Skip to main content

Notice

Please note that most of the software linked on this forum is likely to be safe to use. If you are unsure, feel free to ask in the relevant topics, or send a private message to an administrator or moderator. To help curb the problems of false positives, or in the event that you do find actual malware, you can contribute through the article linked here.
Topic: ReplayGain / EBUR128 True Peak Real World Impact / Killer Samples? (Read 2746 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

ReplayGain / EBUR128 True Peak Real World Impact / Killer Samples?

EBUR128 includes a true peak requirement.

Quote
Many loudness meters have a built-in true-peak meter, and what sets the true-peak meter apart from sample-peak meters is a special algorithm - donated by TC - that not only looks at the actual samples, but also inter sample peaks. In effect, the true-peak mater can unveil peaks in between actual samples that would otherwise cause distortion. Therefore, a true-peak meter actually 'goes beyond 0 dB'. A reading using a traditional sample-peak meter that displays a max of e.g. -0.2 dB could go as far as +3 dB on a true-peak meter reading.

https://www.tcelectronic.com/loudness-explained.html

With my very basic understanding of sampling theory, there could be samples that do not exceed the peak, but when they are played back, the continuous playback could compute to a peak above 0 dB. To counteract this you have to compute what the peak would be, which is computationally more expensive than just looking at the samples.

What's the practical impact of this? Are there any samples that demonstrate the impact of true peak vs. sample peak? Is there a mathematical maximum true peak that could exist based on every sample being below 0 dB?


Re: ReplayGain / EBUR128 True Peak Real World Impact / Killer Samples?

Reply #2
What's the practical impact of this? Are there any samples that demonstrate the impact of true peak vs. sample peak? Is there a mathematical maximum true peak that could exist based on every sample being below 0 dB?

There is no maximum. You can construct circumstances with arbitrarily high true peak. In theory. In reality, the signal has a peak before it was downsampling to the end-user format (say CD) - and, if you try to construct a signal with these very extreme true peaks, the peak will be narrow too. Very little sound pressure lost.
Furthermore, if you use ReplayGain to lower the volume of loud tracks, you guard against it.

Fiddling around back and forth with the resamplers included in foobar2000, I can reach the +10 dB range for two tracks off Merzbow's infamous "Veneorology" album and one track off Chicks on Speed Will Save Us All. Noise music.
ReplayGain volume adjustment is however larger except if you use "Album Gain" mode on the latter track, then you are 1 dB short. So if you play at volume 9 rather than volume 10, or your DAC allows those 3 dB headroom, it still doesn't peak over.

Listen from 1:40 here: https://www.youtube.com/watch?v=gr5mQTXkFOI . Listen from 1:40. The intentional distortion will mask "unintended" clipping quite well. (This is an exception from the trashy punk Chicks on Speed would normally do. Merzbow, on the other hand, does noise for a living.)

Re: ReplayGain / EBUR128 True Peak Real World Impact / Killer Samples?

Reply #3
What's the practical impact of this? Are there any samples that demonstrate the impact of true peak vs. sample peak? Is there a mathematical maximum true peak that could exist based on every sample being below 0 dB?

There is no maximum. You can construct circumstances with arbitrarily high true peak. In theory.
I'm curious how you'd do that. The highest true peak I get through experimentation is +8.4 dB, with a waveform like this (normalized to a maximum value of ±1):
1, -1, 1, -1, ... 1, -1, 1, -1, 1, 1, -1, 1, -1, 1, ... -1, 1, -1, 1 or the negation thereof.

Chris
If I don't reply to your reply, it means I agree with you.

Re: ReplayGain / EBUR128 True Peak Real World Impact / Killer Samples?

Reply #4
An interval with oscillations between -1 and 1 (N times). Zeroes outside the interval.
Halfway between the last "1" and the "0", true peak is logarithmic in N. See https://techblog.izotope.com/2015/08/24/true-peak-detection , the end.

Re: ReplayGain / EBUR128 True Peak Real World Impact / Killer Samples?

Reply #5
I'm curious how you'd do that. The highest true peak I get through experimentation is +8.4 dB, with a waveform like this (normalized to a maximum value of ±1):
1, -1, 1, -1, ... 1, -1, 1, -1, 1, 1, -1, 1, -1, 1, ... -1, 1, -1, 1 or the negation thereof.
The attached files used the command:
sox before.wav after.wav rate -b99.7 88200
The reference filter in BS.1770 is not long enough to produce a higher peak, and...
https://izotope-rx.livejournal.com/5760.html?thread=1920#t1920
Quote
In my opinion, this matter of ISP's was blown out of proportion by companies desiring to sell meters and limiters. The 1 or 2 dB range reserved for peaks would be better used for encoding musical dynamic range.
https://www.tcelectronic.com/loudness-explained.html
Quote
Many loudness meters have a built-in true-peak meter, and what sets the true-peak meter apart from sample-peak meters is a special algorithm - donated by TC
TC Electronic sells meters and limiters.

Re: ReplayGain / EBUR128 True Peak Real World Impact / Killer Samples?

Reply #6
Thanks to the both of you.
The reference filter in BS.1770 is not long enough to produce a higher peak...
Ah, yes, neither is Audition's sinc-like waveform interpolator, from which I read off those +8.4 dB. With higher-quality 2x upsampling, I get near +12 dB overshoot on that signal of mine, so it seems to behave similarly to the experiment Porcus linked to.

But then 12 dB of headroom, which seems to be quite popular in the industry, is more than enough. Ignoring our pathological test cases, 6 dB (i.e., one extra PCM bit) should suffice as well.

Chris
If I don't reply to your reply, it means I agree with you.

