SoylentNews is people
SoylentNews
Vice Noisey reports on a musician who isolates MP3 artefacts by finding the differences between an MP3 and a lossless recording, then samples them to create his own music (N.B. the examples are hosted on Soundcloud; Javascript is needed to listen to them).
These days though, in our rush to listen to all music everywhere at all times, we often sacrifice these layers by listening to the most readily available streams or downloads, which are usually relatively crappy formats like MP3, AAC, or whatever the hell Grooveshark uses, which can sometimes sound like the recording of a song being through a coke can in a garden shed.
Often, we're losing out on a significant amount of what the artist intended, because when the original analog music is converted to one of these formats, certain layers of sound are lost in the digital compression. Translation: there's a lots of bits to your favourite albums that you may have never even heard.
Exploring this, is the Ghost in the MP3 project by doctoral music student Ryan Maguire from the University of Virginia's Center for Computer Music. He investigates these lost layers of sound, what they sound like when rescued, and then tries to make new music with them. For an example in his study, he took the layers of sound lost to compression from the acapella song "Tom's Diner" by Suzanne Vega, which was also the template song used by Karlheinz Brandenburg, the pioneer of the MP3, to test whether the compression of MP3s worked. You can hear the track he made from those bits below.
(Score: 2) by RS3 on Monday May 15 2017, @05:01PM (3 children)
Most of the problems and artifacts are caused by aliasing. https://en.wikipedia.org/wiki/Aliasing/ [wikipedia.org] The article contains 2 links to soundcloud-hosted sound files- first one is clean noise, and the second is the resulting aliasing you hear because of digital "downsampling" that is part of the .mp3 conversion.
I'm somewhat involved in the pro audio / recording world and I know of nobody who uses proper anti-aliasing technology when doing A/D. It needs to be done before the A/D conversion and would be a "brick-wall" filter. Analog brick-wall filters are difficult and expensive to design and build. If not done correctly, they will add artifacts (distortion) of their own.
A great way to get around the problem is "oversampling" https://en.wikipedia.org/wiki/Oversampling/ [wikipedia.org]. Fortunately most A/D conversion (recording sample rate) is being done at 88.2 or 96 KHz so we're OK now and going forward, ...
but, aliasing happens in the digital domain too, so you still need to do a digital brick-wall filter before "downsampling" to 44.1K, or whatever your final bit-rate is. DAW https://en.wikipedia.org/wiki/Digital_audio_workstation/ [wikipedia.org] software may be smart enough to do this for you, I don't know, nor am I aware of any plugins or filters to do digital anti-aliasing; IE: digital brick-wall filter.
I contend that anything which has been sampled at 44.1K starts having aliasing problems. You might not be able to hear above 16 or 20 Khz, and even if your mic or source can't do it, there's electrical noise which can certainly contain energy at frequencies above Nyquist https://en.wikipedia.org/wiki/Nyquist_frequency/ [wikipedia.org].
(Score: 0) by Anonymous Coward on Monday May 15 2017, @06:15PM (1 child)
It's not just aliasing.
One of the things that MP3 does is to apply a psychoacoustic algorithm to simplify what it will actually capture and replicate.
Things don't actually get aliased if they never show up in the signal to be captured.
(Score: 0) by Anonymous Coward on Monday May 15 2017, @06:42PM
> It's not just aliasing.
What is the first word of the post you replied to?
> Things don't actually get aliased if they never show up in the signal to be captured.
Welcome to SoylentNews, Captain Obvious!
(Score: 0) by Anonymous Coward on Monday May 15 2017, @08:12PM
What?
Yes, this is why early CD players sounded so bad and why we developed oversampling...
Err, no. The earliest CD players did this by inserting three zero-valued samples between each discrete sample, making the sample rate for the reconstruction filter 176.4kHz. 88.2 and 96kHz are recording rates, they are oversampled at much higher frequencies.
What does this have to do with mp3's though? If you want to talk audio quality and sample rate conversion with digital filters, Shannon/Nyquist specified a sinc (ringing) filter to produce a non-ringing output. Sinc filters are commonly approximated in DSP using FIR and most converters these days are okay, audition being quite good. I often use SRC 'sinc' filter which is quite slow but adequate, I'll then use lame to encode an mp3 after I have done this. Not that I'm required to change from the professional working sample rate of 48kHz often.
PS: 44.1kHz was only ever used for CD's, the original spec was 44.3kHz but Philips changed it because Japanese broadcast processors ran at 44.1 or 44.056 drop frame. It's an obsolete format, there are no devices that will fail to play a 48kHz MP3 and only a handful of specialised reasons to be working at higher sample rates.