FLAC vs. 320 Mp3

[Joe Bloggs]

What do people think about this? he specially mentions how encoders can mangle intersample clipping, baking it permanently into the music no matter how much headroom you give

i tested this with his mentioned song "Red Hot Chilli Peppers - Monarchy of Roses" and listening on MP3 quality on tidal sounds, it sounds even more dreadful then MP3 usually do IMHO

Even lowly old mp3 could be encoded and decoded in floating point. I would be VERY surprised if this is not the case with newer codecs.

Saying that a crappy encoder CAN mangle intersample clipping for the sake of arguing for lossless or even high res files is akin to saying that CD format is bad by saying that there's a crappy DAP out there that mangles oversampling and can only play decent music if it's encoded in hi-res.

 

[danadam]

5:50

because several of the most popular decoders for streaming services and lossy audio
reduce the file to fixed point immediately after decoding
so that means that all this extra peak information that we can hear gets sliced off it gets clipped
before the file gets to the volume control the normalization

Spotify does the right thing and the normalization happens before converting to fixed point:
https://www.audiosciencereview.com/...-its-been-sent-to-spotify.58449/#post-2143888

it simply shows one of the flaws of MP3 IMHO

It shows those artists/producers cared more about loudness than the potential clipping, be it from ISPs or lossy encoding (or all kinds of DSP that could happen in the playback chain).

If you do your own conversions and you care about such things, you can always reduce the volume before conversion yourself.

 

[gregorio]

What do people think about this?

It has a somewhat complicated history but modern standards eliminate it in most circumstances. I’m very surprised that he’s only recently actually tested the audibility of ISPs, I first did that about 20 years ago and then again in far more detail about a dozen years ago. It’s a well known/studied phenomena, ITU BS.1770 standardised the dBTP measurement in 2006 and global standards (EBU R128 and ATSC A/85) eliminated the possibility starting in 2012, although not all distributors necessarily implement these standards.

he specially mentions how encoders can mangle intersample clipping

Huh? He does the exact opposite and demonstrates that AAC encodes and rebuilds the intersample peaks! In fact, he had to use a plugin to introduce hard clipping because the decoded AAC did NOT mangle the intersample peaks. Did you simply miss that part?

G

 

[Ghoostknight]

Huh? He does the exact opposite and demonstrates that AAC encodes and rebuilds the intersample peaks! In fact, he had to use a plugin to introduce hard clipping because the decoded AAC did NOT mangle the intersample peaks. Did you simply miss that part?

watch again 5:40 and going forward

he doesnt mention the streaming file AAC were correct and thats why he introduced a clipping plugin, my guess is it was probably just a measure to get a more accurate null test going from both lossless files

he also mentions right after 5:40 that the problem occours when encoders use fixed point instead of floating point and that many popular encoders still use fixed point

Even lowly old mp3 could be encoded and decoded in floating point. I would be VERY surprised if this is not the case with newer codecs.

Saying that a crappy encoder CAN mangle intersample clipping for the sake of arguing for lossless or even high res files is akin to saying that CD format is bad by saying that there's a crappy DAP out there that mangles oversampling and can only play decent music if it's encoded in hi-res.

well there are dacs out there that suffer from this, but only if they dont have internal headroom and you use 0 to -2.9 dBfs
so its fairly easy avoidable.... in the case of sub-par encoders you can nothing do about it but use another one, and that might not be a option if we talk about youtube/soundcloud or any other streamingservice offering some kind of lossless

it would be interesting to know which streaming services could still use integer/fixed point , it might not be a thing anymore?

tho my subjective expierence specially with this rhcp song is still that something is going on, specially with heavy distorted/compressed music that just sounds less worse with lossless, i thought this ISP problem could be an objective explanation, but it might not be with most streaming services as AAC is fairly common and hopefully in the right way implemented

 

[danadam]

he also mentions right after 5:40 that the problem occours when encoders use fixed point instead of floating point and that many popular encoders still use fixed point

Decoders, not encoders.

 

[gregorio]

watch again 5:40 and going forward
he doesnt mention the streaming file AAC were correct

Yes he does! As I stated, you simply missed out the relevant part, that’s actually around 5:20, where he specifically states “these peaks are cleanly stored and by reducing the gain we can get them back”!

he also mentions right after 5:40 that the problem occours when encoders use fixed point instead of floating point and that many popular encoders still use fixed point

Again, NO he does not, he states that “AAC and other lossy codecs use effectively a crude version of floating point”, nothing at all about any encoders using fixed point! The potential problem actually occurs because some players, as he states at around 6:00, “reduce the file to fixed point immediately after decoding” and he actually demonstrates that the encoder and decoder preserve the ISPs using decoded AAC files in WaveLab.

