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: occasional high-frequency dropouts in Lame (Read 7710 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

occasional high-frequency dropouts in Lame

well, I know that CBR is more and more getting out of interest while VBR is improving, but I would like to point out quite annoying artifact in Lame CBR.
I took a part from the scooter sample and looped it (because the dropouts are only occasional) and then encoded with 3.98a4 --preset cbr 128.
This is a zoomed original wave file:



This is the same part zoomed in Lame encoded file:


It adds a short "water splash" sound to the clap.

Just to mention, this type of artifact is not present in lame pre-3.93, but is actually worse in 3.96.1.

J.M.

[edit> wrong lame version in topic title; it should be also 3.98a4]

occasional high-frequency dropouts in Lame

Reply #1 –
don't worry about it, it is part of Lame's audio model. If determined not being perceivable for humans it will drop certain frequencies, because high frequencies are the most expensive ones when it comes to compression since they need much more bandwidth than lower frequencies. The bandwidth saved can be used instead to make sure that the perceivable part is as indistinguishable as possible to the original.
--alt-presets are there for a reason! These other switches DO NOT work better than it, trust me on this.
LAME + Joint Stereo doesn't destroy 'Stereo'

occasional high-frequency dropouts in Lame

Reply #2 –
It adds a short "water splash" sound to the clap.

did you not read this?

maybe it is a lame enhancement

occasional high-frequency dropouts in Lame

Reply #3 –
MP3 + 128 kbps + CBR = occasional artifacts.
It's well-known. A bit more worrying are possible regressions compared to previous LAME version.
I noticed similar problems with micro-impulse sample (trombone): 3.90 was unsharp and noisy whereas 3.97 alpha were crisper (more detailed, hence better) but with ugly frequency dropouts. It reminds me old DivX SBC encoders: crisper, detailed but blocky compared to oversmoothed and washed RV9-like encoders. It's maybe a matter of taste, I don't know.
Perhaps your problem is different. Anyway, if you've looped the same sample several time to get this problem, I wouldn't worry too much about it

occasional high-frequency dropouts in Lame

Reply #4 –
Quote from: Jojo on
don't worry about it, it is part of Lame's audio model. If determined not being perceivable for humans it will drop certain frequencies, because high frequencies are the most expensive ones when it comes to compression since they need much more bandwidth than lower frequencies. The bandwidth saved can be used instead to make sure that the perceivable part is as indistinguishable as possible to the original.

It is actually quite annoying. The strange thing it that it is fine on 15 claps but the 16th is destructed 
But it has improved since 3.96, in which the dropout was extending down to 7kHz (or so) and was very annoying.

Quote from: guruboolez on
Anyway, if you've looped the same sample several time to get this problem, I wouldn't worry too much about it

well, it occured once (or twice) in the original track, I just wanted to try to squeeze the worst artifact from it
actually it makes similar artifacts on more of my tracks. It's always like the sharp transient is encoded properly, then there is a dropout, and then the spectrum again "returns".
As I said, this type of artifact is not present in lame 3.93.1 and lower.

J.M.

occasional high-frequency dropouts in Lame

Reply #5 –
Quote from: jmartis on
It is actually quite annoying. The strange thing it that it is fine on 15 claps but the 16th is destructed 
But it has improved since 3.96, in which the dropout was extending down to 7kHz (or so) and was very annoying.

I misread your question. Anyway, as guruboolez said, --preset 128kbps is not designed for full transparency. Lame is constantly tweaked, which might show some regressions to previous releases but at the same time fixes many other problems. Try --preset standard and see if it still happens.
--alt-presets are there for a reason! These other switches DO NOT work better than it, trust me on this.
LAME + Joint Stereo doesn't destroy 'Stereo'

occasional high-frequency dropouts in Lame

Reply #7 –
Just noticed a very nasty dropout on the first kick, extending down to 2kHz. I havent found this previously because I was focusing only on claps.
(I really dont know much about it, but maybe it has something to do with the "Adaptive ATH" ??)

orig. sample:



the nasty dropout:

occasional high-frequency dropouts in Lame

Reply #8 –
Here at HA.org we usually do not evaluate audio codecs by comparing spectral graphs. 
Apart from that, you should not expect perceptual transparancy at 128 kbps CBR

occasional high-frequency dropouts in Lame

Reply #9 –
Quote from: Hanky on
Here at HA.org we usually do not evaluate audio codecs by comparing spectral graphs. 
Apart from that, you should not expect perceptual transparancy at 128 kbps CBR

I know. I said that it sound really nasty, like a "very short burst of underwater sound"  or I dont know how to better express it.
As I posted in the Problem samples discussion, adding -q 4 to the command line fixes it.

J.M.