Can you tell MP3 from CD? Are you sure?

[Rob Babcock]

Check out this interesting article from MaximumPC magazine. Granted, it's not a purely scientific test and has a few problems (eg- it was single blind, not double and the soundcard, while tolerable, is not really very high end) but is a fun read.

Has anyone here (besides obsessive-compulsive DBT'ers like Wmax ) tried to do an experiment like this?

 

[mike c]

I wish these guys had included the all famous 128kbps bitrate

i haven't been ripping to mp3's for some time now ... and I'm not familiar with this VBR thing ... during my time it was a fixed ceiling bit rate ...

hmmm, what do you guys use to rip mp3's nowadays that have this VBR thing? I'm going to build me a new mp3 CD/ipod collection.

 

[MDS]

I've never been involved in a formal DBT (though I wholeheartedly agree with the idea) but I can say from personal experience that it is possible to detect subtle (emphasis on subtle) differences between bit rates.

The main example that comes to mind is Triumph - Ordinary Man, one of my favorite Triumph songs and one that I am rather familiar with. It has a lot of acoustic guitar in the beginning and when encoded at 160 kbps or lower it just doesn't sound 'right' to me. At 192 kbps or higher it sounds just fine and as I would expect, having heard it hundreds of times. My own experiment was to encode it at 256 kbps and 320 kbps and to my ears they sound no different than the 192 kbps version. For what it's worth I encode all my MP3s at 192 kbps.

Mike C: 'Variable bit rate' seeks to allocate more bits to those passages that are more demanding than other parts with the idea being that it is a waste to use 192 kbps (or whatever) for passages that can be fully captured without audible loss using a lower bit rate and instead allocating extra bits for the more demanding passages that cannot be fully captured with the target bitrate. The average will still be the target bitrate but with some sections using more and some using less - in contrast with 'constant bitrate' where every second uses the target bitrate. It should be obvious that encoding one second of silence as 192K zero samples is not the most efficient use of a constant number of bits and that is what VBR overcomes.

Does VBR 'sound better' than CBR - as with anything else, sometimes yes and sometimes no.

Because I also edit audio rippped from CD at times, I use Sound Forge to create my MP3s and although it offers the option of VBR vs CBR, I don't bother - I've found that 192 kbps CBR works great for the vast majority of my music collection.

 

[mike c]

thanks MDS, I guess I'll stick with 192kbps CBR since that's going to be more than overkill

 

[Rob Babcock]

I don't use MP3 much but I'm a big fan of VBR when I do. Note that I normally use bitrates of over 300 kbps with VBR- much of the file will use lower rates so the file size isn't excessive for my purposes.

 

[Yamahaluver]

I have been using LAME since the day it became popular on the net, I use VBR new with the r3mix settings which include JS, quality at 2 and bit rate between 92-320kbps, comes out real good and good size as well, perfect for my car stereo, yes my ears can still discern between original and mp3 although the quality of encoding is going up considerably and I don't know for how long my ears with age would be able to do so, so far I don't use mp3 at home, only for car.

 

[Pwner_2130]

Mp3 vs. Cd

By the way, i can tell the difference between an mp3 and a cd but only when the mp3 is encoded at a bit rate lower than 192.

I listened to the intro of Bob Dylan's "man in the long black coat" in 320 mp3 and then on cd, and there was virtually no difference.

 

[WmAx]

Check out this interesting article from MaximumPC magazine. Granted, it's not a purely scientific test and has a few problems (eg- it was single blind, not double and the soundcard, while tolerable, is not really very high end) but is a fun read.

Has anyone here (besides obsessive-compulsive DBT'ers like Wmax ) tried to do an experiment like this?

Pathetic attempt at a test.

For well founded answers, visit hydrogenaudio.org, where a lot of the real perceptual DBT for loss audio encoders originates and is discussed regularly.

As for the soundcard? That sound card measures superbly. But it's merely tolerable to you? This impression, I suspect, is based upon the forum discussions by ill-founded, ignorant opinions of select audiophiles, not by factual data. Now, I don't know anything about it's internal headphone amplifier, and sometimes when using a headphone load, things can change. But in the past, Creative cards usually use good headphone amplifier circuits, at least for moderately voltage sensitive headphones.

-Chris

 

[emmaco]

Can you tell the difference?

Easily. Really depends on the quality of the reproduction system and one's listening acuity. As an example, few folks have their PCs hooked up to electrostatic speakers in a controlled listening environment. The vast majority of listeners are making their judgement on ear buds in a noisy car with a manufacturing cost well under a quarter. The question has always been "what's good enough for most folks in a real world application?"

The Frauenhofer folks did a great job with the algorithm, but have never sold their performance as identical- merely "good enough for most applications", i.e. mobile audio. Your standard 128k fixed bit rate mp3 throws out what- 75% of the audio bits? I suppose one could argue a VBR approaching a CD's 44.1 sampling could be equivalent- but that's somewhat defeating the purpose of .mp3, isn't it? The days of a personal player holding 25 songs have long since left us.

One other commercial manifestation of the Institute can be heard at www.iosono-sound.com, intended for localizing sound in large cinemas. I heard the demonstration system over at the Mouse a year or so ago and it's very impressive as well- hundreds of speakers and amplifier channels.

 

[MDS]

Easily. Really depends on the quality of the reproduction system and one's listening acuity.

That is the typical audiophile type response - your equipment isn't 'resolving' enough or your hearing isn't is good as mine. Yet the claims never hold up under rigorous testing.

Your standard 128k fixed bit rate mp3 throws out what- 75% of the audio bits? I suppose one could argue a VBR approaching a CD's 44.1 sampling could be equivalent- but that's somewhat defeating the purpose of .mp3, isn't it?

Bit rate isn't everything and it doesn't quite work the way that statement would imply. The premise behind all 'perceptual coding' algorithms is that much of the music won't be perceived because it is masked by surrounding tones. The algorithm doesn't just take the 1.4 Mbps from an uncompressed CD and chop off 1.3 million bits of each second. The bits that remain are not a simple subset of the originals.

The algorithm chops up each second of audio into 32 or more sub-bands and then analyzes them to determine where masking would occur. It discards the bits that contain sound content that it deems to be masked. Lower bit rates, like 128K, also chop off all of the high frequency content above a certain point because it is well known that most people over 21 cannot hear beyond about 17 kHz.

 

[pikers]

Of course, the main difference between CD and 128/192 is high-frequency roll-off. The older we get, the less important that becomes.

In a blind test, it would be fascinating to see if any male over 30 could reliably discern the difference. I've been doing DBTs for years, and probably couldn't if the material was of sufficient quality.