The state of lossy audio in 2022.

[Aerith Gainsborough]

Lossy audio is one of those sobering, even humiliating things, audiophiles are bound to face

Not really.
Classic audiophiles are masters at self deception.
If they can convince themselves to be able to hear the difference between frikken cables, convincing themselves to easily spot even 320KBps MP3 from a mile away while a jet engine is running next to their ear is easy. :'D

Contrary to cables, the difference between lossy and lossless can be heard and ABX/d, given the right samples (something the encoder struggles with) and listening conditions.

 

[Palladium]

You can pretty much beat any lossy codecs at low bitrates with killer samples, but I view that as a testament of how shockingly good the codecs are at everything else and how terrible our hearing is.

 

[FrantzM]

Not really.
Classic audiophiles are masters at self deception.
If they can convince themselves to be able to hear the difference between frikken cables, convincing themselves to easily spot even 320KBps MP3 from a mile away while a jet engine is running next to their ear is easy. :'D

Contrary to cables, the difference between lossy and lossless can be heard and ABX/d, given the right samples (something the encoder struggles with) and listening conditions.

This requires training, specific (ideal, rarely encountered , even in audiophiles settings) , conditions, specific samples. For the vast majority of cases and people, these codecs are audibly transparent.

Peace.

 

[Sgt. Ear Ache]

This requires training, specific (ideal, rarely encountered , even in audiophiles settings) , conditions, specific samples. For the vast majority of cases and people . These cases are audibly transparent.

Peace.

...and, the people who make statements about how awful, unlistenable and just down right dirty lossy audio is are in 99.999% of the cases absolutely not listening and comparing under anything even remotely close to those very specific circumstances. For general listening (and by that I mean ANY listening aside from actively trying to identify any difference between lossy and lossless under extremely controlled circumstances) a 320kb mp3 is indistinguishable from lossless. I mean asuming we are talking about an mp3 that is compressed from a quality original. An mp3 of an original track recorded from vinyl or whatever is a whole different story if it's being compared to a digital lossless original.

 

[Galliardist]

...and, the people who make statements about how awful, unlistenable and just down right dirty lossy audio is are in 99.999% of the cases absolutely not listening and comparing under anything even remotely close to those very specific circumstances. For general listening (and by that I mean ANY listening aside from actively trying to identify any difference between lossy and lossless under extremely controlled circumstances) a 320kb mp3 is indistinguishable from lossless. I mean asuming we are talking about an mp3 that is compressed from a quality original. An mp3 of an original track recorded from vinyl or whatever is a whole different story if it's being compared to a digital lossless original.

I use lossless, but it's more as a security blanket than anything else. I have been able to pick 320kb mp3 from lossless with close comparison in the right circumstances. Lower bitrates are easier. In normal use for listening to music, there's no reason to bother.

Because I listen a lot to classical guitar and lute music, I do know that I sometimes find it harder to pick up the acoustic/environmental information on some lossy recordings if I'm listening really intently for that. However, outside of that condition I'm happy to listen to lower bitrates. The changes are rarely if ever "awful, unlistenable".

The few bad cases I've had with lossy compression have come with DAB+. In Sydney, the "+" stands for... added compression woes. 64bit AAC "with spectral replication" is the culprit here. As I understand it, the spectral replication element has to be matched to the music being compressed, and radio stations don't do that, so we hear it at its worst.

I attended a UK hifi show where DAB was shown off by the BBC, back in the late '90s. At 256kB, it was way better than their FM broadcast at the time. Digital radio was hobbled by putting too many stations on the same multiplex. Fortunately we will get internet radio generally at decent enough rates over time where it isn't already, and in most environments we'll be able to turn off DAB/DAB+.

 

[danadam]

320 kbps , mp3 removes at least 75% (!!!) of what was in the file containing the lossless signal. "removes" as in discards...

It's 75% when compared to uncompressed PCM data, but it's usually less than 75% when compared to the same data compressed with flac.

 

[FrantzM]

It's 75% when compared to uncompressed PCM data, but it's usually less than 75% when compared to the same data compressed with flac.

FLAC doesn't remove anything from the original signal. When uncompressing FLAC files, the original signal is completely recovered , no loss. OTOH, mp3, removes, part of the original signal, the decoded signa l from an mp3 file has lost part of the signal , in a non-recoverable fashion.
Not the same.


Peace.

 

[Sgt. Ear Ache]

I use lossless, but it's more as a security blanket than anything else. I have been able to pick 320kb mp3 from lossless with close comparison in the right circumstances. Lower bitrates are easier. In normal use for listening to music, there's no reason to bother.

