Mastering Engineer crushing it for the sake of it

[ptalbot]

I received a mp3 file from a young artist (that shall remain unknown), asking for my opinion.
I noticed that the sound was pretty “crunchy” and distorted in a bad way (especially during the last chorus), so I’ve checked with the free Orban Loudness Meter to confirm, and this is what it looks like (note the end part of the sound wave which is flat):

The song is mastered at -9.7 LUFS which sound OK, although I believe it would sound better with less compression. But what boggles me is that the file has 6911 reconstructed (inter-sample) peaks over 0dBFS, and up to 3.2 dBFS - which to me is pretty crazy.
I can imagine that the mp3 compression could produce a few overs (unless you give it a margin), but not up to +3.2dBFS!

To note that this has been done by a Mastering studio which has 6 seasoned mastering engineers, and a long list of professional productions…

Am I right telling this artist that she should ask for a new master, with a proper brickwall limiter, and a lower output of say -.5dBFS before mp3 compression, or do you think it doesn’t matter???

[bozmillar]

Am I right telling this artist that she should ask for a new master, with a proper brickwall limiter, and a lower output of say -.5dBFS before mp3 compression, or do you think it doesn’t matter???

If you want a new master, then yes, you are right to ask for a new one. But…

But what boggles me is that the file has 6911 reconstructed (inter-sample) peaks over 0dBFS, and up to 3.2 dBFS - which to me is pretty crazy.

Intersample peaks are completely meaningless. They, by definition, create harmonics outside the audible range. In order to prevent intersample clipping you have to compress harder, which tends to sound, well, more compressed. People preach about the importance of preventing intersample clipping, but I’ve never actually heard a valid argument as to why they are bad.

3.2 dBFS - which to me is pretty crazy.
I can imagine that the mp3 compression could produce a few overs (unless you give it a margin), but not up to +3.2dBFS!

3.2dB over is perfectly normal for a limiter that doesn’t have true peak detection. And it’s very normal for an MP3 to reach that level, even if the limiter does have true peak limiting. Now, the audibility of distortion caused by mp3 conversion is a different story to intersample clipping, and may be something you need to watch out for.

[Cirrus]

The end does indeed looked smashed to hell and back, but you can’t really tell much or judge mastering from looking at a waveform.

For example… maybe this is a song that really builds, and it’s preceded by a really loud wall-of-noise song. In that case, the mastering engineer might decide that he couldn’t have the intro of this song too quiet, because otherwise it sounds weedy following the loud song.

And maybe the mixer has left the mix too dynamic, and relied on a massive increase in volume over the song to provide energy and musical drama. In that case, the mastering engineer might find compressing the track over its length causes the song to lose too much energy.

So they need to find a compromise between having the intro loud enough to sit in context of the album’s flow, and getting the end as loud as possible without sounding compressed.

In that case, plain ol’ clipping at the end, which is what this track looks to have, might well sound better, more dynamic and punchy, than reducing the dynamic range and trying to preserve headroom and transients.

Now, obviously I don’t know the context this song was mastered in, so there’s every chance all that is a load of rubbish, but that’s just illustrating that sometimes mastering people need to work compromises to get the best out of what they were given.

But obviously, if it actually sounds bad, then that’s not good.

[ptalbot]

They, by definition, create harmonics outside the audible range.

Could you elaborate? My understanding of intersample peaks is that they can happen at any frequency, as a result of the sampling process. Say you have a peak at 500hz where the level goes over 0dB, the sampling process cannot store the value it so it clips it to the max available value at the bit rate, but the peak still exists (as a kind of “phantom” result of the waveform) and the DA converter will attempt to recreate it.
Now all DA converter are not created equal and thus I would think it can create distortion, but I would think that the distortion happens at this frequency (and perhaps some upper harmonics) but still in the audible range.

3.2dB over is perfectly normal for a limiter that doesn’t have true peak detection. And it’s very normal for an MP3 to reach that level, even if the limiter does have true peak limiting.

It’s not been my experience. If I output a WAV file at 0dB with some limiting, even with a slow attack, I might see a few peaks over 0dB during mp3 compression, but at most around +0.5dB. If I want to keep these peaks and not limit more, I will output at -0.5dB and then it should be fine.

