I have just conducted a quick blind listening test on a few reggae (Bob Marley) MP3 tracks encoded with --alt-preset extreme and was able to recognize the MAD decoder 100% of the time!

For decoding I use Winamp 2.80 with the built-in decoder, and CoolPlayer 2.05 which uses MAD, and listened on my Koss Porta Pro headphones (which I consider of very good sound quality).

I found the sound of the MAD decoder far superior to the Nullsoft FhG one! These are my observations:


* FhG produced some heavy basses. Sound was well rounded and pleasing, but muddy and unfaithful to the original. In contras, the MAD decoder basses were tight and slightly aggressive but very faithful to the original sound

* FhG mid range was over emphasised. It is especially easy to spot when listening to human voice, with both samples side by side: the FhG version sounds very poor and hardly recognisable when comparing with the MAD version.

* High frequencies were completely muffled by FhG. I could not believe the loss in details, especially when comparing with MAD.


In conclusion, I found the MAD sound very faithful to the original with a highly detailed sound. The FhG sound appeared heavily distorted and muffled next to it, as it some filters and equalisation had been applied. Indeed I had to check a few times if I had not any DSP filter or EQ turned on! The only advantage of FhG is due to the softness and loss of details, I would think it would cause less hear fatigue when listening through headphones; I know which one I would rather use though

Conclusion2: I have got ears I thought my hearing was below average, but after reading so many people writing there is no listenable difference between MAD and the latest Nullsoft decoders, I now know many forum members have a worse hearing than me


Quick awful pun: you would be mad not to use MAD!

The FhG and MAD decoder will only differ in the least significant bit.

That difference would not produce anything like what you describe, so you must have screwed up somewhere else.

--
GCP

I have just finished running another test where I compared the output of the winamp MAD plugin and coolplayer MAD decoder: I could not distinguish the 2...

The only thing I changed from the previous test, was using the MAD plugin instead of the Nullsoft FhG one!

I agree that if the output differs only by 1bit, the least significant one, it should be very difficult to hear any difference at all. However the bit comparison was done quite a while ago and since then there has been many versions of both encoders. I would like to see a bit comparison using both latest versions.

The comparison was done vs the reference FhG decoder. The one that is currently used in WinAMP is also a FhG reference decoder.

There will not be any differences bigger than the LSB, unless your test setup is screwed (as I said before), so I suggest you take a closer look to what is wrong.

--
GCP

What soundcard/settings were you using?

--
GCP

If you want your claims to be taken seriously, make sure your test file isn't clipping and decode it to wav file for ABX testing. You can use mp3gain and Winamp's disk writer for the tasks.

I just did a test for myself using MAD 0.14.2b and FhG 2.8a and the RMS power of the inverse mix between FhG and MAD was -90dB...as expected.

The differences were mostly > 16kHz, which suggests they are simply the noise shaping dither.

None of this would cause any difference remotely to what you describe.

--
GCP

Hi arman68,

Just to let you know that you are not the only one to hear the differences.

It becomes very obvious when you configure the MAD plugin to 24-bits - no comparison whatsoever - at least on my system (SB Audigy, 100 Watt/channel home stereo, with DIY speaker/interconnect cables). And yes, I know that the audigy sound drivers don't propogate the 24-bits all the way through to the DAC, but the sound is still superb despite this (and a heck of a lot better sounding than the SB live card I used to have).

Incidentally, if anyone else has a home stereo attached to their computer and wants to improve their sound quality very significantly, take a look at this link:

http://www.venhaus1.com/diycatfivecables.html

and be sure to read some of the testamonials on this link:

http://www.venhaus1.com/webreviews.html

I built modified (simplified) versions of both of Chris VenHaus's cables (the speaker and the interconnect ones) and the difference was amazing - just like the testamonials. The best way to describe it is that the speaker system became more "headphone-like".


- Spartane.

I will say *again*, if the difference is very obvious, your setup is screwed. Magic speaker cables wont solve that.

To put things into perspective, you're basically claiming the noise floor of a CD is totally insufficient for your listening needs.

--
GCP

Could it be someone forgot to disable WinAMP's built in equalizer when using the default plugin? This is a common mistake, and could easily result in the described differences.

Edit -- the first posting on this thread suggests the WinAMP equalizer was NOT used, but thought I'd check anyway. I actually use the default WinAMP EQ with my crummy PC speakers (they can't reproduce high-end detail anyway)... and have forgotten to disable it before when listening with good quality headphones (uggh).