What are the right circumstances?

And anytime I hear a statement like "harder to pick up acoustic/environmental information on some lossy recordings if I'm listening really intently for that" I'm thinking there's no way you're experiencing that unless you actually know you're listening to lossy recordings (if we aren't talking about 192kb or less that is). AKA confirmation bias.

 

[Ricardus]

Not really.
Classic audiophiles are masters at self deception.
If they can convince themselves to be able to hear the difference between frikken cables, convincing themselves to easily spot even 320KBps MP3 from a mile away while a jet engine is running next to their ear is easy. :'D

My favorite line is "Even my wife can hear it.." which isn't sexist at all.

 

[danadam]

FLAC doesn't remove anything from the original signal.

Yes, that's my point. We take uncompressed PCM data at 1411 kbps and:

• compress it losslessly with flac to let's say 960 kbps,
• compress it lossy with mp3 to 320 kbps.

Flac obviously didn't discard anything, but did mp3 discard 1411 - 320 = 1091 kbps (77%) or only 960 - 320 = 640 kbps (66%)?

If we really discarded 77% of uncompressed data, we would end up with 320 kbps of uncompressed data. This could be further compressed losslessly and we would get less than 320 kbps. That's why, imo, it's wrong to say that 320 kbps mp3 discards at least 75% of data. Maybe in some worst-case scenario, but usually it will be less than that.

 

[danadam]

If we really discarded 77% of uncompressed data, we would end up with 320 kbps of uncompressed data. This could be further compressed losslessly and we would get less than 320 kbps.

On second thought, let me amend this because if that 77% included all redundancy in the data, then we would not be able to compress remaining data any more.

But that's the thing, flac only removes redundancy in data, and we don't call it discarding. So when mp3 "shaves off" 77%, some of it is redundancy and likewise should not be counted as discarding.

 

[charleski]

I decided long ago to use high-bitrate AAC for my library so I could fit most of it on an SD card in my phone. Whenever I feel any anxiety about using AAC instead of FLAC I go hunting for some 'killer samples' where people claim they can easily hear a difference with compression (and sometimes prove that with an ABX log). Then I spend some time doing an ABX myself.
Net result: I'm still using AAC and don't have any anxiety.

 

[bkdc]

Yep. I decided years ago on lossy for my library too.
I decided on variable bit rate MP3. My setting is APS or V2 on LAME encoder (well known to be the best MP3 encoder in the world). It can dip as low as 128kbps at times when it needs or go as high as it needs to 256kbps or to 320kbps if there is a lot of data. It is audibly transparent and results in a file about half the size of constant bit rate 320kbps. It's interesting to see LAME encoding at variable bit rate as you can see the different frequency regions being encoded at different bit rates.

LAME MP3 Encoder

Well mastered records still give me eargasms and if I hear anything bad, it's from the original mastering.

And for archival purposes, I have the original CD. Because well.... 90% of my rips are of classical music. And I have so many thousands of albums in my library that I have no interest in ripping re-releases of 1960's recordings from Deutsche Grammophon and Decca but of modern glorious sounding recordings (post 2000) recorded by conscientious technicians and engineers using modern technology, good recording venues, good mic placement, and mastering at 24 bit or better from labels that actually still record in studios or have very good control of the recording location (BIS, Pentatone, Channel Classics, Aparte, Mirare, Oehms, Naive etc) instead of cost-cutting and selling a crappy sounding live recording with audience hacking in the background.

 

[Mulder]

When it comes to LAME, AAC or OGG Vorbis. Or any other codec. Is there any consensus on which lossy codec is best, or is the consensus that they are all equal?

 

[danadam]

When it comes to LAME, AAC or OGG Vorbis. Or any other codec. Is there any consensus on which lossy codec is best, or is the consensus that they are all equal?

From this 2014 multiformat test (96 kbps, 128 for mp3):

The Opus encoder 1.1, the AAC encoder of iTunes, and the aoTuV at 96 kbps and the LAME at 128 kbps were tested. The most common response was on or between "Imperceptible" and "Perceptible, but not annoying".

The Opus codec was the clear winner, and the AAC encoded by Apple iTunes won the second place. The Ogg Vorbis was the third. The MP3, which was allowed to use 30kbps more bitrate than those contenders, was tied to Ogg Vorbis in joint third place.

 

[tomtoo]

If there is a big difference? Maybe but i cant hear it in everyday listening. On the other hand with cheap storage and high data bandwide why use it?

 

[bkdc]

Pick a codec and stick with it. I chose the most popular codec at the time I made my choice over two decades ago. MP3. One which would be recognized by any digital music player and program. Now the issue is moot. You can pick any. I don't know of any modern software/hardware combination that would have issues with any major codec.

