The state of lossy audio in 2022.

[Jack Harrison]

I pay for Spotify and Amazon HD. I listen to Spotify on everything except my main system and then I can’t tell the difference between it and any other source. I much prefer its interface to Amazon’s and often enough use Spotify to choose something to listen to on Amazon HD whose search system is bizarre. But on my main system Spotify just doesn’t sound as good as Amazon HD or my losslessly ripped CDs or the CDs themselves. if this really is the case then I’d say it was because the system is more revealing and also I listen to it more intentionally. I definitely don’t have golden ears and if there was some way of getting Spotify’s ‘high quality’ 320kb to sound as good as red book I would do it happily and save my money. I’d maybe spend it on a DAC with a totally unhearable 120+ sinad.

 

[Mulder]

I pay for Spotify and Amazon HD. I listen to Spotify on everything except my main system and then I can’t tell the difference between it and any other source. I much prefer its interface to Amazon’s and often enough use Spotify to choose something to listen to on Amazon HD whose search system is bizarre. But on my main system Spotify just doesn’t sound as good as Amazon HD or my losslessly ripped CDs or the CDs themselves. if this really is the case then I’d say it was because the system is more revealing and also I listen to it more intentionally. I definitely don’t have golden ears and if there was some way of getting Spotify’s ‘high quality’ 320kb to sound as good as red book I would do it happily and save my money. I’d maybe spend it on a DAC with a totally unhearable 120+ sinad.

I have a friend who had nearly exact the same experience as you, i.e. Spotify is good but not as good as CD on his high-end system. Until I performed a ABX test with him (with foobar2000), on his own high-end system, and he could not distinguish between OGG-Vorbis, AAC and Red Book. You could try the same if you are curios. Compare OGG (Spotify is OGG) and Red Book in an ABX-test. It is easy to do. Just remember to make your own OGG from the same CD so you eleminate the risk of comparing different masters.

 

[Koeitje]

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.


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).

12TB is a single hard drive these days.

 

[Deleted member 58722]

What type?

BTA30 Pro-FiiO official website

FiiO HiFi Bluetooth Transceiver BTA30 Pro is a Bluetooth decoder with three modes of Bluetooth reception, Bluetooth transmission and digital decoding. It is equipped with Qualcomm CSR8675 Bluetooth chip, independent high-performance DSP chip and high-performance DAC AK4490. It supports...

www.fiio.com

 

[Jack Harrison]

I have a friend who had nearly exact the same experience as you, i.e. Spotify is good but not as good as CD on his high-end system. Until I performed a ABX test with him (with foobar2000), on his own high-end system, and he could not distinguish between OGG-Vorbis, AAC and Red Book. You could try the same if you are curios. Compare OGG (Spotify is OGG) and Red Book in an ABX-test. It is easy to do. Just remember to make your own OGG from the same CD so you eleminate the risk of comparing different masters.

Honestly because all my CDs are ripped to alac and I’ve so much spare storage the only reason I have for wanting to listen to 320kb is so I can stop my Amazon HD subscription and just listen to Spotify. And I think you may well be right and it might be the mastering that is making Spotify sound so different on my main system. When I’m listening casually I actually often enough like or even prefer the Spotify sound especially with rock or small groups of musicians as in jazz or chamber music. But perhaps with big orchestral or choral works I think their default normalisation maybe stuffs up reducing the dynamic range too much? But I also wonder whether they somehow sweeten the sound by adjusting the highs also? Because it’s not just dynamic range that sounds different to me but also tonality. I think with Amazon HD you can turn off normalisation which I’ve done, and they don’t normalise so aggressively either?

 

[charleski]

On a phone, implies you use to listen to music in a noisy environment.

Like ... hotel rooms? No. Frankly, for several years I was using my phone as the source for my home listening as well.
You know what they say about assumptions ...

 

[Mulder]

Honestly because all my CDs are ripped to alac and I’ve so much spare storage the only reason I have for wanting to listen to 320kb is so I can stop my Amazon HD subscription and just listen to Spotify. And I think you may well be right and it might be the mastering that is making Spotify sound so different on my main system. When I’m listening casually I actually often enough like or even prefer the Spotify sound especially with rock or small groups of musicians as in jazz or chamber music. But perhaps with big orchestral or choral works I think their default normalisation maybe stuffs up reducing the dynamic range too much? But I also wonder whether they somehow sweeten the sound by adjusting the highs also? Because it’s not just dynamic range that sounds different to me but also tonality. I think with Amazon HD you can turn off normalisation which I’ve done, and they don’t normalise so aggressively either?