I just got finished wasting time on alt.music.mp3 "debating" this issue. I agree with Garf: if you find the difference "obvious" between MAD (either 16 bit or 24 bit) and FhG, my first reaction is that you must have done something wrong. It's not personal, and the person I took issue with was instantly insulted, and things degraded into a flame-fest. But hearing the difference between dither and truncation at 16 bits (or between 24 bits and 16 bits) are extraordinary claims. And extraordinary claims demand extraordinary proof.

If you haven't done so already, please decode to WAV first, and verify that your two WAV files are different only by 1 bit. If you like, I can do this for you. You can anonymously ftp your files (preferably losslessly compressed) to:

ff123.net
userid:
use anything for a password
go to the "incoming" directory

Use PC-ABX to conduct your blind tests. If the difference is as obvious as you both claim, you should have no trouble, getting 5 out of 5 trials immediately correct. See my MAD challenge page at:

http://ff123.net/madchallenge.html

ff123

I can definitely say the MAD decoder sounds better, more true to the original, but this is probably due to the clipping attenuation.

Although I do have to agree with arman, the mid-range in the FhG decoder is glarry and over-emphasized.. that's just what I've noticed. This could be due to the clipping attenuation, but who knows, *shrug*, MAD sounds the closest to the original to me and that's what matters

Although being purely psychological...

You're probably right.. the same rule applies to a lot, especially cars. I can't tell you how many times someone posted something on a car message board exclaiming the power gains they achieved by doing a modification because they "felt" the difference, when in reality, there's no way they could have gained any.

Oh well, weird how the mind works like that

It's a rather strange subject. What's "psychological" can be just as "real" as anything else -- for example, taking placebo painkillers (sugar tablets) can kill pain just as well as narcotics, in some situations. So does that mean the placebo is "real" because the mind makes it so? After all, there was no pain.

And does it mean the difference with MAD is "real" because it *sounds* real when listening (even though standardized blind tests may fail)?

Questions for a philosopher maybe... not audio testers/programmers...

I actually did the test because of so many people claiming the FhG decoder was as good as MAD and they could not hear any difference. I went into it believing it, so I do not think it is psychological. Or then with the sound differences I heard I am seriously screwed

The samples I used did not clip, MAD clipping attenuation did not come into effect.

Decoding was on 16bits 44.1kHz in each case.

As I said before, I did not expect to hear any difference and I was shocked by the results. I agree that there must be something flawed as only 1 bit could not account for such a huge sound difference. Unless it is not the least significant bit...

The only thing that could be flawed is I did not decode to WAV, but listened straight through the players (I asked someone to play the samples without telling me). Could this account for the sound difference?

Likely, some player is doing some processing to the sound that changes things. e.g. WinAMP will still mutilate the output if the equalizer settings are all flat but it is still enabled.

--
GCP

This is why I asked to take a look at the wav files. Just use Winamp's wave write plugin to write the output to wav instead of routing the output to your speakers. Then upload the wavs (or preferably losslessly compressed wavs) to my ftp account. A 10 second excerpt should be fine. The alignment doesn't have to be exact; I can trim the beginning and end points.

ff123

I see deaf people.

And how did you calibrate the volumes?

That's a good point. If one uses two different applications with two different volume controls, there's no guarantee they'll have the same volume. The one which is louder will almost always sound better. One more reason to use PC-ABX to perform the blind tests.

ff123

Under what circumstances? I can't reproduce this.

With every control at 0dB the winamp EQ does absolutely nothing, even when enabled.


David.

P.S. the test would be fairer if you were using the same player for both decoders. Winamp can use either decoder, so why not try this? There may be some additional audio processing built in to the other player that isn't documented.

You are right. Either I remember this wrong or it has changed.

--
GCP

What about the decoding engine used in Apollo (http://www.hut.fi/~hylinen/apollo/). I find the low frequencies not that natural, does anyone agree in this? Are there some tests online concerning Apollo's decoder?

Does PC-ABX make sure the volumes of the files to compare are equal?
What would be a safe way to adjust the volumes without introducing more "processing".

I know, just questions that jumped at me
--
Ge Someone

No



If the decoders work, they will be equal to start with.

--
GCP

It's more a matter of PC-ABX not changing the volume, which I know Winamp can do, at least. So two files decoded to WAV will be played exactly as is in PC-ABX, with only differences at the 1-bit level.

ff123

I just did a quick listening test, not blind, and this time the output of FhG and MAD sounded very similar. I think I have found the culprit:

In Winamp I was using the Direct Sound ouput plugin from http://www.blorp.com/~peter/zips/ which caused all the problems.

After switching to the normal WaveOut plugin the sound went back to 'normal'. When I have some time, I will decode the files to wav and do a proper listening test.

