The third installment in the pre-release series...
http://www.mars.org/home/rob/proj/mpeg/mad-plugin/in_mad-0.12.3b-pre3.zip
This version adds configurability to the automatic attenuation, and also fixes the problem with the joint stereo pie graph.
I decided it might be better to control the sensitivity of the auto attenuation rather than use a simple hard limit. The default is somewhat less aggressive than it was before (which would correspond to the "Most" setting.) The idea is to control how gradually the attenuation is increased. If only a few isolated samples clip, a less sensitive setting will attenuate less than a more sensitive setting would. Continuous clipping will continuously increase the attenuation until the signal no longer clips; the sensitivity controls how fast this occurs.
Let me know what you think.
Cheers, -rob
The pre-releases just keep getting cooler, thanks Rob! The auto attenuation is fantastic. On one track I have, ZZ Top's 'La Grange', taken from the Armageddon soundtrack (which has incredible amounts of clipping on the CD itself), and compressed into a VBR file by LAME, on MAD's default clipping setting it went from 32661 clipped samples to a mere 31 clipped samples with -1.1dB attenuation (23 at the Most setting, but still -1.1dB attenuation). On another track, this time by Christina Aguilera, there was a point in the song where she goes to stratospheric heights with her vocals, and it always seemed to distort with speakers and even with my good headphones, but after attenuation, smooth as silk. The top end scratchiness during loud segments were decidedly smoother, which is the whole point of attenuation I suppose. Well, it works!
I have a question though, how does the gradual attenuation work? How about something that clips yet isn't continuous, like a cymbal crash? And are all those graduations for the setting necessary? I think instead of this, a self-adjusting attenuation would be even better. It should work at an aggressive level however, and step back or un-attenuate as soon as the clipping stops. What I'm thinking is something like the concept of variable bitrate MP3 encoding. When the decoder sees the first instance of a clipped sample, it automatically attenuates the full level, but after so many samples of no clipping, it steps back to no attenuation. And so on. What I fear is that a large spike could occur in the beginning or middle of a file, setting a high attenuation level, and then muffling the rest of the song which might not clip at all.
And the Joint Stereo pie graph is very neat. There's just so much to look at now. The MAD plug-in is becoming a tweaker's dream come true. Is all this pre-releasing building up to a Christmas day present for all the devoted good little MAD users out there? If so, reading Xing VBR headers is on top of my Santa list!
:)
Thanks,
naw
Rob Leslie wrote:
The third installment in the pre-release series...
http://www.mars.org/home/rob/proj/mpeg/mad-plugin/in_mad-0.12.3b-pre3.zip
This version adds configurability to the automatic attenuation, and also fixes the problem with the joint stereo pie graph.
I decided it might be better to control the sensitivity of the auto attenuation rather than use a simple hard limit. The default is somewhat less aggressive than it was before (which would correspond to the "Most" setting.) The idea is to control how gradually the attenuation is increased. If only a few isolated samples clip, a less sensitive setting will attenuate less than a more sensitive setting would. Continuous clipping will continuously increase the attenuation until the signal no longer clips; the sensitivity controls how fast this occurs.
Let me know what you think.
Cheers, -rob
The pre-releases just keep getting cooler, thanks Rob! The auto attenuation is fantastic. On one track I have, ZZ Top's 'La Grange', taken from the Armageddon soundtrack (which has incredible amounts of clipping on the CD itself),
There is something I don't understand. Clipping is sound above the maximum allowed sound. As the sound of a CD is stored on a CD, and the maximum amplitude sound that can be stored on a CD is obviously the maximum allowed intensity on a CD, how can there be clipping on a CD, ie the sound of the CD exceding the sound that can be stored on this CD. That seems strange to me.
Regards,
--
Gabriel Bouvigne - France bouvigne@mp3-tech.org mobile phone: gsm@mp3-tech.org icq: 12138873
MP3' Tech: www.mp3-tech.org personal page: gabriel.mp3-tech.org
On Thu, 14 Dec 2000 20:35:55 +0100, Gabriel Bouvigne wrote:
The pre-releases just keep getting cooler, thanks Rob! The auto attenuation is fantastic. On one track I have, ZZ Top's 'La Grange', taken from the Armageddon soundtrack (which has incredible amounts of clipping on the CD itself),
There is something I don't understand. Clipping is sound above the maximum allowed sound. As the sound of a CD is stored on a CD, and the maximum amplitude sound that can be stored on a CD is obviously the maximum allowed intensity on a CD, how can there be clipping on a CD, ie the sound of the CD exceding the sound that can be stored on this CD. That seems strange to me.
I'll repeat what I sent to Peter.
You're thinking lossless compression here, 'cause you won't get a signal different from the original with lossless compression such as Monkeyaudio. However, MP3 reconstructs an audio signal from less than the original signal plus noise. Add for example one frequency x and another frequency 2*x, then you'll have some areas that are boosted, and some that are attenuated. When 2*x is filtered out again(or gets discarded in some places, because the encoding algorythem thought you won't hear it) then some places will get attenuated that were peaks and some will be boosted that weren't peaks before. And sometimes in complex signals that are mastered to use the maximum of the 16 bit resolution on a CD and are mastered to extreme loudness(single releases are mangled this way a LOT), the signal's reconstruction overshoots, because the stuff that got filtered out in the encoding process meant that the tightly controlled frequencies of the signal will have some things missing. You get some peaks and attenuations you previously didn't have. You can check this easily, by taking a ripped track and using an EQ in a sample editor to EQ 100Hz by -6 dB for example. You almost always get clipped samples, because of the great energy that lower frequencies take and thus command more amplitude and heavier changes to the signal.
