percussive genres.

To help squeeze every last drop of perceived dynamic range out of a track you can use a technique known as sidechain compression (http://samplesfrommars.com/blogs/tips-tricks/18999227-how-to...) to dynamically compress other components of the mix in response to (usually) a kick drum.

Is this what causes the volume to dip a bit after strong bass/drum hits?

Yes, exactly that; the compressor turns down the rest of the mix as it is triggered by the kick drum (which isn't going through the compressor). Easily heard at the beginning of "Call on me" by Eric Prydz [1] if you concentrate on the synth string sound. Lots of people think this is the only use of it (it's not), but it's probably the most popular.

[1] - https://www.youtube.com/watch?v=L_fCqg92qks

Thanks for that informative link.. I would say the advice to simply re-write a musical composition so that bass and kick don't happen at the same time seems to put the cart of mix aesthetics before the horse of musical intent! :)

"Sideband compression" sounds like an interesting dsp project! But I think the term is "sidechain compression", right?

Yep, nice catch. Edited.

Re-arranging is definitely one 'solution' and the only real approach when you're talking music played live on acoustic instruments. It's one of the elements of good orchestral music I find most fascinating - that many sources of noise, each playing their part and either perfectly blending or completely juxtaposing one another.

I'd argue that using all the various forms of DSP at your disposal when working with sound in the recorded / amplified sense (and especially electronic music) is an equally valid aesthetic choice. Different limitations, different capabilities and difference aesthetic possibilities for different mediums.

In electronic music, mix aesthetics is quite often exactly the musical intent - you're trying to create a track that sounds a certain way first, and think about melodies and harmonies (if you even have them) second. You would often pick specific key and bassline because of how low frequencies (55-110 Hz octave and below) sound in the mix on your intended equipment. You would even distinguish between open-air and club tracks sometimes because of how open space affects sonic qualities of your mix.

Sidechain compression doesn't do anything more for overall dynamic range than regular compression. Rather it provides a channel for some audio to completely bypass the compressor while the rest gives way to it. The point of it really is purely stylistic, the fact that it reduces the overall dynamic range is incidental.

Keyword there was perceived dynamic range. Completely agree that it's a stylistic choice, the goal of which is usually a 'huge kick/bass' etc. It's simply a technique to create that illusion without needing to increase actual dynamic range.

A great example of this technique being used stylistically is Titanium by David Guetta (ft. Sia) - listen for the kick drum masking out all of the other instruments in the mix at the end of the chorus. In this case, it's used more as an obvious effect than a transparent mixing technique (as compressors usually are).

Doesn't all lossy compression use models of human sensory processing to determine which information is important?

Yup! This is called temporal masking [1].

There's also simultaneous or frequency masking. If you've ever wondered why loud noises prevent you from hearing something that would otherwise be perfectly audible (e.g. a loud motorcycle driving by your apartment can temporarily drown out a noisy air conditioner), that's why.

Both of these psychoacoustic concepts are used in perceptual audio encoders like MP3 - the idea being that certain bits can be discarded because the listener wouldn't be able to hear them anyway.

https://en.m.wikipedia.org/wiki/Auditory_masking

Interesting! Can you completely discard it (as in, complete silence immediately after) or just compress the immediate bits after a lot more than normal?

I'm guessing that the data couldn't completely be silenced because then any heavy dynamic compression at playback would bring out the optimization and break the sound.

Sadly I don't know the answer. I was just the GUI engineer fascinated by the cleverness of how to compress audio.

This is why I only listen to music in PCM 24-bit WAV's instead of compressed audio. The difference is huge due to the missing frequencies in MP3 and similar formats.

Reply on news.ycombinator.com