No wonder we keep having arguments, you watch a video and then your brain apparently misses out relevant parts and invents something completely different from what the video actually states/demonstrates!!

G

 

[FunkyBassMan]

It has a somewhat complicated history but modern standards eliminate it in most circumstances. I’m very surprised that he’s only recently actually tested the audibility of ISPs, I first did that about 20 years ago and then again in far more detail about a dozen years ago. It’s a well known/studied phenomena, ITU BS.1770 standardised the dBTP measurement in 2006 and global standards (EBU R128 and ATSC A/85) eliminated the possibility starting in 2012, although not all distributors necessarily implement these standards.

Huh? He does the exact opposite and demonstrates that AAC encodes and rebuilds the intersample peaks! In fact, he had to use a plugin to introduce hard clipping because the decoded AAC did NOT mangle the intersample peaks. Did you simply miss that part?

G

I'm a bit suprised at how dismissive you seem about this, though i guess I shouldn't be. Sure, standards like ITU-R BS.1770 and EBU R128 introduced true-peak limits and recommendations—but as you know from your own work, realworld practice rarely follows theory perfectly. Ian Shepard's recent test wasn't about finally realizing ISPs exist; it was about showing clearly that even today- with current codecs and playback hardware, ISPs remain audibly problematic.

You mentioned your own tests from years ago, cool, but implying the issue is essentially solved just doesn't reflect reality. Shepard explicitly demonstrated that although AAC encoding does preserve peaks in theory (due to floating-point processing), the practical issue emerges precisely at PLAYBACK—when many decoders convert audio immediately back to fixed-point ---causing exactly the clipping distortion your saying shouldn't exist. That isn't theoretical; it's a real, audible phenomenon experienced by countless listeners every single day.

Yes, standards like ITU BS.1770, EBU R128, and ATSC A/85 exist, but standards alone don’t eliminate problems unless universally followed, and clearly they're not—especially given the ongoing loudness wars and consumer preferences for louder masters. Professionals are still regularly delivering files with ISPs because artists and labels demand it.

So, respectfully, dismissing this as a settled issue feels disconnected from real-world practices. Shepard’s demonstration isn't just theoretical speculation; it's evidence that ISPs are not only relevent but consistently audible. It’s crucial we acknowledge this reality rather than downplaying its significance. Standards alone don't solve the issue—real-world adherence and awareness do.

 

[gregorio]

Sure, standards like ITU-R BS.1770 and EBU R128 introduced true-peak limits and recommendations—but as you know from your own work, realworld practice rarely follows theory perfectly.

As I know from my own work, if EBU R128 or ATSC A85 are stipulated in the deliverables and that isn’t adhered to in practice, then the deliverables will typically just be rejected (and have to be corrected and resubmitted) or in the case of many streaming services, will simply have it’s level reduced to comply with the standard. There are no exceptions (because the ingest process is automated) and therefore the practice DOES follow the theory perfectly, except of course in those services/distribution channels which do not implement a dBTP standard, as I already stated!

Ian Shepard's recent test wasn't about finally realizing ISPs exist; it was about showing clearly that even today- with current codecs and playback hardware, ISPs remain audibly problematic.

Huh? He did NOT show “clearly that even today-with current codecs and playback hardware ISPs remain audibly problematic”. Didn’t you even watch the video or maybe you just missed the fact that after the AAC was decoded he used a hard clipping distortion plugin? What he actually therefore “clearly showed” is that a clipping distortion plugin can cause audible problems! The question remains about whether it would be rendered audible by the DAC but for more than 30 years DACs routinely have had headroom built in.

G

 

[FunkyBassMan]

As I know from my own work, if EBU R128 or ATSC A85 are stipulated in the deliverables and that isn’t adhered to in practice, then the deliverables will typically just be rejected (and have to be corrected and resubmitted) or in the case of many streaming services, will simply have it’s level reduced to comply with the standard. There are no exceptions (because the ingest process is automated) and therefore the practice DOES follow the theory perfectly, except of course in those services/distribution channels which do not implement a dBTP standard, as I already stated!

Huh? He did NOT show “clearly that even today-with current codecs and playback hardware ISPs remain audibly problematic”. Didn’t you even watch the video or maybe you just missed the fact that after the AAC was decoded he used a hard clipping distortion plugin? What he actually therefore “clearly showed” is that a clipping distortion plugin can cause audible problems! The question remains about whether it would be rendered audible by the DAC but for more than 30 years DACs routinely have had headroom built in.