Re: ReplayGain / EBUR128 True Peak Real World Impact / Killer Samples?

Reply #7
Searching for the highest peaks among my lossless files, I find a handful that are not noise music with <replaygain_track_peak> value between 1.7 and 1.9. I used a slow resampler, fb2k's SSRC. (Its auto 8x doesn't give me anywhere that loud.)

Disregarding the noise music, I find Rush, Royal Hunt and Scorpions - and Depeche Mode: Dream On.

Jazz: a live track from Esbjörn Svensson Trio is tagged at > 1.6

Re: ReplayGain / EBUR128 True Peak Real World Impact / Killer Samples?

Reply #8
I recorded a Merzbow clip from the output of my soundcard. The first 5 seconds followed by a 1 second of silence are not part of the Merzbow, but a tone at 0dBFS for alignment purpose, so everything higher than the tone are overshoots. The DAC chip on the soundcard (PCM1794A) does not have any headroom above 0dBFS, I reduced the digital playback level to allow overshoot without clipping, about 10dB of reduction is required for this track. The sample pattern on the highest peak looks very similar to the test signal being discussed, makes me believe Merzbow is a mathematician.

The BS.1770 filter in the iZotope article only contains 48 samples, many DAC chips use 2-3x more samples (so, somewhere between 96-144) for the first stage of 2x upsampling. Some software resamplers allow very long filters but extreme settings cannot reflect behaviour of typical chips.

Re: ReplayGain / EBUR128 True Peak Real World Impact / Killer Samples?

Reply #9
makes me believe Merzbow is a mathematician.
This particular album was created under the influence of significant amounts of alcohol, with Merzbow throwing in these singularities to observe the don't drink and derive rules.

(The alcohol part is from an interview with Akita. Oh, and Slave New Desart is one of the two tracks I referred to in reply #2, the other is Klo Ken Phantasie.)

Re: ReplayGain / EBUR128 True Peak Real World Impact / Killer Samples?

Reply #10
In my opinion, this matter of ISP's was blown out of proportion by companies desiring to sell meters and limiters. The 1 or 2 dB range reserved for peaks would be better used for encoding musical dynamic range.
Sorry for the noob question, but what does this mean? You're saying reserving e.g. 1 dB for peaks (true peaks going beyond sample peaks) means you're left with 1 less dB of dynamic range, for e.g. a classical piece that may have very quiet and very loud sections?

Possible hints at the answer:
https://wiki.hydrogenaud.io/index.php?title=ReplayGain_specification#Clipping_prevention
http://forum.cakewalk.com/Why-extra-headroom-is-needed-for-MP3s-m1689770.aspx

Re: ReplayGain / EBUR128 True Peak Real World Impact / Killer Samples?

Reply #11
Disregarding the noise music, I find Rush, Royal Hunt and Scorpions - and Depeche Mode: Dream On.
loudness-scanner vs sox rate 8x:
Code: [Select]
Sample peak  True peak  True peak (sox)
      -0.01        1.7            1.77  - Dead Can Dance  Anastasis               02. Anabasis.flac
       0.00        1.6            1.73  - Satriani        Is There Love in Space  02_UpInFlames.flac
      -0.10        1.7            1.68  - Iron Maiden     Rock in Rio             06. The Number Of The Beast.flac
      -0.52        2.3            2.74  - Nightwish       Highest Hopes (Limited Edition)  11_DeepSilentComplete.flac
      -0.33        2.0            2.42  - Nightwish       Highest Hopes (Limited Edition)  12_SacramentOfWilderness.flac

Re: ReplayGain / EBUR128 True Peak Real World Impact / Killer Samples?

Reply #12
In my opinion, this matter of ISP's was blown out of proportion by companies desiring to sell meters and limiters. The 1 or 2 dB range reserved for peaks would be better used for encoding musical dynamic range.
Sorry for the noob question, but what does this mean? You're saying reserving e.g. 1 dB for peaks (true peaks going beyond sample peaks) means you're left with 1 less dB of dynamic range, for e.g. a classical piece that may have very quiet and very loud sections?

Possible hints at the answer:
https://wiki.hydrogenaud.io/index.php?title=ReplayGain_specification#Clipping_prevention
http://forum.cakewalk.com/Why-extra-headroom-is-needed-for-MP3s-m1689770.aspx
https://izotope-rx.livejournal.com/5760.html
You can scroll up and read the main article above the comment, because the quoted text was a reply to the article, and the article is much more detailed than the two links you provided. Also, because I only quoted the text but I am not the same person who wrote that comment, so I can't tell if that person was talking about highly dynamic classical music or not.

Just in case you did not actually click the link to read the whole comment of that person, click this and read it:
https://izotope-rx.livejournal.com/5760.html?thread=1920#t1920

Re: ReplayGain / EBUR128 True Peak Real World Impact / Killer Samples?

Reply #13
Just for fun, download the attached m4a file in the post below and analyze it. I made the arrangement at around 2010 as the game (Final Fantasy XIII) was released at that time. When I made that arrangement, the tools I used don't have any facility to measure "true" peak (based on a 48 samples filter?) or codec induced overs.
https://hydrogenaud.io/index.php/topic,114240.msg940944.html#msg940944

Re: ReplayGain / EBUR128 True Peak Real World Impact / Killer Samples?

Reply #14
Disregarding the noise music, I find Rush, Royal Hunt and Scorpions - and Depeche Mode: Dream On.

Jazz: a live track from Esbjörn Svensson Trio is tagged at > 1.6
The highest one in my collection that falls into this category:
https://www.audiosciencereview.com/forum/index.php?threads/correlation-between-sample-rate-and-audible-frequency.9411/post-304560