Tony
Gabriel Bouvigne wrote:
The pre-releases just keep getting cooler, thanks Rob! The auto attenuation is fantastic. On one track I have, ZZ Top's 'La Grange', taken from the Armageddon soundtrack (which has incredible amounts of clipping on the CD itself),
There is something I don't understand. Clipping is sound above the maximum allowed sound. As the sound of a CD is stored on a CD, and the maximum amplitude sound that can be stored on a CD is obviously the maximum allowed intensity on a CD, how can there be clipping on a CD, ie the sound of the CD exceding the sound that can be stored on this CD. That seems strange to me.
Strange but true. Modern CD's are mastered to sound very loud and sometimes sound engineers go overboard with it. A lot of pop CD's are compressed as well, to magnify their sound even more. So on a prudently engineered CD track, you have a wave form with peaks and lows, how it should be; however on one of these compressed and amplified CD tracks, you see virtually no peaks and lows, but a wave form that is pretty much solid throughout the entire dynamic range, and a lot of times goes beyond the dynamic range (clips). Actually making an audio CD clip is fairly easy. All that's stored on a CD is 16-bit wave files. You can open up a 16-bit wave file in any sound editing program and amplify it to induce clipping. Nothing too hard about it. However, I can compress and amplify a 16-bit wave file without inducing clipping, so why there are CD's out there with this problem is indeed strange. I think it's more often than not CD's being mixed for low-end audio equipment, which don't necessarily reveal these deficiencies. Or even just deaf audio engineers. It's pretty well known that many 'experienced' audio engineers have very bad hearing, after years of loud music all day long, so they use a lot of equalization and special equipment to help them hear the music. It's a wierd reality, but true as well. Maybe it's a combination of both of these factors which account for the current sad state of digital recordings.
David Shin naw@SoftHome.net wrote:
I have a question though, how does the gradual attenuation work?
I thought I explained it succinctly, but I'll try again. :-)
Whenever clipping occurs, the peak clipped value is examined and a new attenuation setting is calculated. Previously (or with the "Most" sensitivity setting) the new attenuation is one that would have avoided the clipping entirely. This can cause sharp drops in the output signal level that may not be desirable, especially if it was only a few clipped samples.
With a less sensitive setting, the new attenuation is somewhere between the previous setting and one that would avoid the clipping entirely. This way, if clipping continues, the attenuation will gradually still be increased to the point that clipping no longer occurs.
Once increased, the attenuation is never decreased unless the "Reset" button is pressed in the Statistics tab.
How about something that clips yet isn't continuous, like a cymbal crash?
A one-time cymbal crash that causes only a few clipped samples will trigger the auto attenuation. How much attenuation occurs as a result is determined by the amount of clipping and the sensitivity setting. Since it's a one-time event, a less sensitive setting would have a less drastic effect than a more sensitive setting. In either case the setting is permanent until clipping occurs again.
And are all those graduations for the setting necessary?
Maybe not. It wasn't clear to me how much granularity would be desirable, so I picked something more or less out of a hat. This is where I value feedback: if people think less (or more) granularity is good, I'll change it; I don't have strong feelings about it.
Is all this pre-releasing building up to a Christmas day present for all the devoted good little MAD users out there? If so, reading Xing VBR headers is on top of my Santa list!
Santa is making a list and checking it twice. ;-)
Cheers, -rob
Rob Leslie wrote:
David Shin naw@SoftHome.net wrote:
I have a question though, how does the gradual attenuation work?
I thought I explained it succinctly, but I'll try again. :-)
Whenever clipping occurs, the peak clipped value is examined and a new attenuation setting is calculated. Previously (or with the "Most" sensitivity setting) the new attenuation is one that would have avoided the clipping entirely. This can cause sharp drops in the output signal level that may not be desirable, especially if it was only a few clipped samples.
With a less sensitive setting, the new attenuation is somewhere between the previous setting and one that would avoid the clipping entirely. This way, if clipping continues, the attenuation will gradually still be increased to the point that clipping no longer occurs.
Once increased, the attenuation is never decreased unless the "Reset" button is pressed in the Statistics tab.
Ah, that clarified it. I just wanted to know where that "between" point was for settings less than Most. This question sort of went along with my "do you need all those gradations?" I think a Most and then a Least and Moderate setting is all you'd really need. But I think 8 gradations is overkill, and doesn't offer any real benefit. Yet I think a coninuously changing attenuation would be even better. It doesn't even need to look ahead, like the limiter that Anthony is talking about.
Is all this pre-releasing building up to a Christmas day present for all the devoted good little MAD users out there? If so, reading Xing VBR headers is on top of my Santa list!
Santa is making a list and checking it twice. ;-)
I'll be sure to leave out my virtual milk and cookies!
---------- | | <------Oreo ----------
-
Anthony Airon Oetzmann -
David Shin -
Gabriel Bouvigne -
Rob Leslie