In this case 3.2dB is a lot, and I can’t believe that this is the process of the mp3 compression alone. When I look at the waveform it looks like this was already the case of the output, and mp3 compression has probably made it worst (and worst sounding).

I would expect a mastering engineer to strive to get me the best sound, without artifact, that translates well over a wide range of DA converters. In that case, I would definitely think that the ME has been pretty careless about the peaks.

[ptalbot]

Unfortunately I cannot post the song here, but it’s mainly acoustic pop, with good dynamic at the start, and then yes, there is a kind of wall of sound distorted guitars at the end, with big toms. To me this doesn’t sound too good at the end, and I would think that a ME could also automate these things to keep a good dynamic overall without crushing it to where there’s no dynamic left in the end.

[AJ113]

Am I right telling this artist that she should ask for a new master, with a proper brickwall limiter, and a lower output of say -.5dBFS before mp3 compression, or do you think it doesn’t matter???

Firstly, we’re looking at an mp3. I would be wary of making any kind of judgement call until I had an uncompressed file in my hands.

Notwithstanding the above:

In my opinion, there is too much made about inter sample peaks. I have never once heard any playback discrepancy that I could attribute to the phenomenon.

Having said that, I always master with no inter sample peaks, because you never know who is going to see the file and analyse it, and you never know what their expectations may be - misguided or otherwise. It’s a very simple process, so I see no reason not to do it.

Mastering to a lower level of -0.5db (some people go as low as -.3db) is not the way to go in my opinion, not least because it offers no guarantee of solving the perceived problem. In addition, is it really worth producing quieter recordings - which is something we can hear - for the sake of trying to resolve a non-issue that we can’t hear?

Every commercial recording I have ever seen has been mastered to 0db, so I think it is wise to adhere to this apparent norm.

[ptalbot]

Firstly, we’re looking at an mp3. I would be wary of making any kind of judgement call until I had an uncompressed file in my hands.

I realize that it’s not conclusive enough, however I’ve never seen a WAV file with no intersample peaks issue that would produce +3.2dBFS after mp3 compression. I would be very surprised if that was possible, unless maybe you compress to a very low bitrate, which was not the case here (320kpbs).

[Tesgin]

Okay, so my thoughts on this might be over-simplistic. I don’t want to put myself forth as an authority with this at all (y’all are way over my head when you’re talking about intersample peaks!), however let me share a few thoughts and observations, and I’d LOVE to hear if I’m off-base.

So, one thing is that it’s not clear to me (I might have missed it) if this mp3 was mixed for streaming, or if it’s just a draft so to speak of the wav file that was mixed for CD.

So, if mixed for streaming, one thing is that the integrated LUFS is a bit high. They dynamic range is unnecessarily compromised. Not bad, but -16 to -14 is more the target. Now that won’t speak to the distorted sound you’re hearing; however, more important is the PLR-S (peak to loudness ratio) of the loudest parts of the song. People like Ian Shepherd say that should be 8 or more. So, for sure going to the loudest part of this song – right at the end – it’s not going to be anywhere near that. It would appear that that part of the song is over-compressed.

Even that doesn’t speak directly to the distorted sound you’re hearing, but I’d be curious at what point in the song you’re hearing that? I am taking it, from your comment, that when you say “last chorus” it’s that part of the song. I would expect, from just looking at the wave form, that it might sound “crunchy” (could you define what you mean by that?).

I’m not familiar with the Orban – it looks pretty cool. I’ll check it out. I’m curious if it’ll tell you the true peak values? Have you checked out the Youlean Loudness Meter? It has some pretty helpful metrics. I have to suspect that the true peak values are over the top. If so, that would speak to the crunchy, distorted sound you’re hearing.

It’s also not unheard of that songs are mastered like this to maximize loudness. However, in this world of streaming audio, at least, the loudness wars are over, as they say.

At least in that world, this song would appear to be overcompressed. A ton is lost in the dynamic range.

So, of course loudness meters and LUFS values and true peak values aren’t the be-all and end-all, but they give you visual data to clue you in to when something’s wrong. I don’t wanna go overboard with numbers and graphs. The bottom line is that, if it sounds good, it is good. And you’re saying it ain’t sounding good. And the loudness numbers are consistent with what you’re hearing, and probably what you’re client is hearing (or he/she wouldn’t be asking for your opinion on the mix).