I'm not advocating Spotify specifically here, but normalising can be turned off in Spotify as well. The question that makes me most curious here is whether MP3, OGG, etc. is transparent, ie to what extent it is possible to hear sound differences between lossless and lossy. Then how a specific streaming service sounds in a comparison can be hard to know I think if you don't know exactly how they manipulate the sound.

 

[Koeitje]

I'd be more worried about the master used and possible processing by the content provider than lossy versus lossless.

 

[kipman725]

I acidently trained myself to hear mp3s fixed block size (limits rate of change of frequency spectrum) when doing an undergraduate coursework using the same technique to perform noise cancellation, its especialy obvious on cymbals.

 

[Jack Harrison]

I'm not advocating Spotify specifically here, but normalising can be turned off in Spotify as well. The question that makes me most curious here is whether MP3, OGG, etc. is transparent, ie to what extent it is possible to hear sound differences between lossless and lossy. Then how a specific streaming service sounds in a comparison can be hard to know I think if you don't know exactly how they manipulate the sound.

My understanding which might be wrong is that all Spotify tracks are normalised in the masters used by Spotify and that although some control is given back to the user in the app all the tracks remain normalised to Spotify specs. Spotify certainly gives advice to artists on how to normalise their masters prior to submission in order it seems to stop Spotify having to do it themselves. And one of the reasons Spotify gives is that if the dynamic range of the masters is too great then it leads to distortion in the loud passages. That seems to suggest they recognise a real difference between the potential quality of 320kb lossy vs red book lossless?

 

[FrantzM]

While I enjoy my lossy , every day, I find that applause sound wrong on Spotify. I haven't made a direct comparison with CD. It could be with my speakers or room... Don't bother me.

Peace

 

[fieldcar]

While I enjoy my lossy , every day, I find that applause sound wrong on Spotify. I haven't made a direct comparison with CD. It could be with my speakers or room... Don't bother me.

Peace

Can you share the specific track on spotify? I wonder if I can capture the lossy audio and look at it vs qobuz or Amazon hd. Pm me if you don't want to share publicly.

I'm pretty sure I've noticed similar with these services. I really doubt that any sort of care or up to date encoders are used on these services. Bad rips, bitrate(QoS) throttling, and old encoders could be normal for the sake of maximum profit. With how Spotify gave up on lossless, I'm sure they realize 99.99% of their customers don't care at all. Oh. Also, back in the day, Spotify was founded on illegal rips from music sharing sites. I wonder if I can find that article.

Edit: https://gizmodo.com/early-spotify-w...uilt-on-pirated-mp3-files-new-book-1795109991

 

[Jack Harrison]

Loudness normalization - Spotify

artists.spotify.com

The first section of this page seems to describe the normalisation applied routinely by Spotify that can be turned off by the user and/or isn’t used by Spotify in certain applications. But the bit at the end ‘mastering tips’ suggests that all their masters are normalised and the reason seems to be because there is inherent distortion in 320kb Ogg Vorbis if full red book dynamic range is used?

 

[danadam]

My understanding which might be wrong is that all Spotify tracks are normalised in the masters used by Spotify and that although some control is given back to the user in the app all the tracks remain normalised to Spotify specs.

They only calculate the loudness, not modify the masters. In Loudness normalization:

Positive or negative gain compensation gets applied to a track while it’s playing.

-> while it’s playing

And one of the reasons Spotify gives is that if the dynamic range of the masters is too great then it leads to distortion in the loud passages.

But the bit at the end ‘mastering tips’ suggests that all their masters are normalised and the reason seems to be because there is inherent distortion in 320kb Ogg Vorbis if full red book dynamic range is used?

I don't see how it suggests anything like that.

Firstly, they say "If your master’s louder than -14dB integrated LUFS". "Louder than" usually means more compressed, i.e. less dynamic range, not more.