The reason I was using the Direct Sound plugin was because some people in this forum had complained about the poor quality of the SqrSoft Crossfader output plugin, and the suggestions were to use the one above... I will test the output of the SqrSoftplugin as well.

Save time and rather do the other tests you wanted to perform.

My experience (at least with Win98SE) is that Directsound has *horrible* sound quality, and is useful only in games, or other cases where multiple streams in software are necessary. Too bad there doesn't seem to be any way to avoid it with Windows Media Player... it tends to play too quiet, and really mangle sound quality (at least with my SB PCI128) -- I think because of lousy software bitrate conversions & other filtering being done.

Thanks Microsoft... for another mediocre attempt at "solving" problems. Mediocrity has to be Microsoft's motto...

IMO:
With WinNT: DirectSound is great and WaveOut sucks
With Win9x: DirectSound sucks and Waveout is great.

I finished redoing blind listening tests. I did not decode to WAV though, because I wanted to test the effect of the different output plugins as well. Here are the results:

* I could not differentiate the output from the MAD and FhG decoders!

However, I found that the Direct Sound winamp output plugin degraded the sound: the version I used im my previous test caused my initial wrong conclusion (see first post in this thread); the one I used now - Direct Sound output 1.4 beta 6 - did not have such a bad effect on the sound quality, but the playback volume was much lower than other plugins.

Output from winamp plugins waveOutput v2.0 RC12, SQrSoft Advanced Crossfading and Coolplayer 2.05 were indistiguishable.

CPU usage for all combinations of the above was 0% on my PIII 1GHz laptop running XP.


I guess the only advantage of the MAD decoder now is the ability to playback at 24bits resolution which should give a better sound if your soundcard supports it. Does anyone knows which output plugin supports 24bits? From looking at the settings from all of them, I would think only SqrSoft ACF does.

Things about DirectSound (that everyone should know but...)

A) There were some Creative Labs drivers (maybe other soundcards) that used to play in 22Khz 8bits with directsound whe using the primary buffers. This was very evident, and I assume it is not the case here

B) For some (strange) reason, some implementations of Directsound CAN filter the sound out at 16Khz or so ( resampling to 32Khz? ). I am the developer of a Tracker (Psycle) and the directsound output has this problem with some Creative cards too. The most funny part is that If I give the focus to winamp (even when not playing), then the "filter" dissapears. Funny huh? ( BTW: This happened under Win9x/Me)

C) Volume of Directsound: When you change the volume using Waveout, you are changing the volume of the soundcard output for the wave channel. When you change the volume using the directsound plugin, the volume is lowered by software by Directsound. So Top volume in directsound equals to the volume of the waveout you have been using before.

Funny, because the SQrSoft plugin is supposed to be horrid in terms of sound quality..

What exactly makes waveOut suck in NT? I run XP, and I'm using the recently released resampling waveOut and DS plugins, which one is "supposed" to sound better, or is there even a difference?

It's really horrid when resampling is involved.

I would believe it's as shitty as Nero's resampling routines.

Just to add a small bit to the discussion:

Did anyone else think it was unfair to use an 8-bit samples for the decoder comparison on the MAD (winamp plugin) page? While its 100% certain that MAD sounds better (less noisy) in this comparison, it doesn't seem logical to assume that this will be the case at higher bit-depths, as the page asserts. Maybe the Fhg enoder just is buggier with non-standard bit-depths samples, which wouldn't surprise me since how many people encode 8-bit material to mp3? I'd assume it would need to be upsampled to 16-bit to begin with, as well.

Anyway, the whole comparison there made me think they couldn't find any 16-bit stuff that actually sounded noticeably better with MAD.

In my period of casual testing, I haven't been able to notice a significant difference either (other than MAD breaking streaming stations). I'd imagine if you were blessed with a 24-bit card then *hopefully* the differences would start to emerge more obviously.

I know it isn't decent, but:

**KICK**

What about the sound quality if you had an SB Audigy and did the following?

MAD Decoder set to 24 bits
PP's DS output plugin set to resample at 48000 16bits?

Might that turn the extra bits into something useful/hearable?

- Agent 86

Well, theoretical it should result into better audio quality (if the resampler in the DirectSound plugin is better than that of the Audigy). I would set the MAD Decoder to 32bit, since I read elsewhere that the 24bit output of MAD has some flaws. The Audigy also eats 24bit and even 32bit of the output plugin, but I don't know if that would improve quality...

The Audigy is a good gamer's card, but I'm beginning to regret my purchase, since I'm not a gamer at all and audio quality & drivers (support) are horrible.