G

Gregorio, come on-- you're seriosly oversimplyfying what's hapening out there. Yes, standards like EBU R128 and ATSC A85 exist, but let's be real: plenty of masters slip thru with ISPs, whether due to deadlines, artist demands, or plain oversite. Acting like automated rejection or perfect adherence is the norm everywhere just isn't acurate. Maybe you've been lucky or disciplined enough to never encouter this, but it doesn't mean it's universal..

About Ian Shepard’s test: Sure, he used a clipping plugin, but you completly missed his point. He wasn't showing some artificial scenario; he was demonstrating exactly what can happen when consumers play audio decoded into fixed-point formats. Pretending every DAC for "30 years" perfectly handles peaks ignores real-world consumer gear that clips regularly, gear people actually listen on everyday.

Honestly, your attitude here isn't helpful. ISPs aren't just theoretical—they're genuinly audible under typical listening conditions, and Shepard's test realisticly proves this. Dismissing it outright helps no one

 

[Elegiac]

MP3 upsampled to 384khz 4 life

 

[gregorio]

Acting like automated rejection or perfect adherence is the norm everywhere just isn't acurate.

Strawman argument! I clearly stated “except of course in those services/distribution channels which do not implement a dBTP standard, as I already stated!” and therefore I am not acting it is the norm everywhere!

he was demonstrating exactly what can happen when consumers play audio decoded into fixed-point formats.

No he wasn’t, he demonstrated what would happen if a consumer played that audio without reducing it’s, through a DAC without headroom.

Pretending every DAC for "30 years" perfectly handles peaks ignores real-world consumer gear that clips regularly, gear people actually listen on everyday.

Strawman argument! I did not pretend every DAC for 30 years perfectly handles ISPs. Why don’t you try not deliberately misquoting/misrepresenting what I state for a change?

ISPs aren't just theoretical—they're genuinly audible under typical listening conditions, and Shepard's test realisticly proves this.

Strawman argument! I never stated or even implied ISPs are only theoretical. Under typical listening conditions most ISPs will not be genuinely audible because most DACs have at least some headroom. And again, Ian’s test did not “realistically” prove this because “realistically” consumers do not deliberately apply a clipping distortion plugin to their system!

Brilliant, you’re going for making yourself look foolish in two threads at the same time, a new low, you must be very happy with yourself! lol

G

 

[FunkyBassMan]

Strawman argument! I clearly stated “except of course in those services/distribution channels which do not implement a dBTP standard, as I already stated!” and therefore I am not acting it is the norm everywhere!

No he wasn’t, he demonstrated what would happen if a consumer played that audio without reducing it’s, through a DAC without headroom.

Strawman argument! I did not pretend every DAC for 30 years perfectly handles ISPs. Why don’t you try not deliberately misquoting/misrepresenting what I state for a change?

Strawman argument! I never stated or even implied ISPs are only theoretical. Under typical listening conditions most ISPs will not be genuinely audible because most DACs have at least some headroom. And again, Ian’s test did not “realistically” prove this because “realistically” consumers do not deliberately apply a clipping distortion plugin to their system!

Brilliant, you’re going for making yourself look foolish in two threads at the same time, a new low, you must be very happy with yourself! lol

G

Gregorio, come on...your accusing me of strawmen while making plenty yourself. You keep insiting you're being misrepresented, but the core issue remains: your repeatedly minimizing how frequently ISPs genuinly impact real-world listening scenarios.

You say "except services that dont implement dBTP," yet act as if that's some tiny corner case rather than a widespread reality. Automated rejection might exist, but it definitely isnt universal—plenty of problematic masters still get through regulary.

And w/r/t DAC headroom—seriously, most consumer gear doesn't offer ideal headroom margins, and clipping is common, especially at high playback levels or lower-quality devices. shepard’s clipping example wasn’t some artificial lab scenario; it closly mirrors what happens daily on real-world consumer setups.

You might not explicitly claim ISPs are purely theoretical, but your constant dismissal of practical tests and scenarios sends exactly that message. Stop hiding behind technicalities and recognize this: ISPs are still frequently audible, still widly problematic, and still relevant to everyday listeners.

 

[Ghoostknight]

So, to avoid further discussion i went ahead and downloaded two tidal files....

Codec: mp4a.40.2 vs flac

while this isnt what we are looking for (i guess the codec works somewhat correct in this case in terms of ISPs) its interesting to see that the clipping behaivour still changed slightly going from a clean hard clip on flac to something a bit more irregular looking on mp4a.40.2

 

[VNandor]

To me, it looks like it clips slightly less.

 

[bigshot]

If a player clips, it clips. That has nothing to do with the file format. It's a problem with the player.