Transparency of LC-AAC at 500kb/s bitrate (stereo file)

I'm recently exploring audio formats, and I found that when encoding to aac at 500kb/s, it's much smaller than lossless formats like flac or alac (which is expected), but audiochecker said that it's CDDA with probability of 94%.

I'm using FFmpeg version N-94001-g1e7a8b92ee with only internal aac support on macOS High Sierra 10.13.6 and using audiochecker 2.0 beta in crossover.

Does this mean that aac is extremely near lossless at 500kb/s?

Solution 1:

There's a lot of talk about 'lossless audio'. There's also a lot of opinion surrounding it, much of which doesn't come from sound engineers but from people who are chasing the wrong part of the rainbow ;)

To really simplify,

• Lossless audio is that which can be uncompressed to produce exactly the same data that was originally encoded.
  You need this if you are going to be further processing the audio.
• Lossy audio can be uncompressed to sound almost exactly as it was originally encoded.
  You need this if you just want to listen to the audio.

I'm not going to dig into any math in this, just skim lightly over the basics.

The difference in the two has become a matter of public perception, rather than hard fact & analytical data.
Very few people on an A/B test can tell the difference, so they use tools which they don't understand to give them data that makes no real sense…
[bear with me, I'm not being insulting, but some of the interweb wisdom on this is way off the mark].

You will often see pictures of spectrograms like this

Telling you the one on the right is obviously lossy because it has a lower frequency cut-off. Well, yes, it does imply that, but it doesn't actually prove it, it is a false premise based on a misunderstanding. What it does prove is that it used a lower sampling frequency, causing a lower Nyquist limit & therefore a lower high frequency shelving limit in the resulting audio.
If you re-compressed the file on the right into a lossless format it would not magically regain the lost data & yet the file itself would now be completely lossless - but only insomuch as no further data would be lost. I think this is probably what the 'checker' apps are trying to test, but in quite a misleading way.

Good lossy compression doesn't look 'truncated' like that on a spectrogram. It doesn't look any different to a lossless file - because the differences are far, far too small to show up on such a blunt tool as a spectrogram.
The whole point of lossy encoding is that it quite cleverly throws away the bits you can't hear anyway. That's its trick. The higher the bitrate, the less it has to throw out, which is why there is a noticeable difference between a 256kbps [or 320kbps] & a 128kbps file, if your speakers & ears are good enough. If you're going to be listening on earbuds from your phone, or in your car, then 9/10 people can't tell the difference between 128kbps & anything higher - the background noise will effectively bury the difference.

You can use a tool that will run the math on an uncompressed lossy file, but how it can decide if any data is missing from the end result without an original to compare to is completely beyond me. Maybe it looks at only the high frequencies, maybe the actual cut off…

I'm going to pick on one of these received wisdom audio checker sites & pick some holes in it… the poor guy probably doesn't deserve being singled out, but he covers the same tired, misinformed ground that many of them do…

From: Audio Checker – Check the True Quality of Your Audio Files

In reference to the above spectrograph…

kHz, a measurement of frequency that greatly affects sound quality, will be displayed on the left. In theory, bitrates work with frequency and higher bitrates preserve higher frequencies, which result in better sound quality.

✔️correct, though things get less confusing if you use the term 'sampling frequency' rather than bit-rate. Bit-rate is really a transmission 'speed' not an accurate gauge of the actual sampling frequency. Only lossy compression formats really have anything to do with bitrates. Completely lossless, uncompress PCM such as WAV of AIF have only a sampling frequency [& bit-depth].
See Presonus: Digital Audio Basics: Sample Rate and Bit Depth for some further info.

Lossless FLAC files don’t lose any of their higher frequencies as the left chart shows, yet lossy audio files don’t have higher frequencies

❌ false premise. Lossy files can have just as much high frequency information as lossless - it's based on sampling frequency & Nyquist, as explained above.

detects upscaling, upsampling and transcoding.

✔️ Ahh… so that's how it does it. It checks to see if someone has cheated by taking a low-grade file & changed the transcoding parameters to make it seem better. All those methods will produce data areas with either 'gaps' or doubled data, which certainly prove the file has been messed with at some point, probably in an attempt to deceive.

❌ What it completely & utterly fails to do is test that a file can be uncompressed to give exactly the same data as when it left the 'factory'. It has nothing to compare to.

OK, let's move away from there & back to your results…

CDDA with probability of 94%.

Well, that's fine if you consider CDDA [16-bit 44.1kHz PCM] to be the pinnacle of audio storage. It hasn't been in probably a decade or so. [Also bear in mind your checker came to that conclusion with absolutely no reference to the original file at all, it's just a guess].

Back in the days everybody had CD players, from the mid-80s onwards, then CDDA was the pinnacle of consumer audio storage - once we'd got over the argument about vinyl, which I'm going to ignore entirely for this exercise.

Since Apple moved to DRM-free AAC in the mid 2000's & introduced iTunes Plus, record companies started to upload their original audio masters to iTunes at 24-bit rather than 16-bit. This coincides with computers getting fast enough that recording studios could start to use higher frequencies in the original recordings - these days going up as far as 24 or even 32-bit 192kHz before downsampling to deliver to the consumer. In a pure numbers game, that's more numbers & it's 'real' because it hasn't been through any process since it left the recording studio/mastering suite. That's the most accurate source available to anyone outside the studio.

By the time this has been encoded to AAC at 256 or even 320kbps, the result is actually better than CDDA…
…& that's where the argument for lossless finally falls down.

Unless the record companies release lossless data at those sample/bit rates, [which did happen with the recent Beatles remasters, only available on USB stick] the 'best' source to the consumer is the AAC.. which was already lossy, but pretty much invisibly lossy.