If you have 20000 albums that you're transporting on a small drive, there is a practical storage difference between uncompressed at 1400kbps and compressed at 180kbps.

But those old tests are useless. Even my flawed ears can hear the limitations of 128kbps constant bit rate MP3 from the original unless they are crappy 1970's or 1980's recordings.

You want to pick something that has been deemed transparent and that you are confident is transparent. I'm anal when it comes to audio, and I am confident that what I'm using is transparent even to pristine 12 year old ears.

LAME - Hydrogenaudio Knowledgebase

wiki.hydrogenaud.io

 

[fieldcar]

I figured I would share some charts. My feeling is that the best we can do to truly objectively observe a lossy codec is through a spectrogram. Group ABX testing is fine and all, but it's far from objective. From what I gather, lossy audio is best when it keeps anything under ~15KHz untouched compared to the source material. I've found that classical music to be the hardest for codecs to manage, and most EDM/pop is easy for lossy codecs.

My criteria for a good sounding lossy codec is as follows:

• Good visual similarity between lossless source and lossy codec, especially below 15KHz. This means, same amplitude and intensity for the meat of the music. Nothing added or lost.
• No new transients and harmonies that were not existent on the source material.
• No masking of "air" in the 6-13KHz range.
• Aggressive omission of non-transient information (noise) only above 15KHz. Most content above here is only harmonic transients and noise anyway.
• If lossy omission removes anything, it should be during louder parts of a track. When the track has less masking events, it should not cut high frequency information.

I really thought AAC would be considerably better than MP3, but follow along with my interesting findings.

Opus 320Kbit codec v1.3.1
Outstanding performance. Almost zero change below 20KHz. No modification of harmonies, and it leaves noise between those harmonies alone. The harmonies don't have any exaggerations or attenuation. I can almost guarantee that you will not be able to tell the difference between this and a FLAC. No spectral holes are visible. It's probably due to how the codec re-synthesizes some of what it removes. Just Magic.
Note: I ran 320Kbit both CBR and VBR, but they appear identical. I think I recall reading something about how this codec is always VBR no matter what.

Apple CBR AAC 320Kbit Apple iTunes encoder
Applies some sort of gradual NR from 6KHz and up, and then has some sort of noise/smearing between harmonics and some of the harmonies are amplified. Sure, it preserves higher frequency transients, but I'd argue that they are a waste of information when the most perceptible part of the audio is below 15KHz.
EDIT: Something was wrong with the Apple encoder. It appears to be much closer to OPUS now.

MP3 CBR 320Kbit Lame encoder 3.101
Looks rough at first glance, but most of the lossy omission is above 16KHz. This may actually be better than Apple AAC, despite it looking kind of rough/grainy from 16-20KHz. Almost nothing below 16KHz is touched, which means it's unlikely to be perceptible. I'm actually surprised by this, as I though AAC would blow MP3 away. I'm sure a lot of this is due to LAME's great encoder.

MP3 VBR V0 lame Lame encoder 3.101
Applies aggressive de-noising above 16KHz, but also appears to infrequently remove noise between loud harmonies from 6-16KHz. Really good results considering it's just an MP3. The other cool thing is that it restores high frequency noise up to 22KHz during silent parts of a track that wouldn't be masked by louder parts of the track. I may actually start encoding my portable libraries for my car in this format since my car doesn't support OPUS. I'm extremely surprised by this result.

So, what are your thoughts? Any requests for lower bitrates?

 

[bevok]

I have an LDAC BT transmitter. Excellent sound, no dropouts.

What type?

 

[Aerith Gainsborough]

I decided long ago to use high-bitrate AAC for my library so I could fit most of it on an SD card in my phone. Whenever I feel any anxiety about using AAC instead of FLAC I go hunting for some 'killer samples' where people claim they can easily hear a difference with compression (and sometimes prove that with an ABX log). Then I spend some time doing an ABX myself.
Net result: I'm still using AAC and don't have any anxiety.

On a phone, implies you use to listen to music in a noisy environment.
Yeah, fat chance of discerning lossless from lossy in that case.

I use a blutooth Transceiver for my old ATH M-50s. No issues here either, despite the protocol being lossy.

If you have 20000 albums that you're transporting on a small drive, there is a practical storage difference between uncompressed at 1400kbps and compressed at 180kbps.

Sheesh. 20K albums, really? Seems kinda excessive but okay.
At ~60 minute length per album that would clock in to around 12TB of storage space (1400kbps) or 1.5TB (180kbps).