I’m eager to hear from y’all if I’m being too simplistic or outright off-base altogether. Heck, I’m new to this. But that’s what I’m seeing.

So, for me the three big pieces for your client are about what the media is being mixed for (CD or streamin or other), and what the true peak values are…

… BUT there is a third thing you might want to check out:

I’d be very curious to see what the Pre-Master looks like that they sent for mastering. If that’s where the damage is, there’s no amount at mastering that’ll fix that. Be sure to check that out before telling them to ask for a new master.

If the Pre-Master is cool, it wouldn’t seem unreasonable to me to suggest they ask for a remastering if they’re not happy. The meters as well as your ears would seem to suggest their concerns are possibly well-merited.

[bozmillar]

My understanding of intersample peaks is that they can happen at any frequency, as a result of the sampling process.

For an intersample peak to shoot significantly above the sample, there needs to be a rapid change in value (high frequency). Higher frequencies will produce higher intersample peaks. Drums with a sharp attack (high frequency content) are the most common culprit of instersample peaks going above 0dB, but they have the advantage of being not being cyclic nor tonal, so distortion is rarely an issue.

Also, pretty much any modern D/A converter isn’t going to clip intersample peaks, so even if they do go avove 0dB, they’re probably not clipping on the playback system anyway.

If I output a WAV file at 0dB with some limiting, even with a slow attack, I might see a few peaks over 0dB during mp3 compression, but at most around +0.5dB. If I want to keep these peaks and not limit more, I will output at -0.5dB and then it should be fine.

It sounds like you are talking about two separate things in the same sentence. Intersample peaks and clipping caused by mp3 distortion are two completely separate things. I can’t speak for mp3 encoders/decoders on whether or not the distortion is audible, but my guess is that it is. It all depends on if the decoder is decoding in such a way that anything ove 0dBFS clips or not.

But preventing intersample peaks above 0dBFS is not going to prevent mp3 clipping from happening. It might reduce it a bit because the output level will be slightly lower.

In this case 3.2dB is a lot, and I can’t believe that this is the process of the mp3 compression alone. When I look at the waveform it looks like this was already the case of the output, and mp3 compression has probably made it worst (and worst sounding).

3.2dB is normal for something that is clipping. Hard clipping can definitely shoot intersample peaks that high above 0dB. But if that’s what’s going on, it’s not the intersample clipping your are hearing, it’s the baked in clipping, which can be very much audible, especially if not done right.

I would expect a mastering engineer to strive to get me the best sound, without artifact, that translates well over a wide range of DA converters. In that case, I would definitely think that the ME has been pretty careless about the peaks.

I agree. Although I don’t know if it’s carelessness or if it’s intentional. Clipping is a pretty common step in boosting levels, although it’s really easy to overdo if you don’t do it right. I’ve heard plenty of masters that clip way harder than I would think is reasonable.

[ptalbot]

So, one thing is that it’s not clear to me (I might have missed it) if this mp3 was mixed for streaming, or if it’s just a draft so to speak of the wav file that was mixed for CD.

As far as I know that mp3 was produced for streaming, yes.

Even that doesn’t speak directly to the distorted sound you’re hearing, but I’d be curious at what point in the song you’re hearing that? I am taking it, from your comment, that when you say “last chorus” it’s that part of the song. I would expect, from just looking at the wave form, that it might sound “crunchy” (could you define what you mean by that?).

It definitely sounded distorted to me, and over compressed, compared to the rest of the song. Personally, I think some of the impact of that last chorus (big distorted guitars and big toms) was lost because of the loss of dynamic. Upon hearing it, I wanted to look at what could cause this, which is why I fired Orban Loudness meter to confirm my suspicion. I’m not familiar with the Youlean Loudness Meter, but when mixing I’m using Izotope Insight.

I’d be very curious to see what the Pre-Master looks like that they sent for mastering. If that’s where the damage is, there’s no amount at mastering that’ll fix that. Be sure to check that out before telling them to ask for a new master.

I don’t have access to the pre-mastered wav file, unfortunately.