Second, the reason why they suggest leaving headroom is that lossy compression may produce peaks higher than in the original file. Without headroom such peaks will get clipped. And the reason why they suggest more headroom when the track is louder, is because louder, more dynamically compressed tracks are more susceptible to that.

 

[danadam]

Without headroom such peaks will get clipped.

And actually, normalization can help in this case. If normalization applies negative gain during playback (which is almost certain for loud tracks), then such higher peaks will get reduced and will not clip.

 

[melvinjames]

Interesting discussion gents. Years ago I was convinced I could hear a difference between flac and WAV and consequently ripped my CDs to WAV using dBpoweramp. Believed it for a couple of years in fact. In the end it was just my own bias, proven with my own non-scientific a/b listening tests. Similarly, I could easily distinguish between those same flac/WAV files and MP3 320 CBR. Yeah, not so much, proved again with the same basic non-scientific a/b testing. If I were put to a proper double blind test I can’t imagine scoring better than 50/50 and we all know what that means.

That said, I’ve been a happy Qobuz subscriber since their USA launch and canceled my Spotify Premium account. The pawltry $10/year price difference keeps me with Qobuz, which appeals to my irrational audiophile tendencies.

 

[tmtomh]

I appreciate the spectrograms comparing lossy formats - I've looked at a lot of those myself. And I also appreciate the (as usual) very detailed and in-depth discussion here.

That said, to me the issue is fun to discuss in all its complexities, but as a practical matter it's very simple: high-bitrate lossy is very difficult to distinguish from lossless, and for most people in most listening situations (including your main, super-revealing home hi-fi setup), with most music, it's totally impossible to distinguish from lossless.

The 320k bitrate is high enough that it appears to preserve almost all of the full theoretically audible spectrum of frequencies up to 20kHz, and in 99%+ of cases preserves 100% of the full spectrum of frequencies that anyone can actually detect based on their hearing ability, the musical content, the listening environment, and perceptual masking.

So to me it all comes down to a simple three-way question of striking a practical balance among:

1. Format preference - lossless vs lossy
2. Collection size/available storage space
3. Interface/source preference - local streamer, a particular subscription service, and so on

If you have a not-gigantic collection, or a ton of storage space, or your preferred subscription service with the best selection and interface offers high-bitrate lossy or lossless at a price you are willing to pay, then you're good to go and don't need to make any tradeoffs. Otherwise, the tradeoffs are simple to understand and generally minor in their impact on our ability to enjoy hi-fi sound reproduction of our favorite music.

 

[Ambient384]

Study comparing lower bitrate Opus, AAC, Ogg Vorbis and MP3 to a lossless source track:
Results of the public multiformat listening test (July 2014)

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.
View attachment 269678

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.
View attachment 269679

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.
View attachment 269680

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.

View attachment 269681

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

Noticed that with Classical & Dark ambient that Apple AAC seems to use a more of NR fliter than a smart 16 ~ 22KHz low pass fliter, At <192kbps It seems to get easily confused as synth sounds/noises end up smeared badly. LAME seems to just use a dynamic 16KHz low pass that checks If It can keep or cut the 16 ~ 22KHz area while lower settings just use a 16KHz low pass filter.

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?

LAME MP3 at V2 Is pretty much a safe a choice with universal suppourt as well.

 

[ehabheikal]

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

Me too, i can even get huge distances and the sq stays great.

 

[Tks]

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?

OPUS actually seems to be the poised successor to MP3 (and AAC if Apple would just stop being Apple for two seconds). I saw one guy on Reddit though challenging this notion with Monkey's Audio Encoder. I remember being convinced in a serious aspect I can't recall though (might be filesize with respect to preserving most of the spectrum). But I personally lean on Opus since it's bandwidth savings are huge given how good the results are. It's deployed on platforms like Twitch and Discord for good reason, unrivaled in voice communications I believe, so not just music frequencies. But yeah, almost all these codecs are good enough to use and never worry about lossless differences during listening (when there are audible differences, you have to be listening for it, and be trained, and listening to specific tracks). It all comes down to compatibility and bandwidth savings and I suppose conversion speeds at the end of the day. I think OPUS achieves this better than any other.

If compatibility is a concern (since not all decades old devices support Opus decoding), MP3 highest quality VBR is good enough (actually better than 320kbps MP3 I recall reading due to allowing whatever bitrate the file calls for during playback).