[ptalbot]

For an intersample peak to shoot significantly above the sample, there needs to be a rapid change in value (high frequency). Higher frequencies will produce higher intersample peaks. Drums with a sharp attack (high frequency content) are the most common culprit of instersample peaks going above 0dB, but they have the advantage of being not being cyclic nor tonal, so distortion is rarely an issue.

I see. Makes sense, thanks for the clarification!

3.2dB is normal for something that is clipping. Hard clipping can definitely shoot intersample peaks that high above 0dB. But if that’s what’s going on, it’s not the intersample clipping your are hearing, it’s the baked in clipping, which can be very much audible, especially if not done right.

That’s quite possible, yes.

[Tesgin]

I don’t have access to the pre-mastered wav file, unfortunately.

Yeah, the more I think of it, I’d want to check that out with him before suggesting he ask for remastering. Did he do his own pre-mastering and send it to be mastered, or did he send the multi-tracks and the engineer mixed and mastered.

In either event, it’d be good to clarify what exactly he sent. Maybe the damage was in the original tracks.

[Tesgin]

3.2dB over is perfectly normal for a limiter that doesn’t have true peak detection. And it’s very normal for an MP3 to reach that level, even if the limiter does have true peak limiting.

Boz, could you clarify what you mean to me: when you say “normal” do mean that as in “no big deal, not necessarily bad,” or do you mean that as in “damaged, but a common mistake if one isn’t looking at true peak”?

My understanding is that if one pushes the limiter right up to 0dB, there can be subtle clipping nonetheless when the digital/audio converter rounds off the waveform, and the rounding off that occurs between samples can push things over 0 dB. Am I correct with that? 3.2 would seem to be quite a bit, even for intersample peaks, which would generally be more subtle and “maybe” not even noticeable. To get it that high, I’m thinking the signal must have been pushed well past 0 dB. I’m thinking that to hit 3.2 it was clipping already, without any consideration of ISP.

Just trying to understand. Is that what you’re saying?

Thanks in advance,
Tesgin

[bozmillar]

Boz, could you clarify what you mean to me: when you say “normal” do mean that as in “no big deal, not necessarily bad,” or do you mean that as in “damaged, but a common mistake if one isn’t looking at true peak”?

I guess by normal I meant that it’s not surprising to see levels that high above 0dBFS. Whether it’s ok or bad depends on the actual sound that it produces, so I can’t really speak to whether or not it’s ok. According to the orignal post, it seems like it has audible distortion, so I would put that in the bad category.

My understanding is that if one pushes the limiter right up to 0dB, there can be subtle clipping nonetheless when the digital/audio converter rounds off the waveform, and the rounding off that occurs between samples can push things over 0 dB. Am I correct with that? 3.2 would seem to be quite a bit, even for intersample peaks, which would generally be more subtle and “maybe” not even noticeable. To get it that high, I’m thinking the signal must have been pushed well past 0 dB. I’m thinking that to hit 3.2 it was clipping already, without any consideration of ISP.

well, for example, a square wave can overshoot by 2dB.

But if you have other high frequency stuff going on, that over shoot can easily hit +4dB. This is what it looks like when I move just a couple of samples around:

But, it’s important to realize that any deviation from this line is by definition distortion taking place above nyquist frequency. Whether it’s clipping, or doing anything crazy, it’s high frequency content that should be inaudible. Of course, different D/A converters could react differently, but I haven’t heard any.

[Descent]

How does the song sound though? Is there any way we can hear it?

[ptalbot]

As I said, unfortunately I don’t have the permission from the artist to share the song, it’s not been released yet. Hopefully when it is, I’ll be able to point you to a place where to listen to it.

[Descent]

That’ll be great. What do you think, once you listened to it, without looking at the screen, does the distortion work with the song?

[ptalbot]

As I said:

It definitely sounded distorted to me, and over compressed, compared to the rest of the song. Personally, I think some of the impact of that last chorus (big distorted guitars and big toms) was lost because of the loss of dynamic.

So no, I don’t think it worked with the song.

[AJ113]

But, it’s important to realize that any deviation from this line is by definition distortion taking place above nyquist frequency.

This is why I said it’s a (potentially) audible solution to an inaudible problem. I mean if it’s inaudible, is it really a problem?