Ringing Redux

Listening Tests by bAdDuDeX, the Digital Ear, Wombat, and Hans Heijden; Notes by ff123

 

Addenda
(4/9/01): It is possible to virtually eliminate narrow spiking in Lame 3.88b with this sample using -q0 in combination with --nspsytune.
(4/14/01): Using --nspsytune with VBR and athtype 2 produces files which sound ok, but could be extremely huge.

bAdDuDeX has performed some listening tests of a sample in which he finds ringing by Lame 3.88b to be especially obvious. I believe that the artifact called "ringing" in this particular sample is caused by loud, high-frequency, and relatively long-lasting sounds above 16 kHz which don't get properly encoded for their full duration, and thus create an artifact when they suddenly disappear.

The file he used was a clip from "Bloodline" by Slayer. You can download the clip (ringing.pac) from my Audio Samples Page. The file has been losslessly compressed using the LPAC format.

Summary Results
Listener Ranking, best to worst
bAdDuDeX 1. mpegplus 1.7.9 -standard
2. Mp3Enc31 -qual 9 -br 192000
3. Lame 3.88b --r3mix -b128 --lowpass 16
4. Lame 3.88b -h --abr 192 --nspsytune --athtype 2
5a. Lame 3.88b -h -b256
5b. Lame 3.88b --r3mix -b128
5c. Lame 3.88b -h --abr 192

Not ranked, but it sounds fine: Lame 3.88b -V1 -mj --nspsytune --athtype 2 -b128 --lowpass 19.5
Wombat 1. Lame 3.88b -h -b256
2. Lame 3.88b --r3mix -b128 --lowpass 16
3a. Lame 3.88b --r3mix -b128
3b. Lame 3.88b -h --abr 192 --nspsytune --athtype 2

Not rated: mpegplus 1.7.9 -standard
Not rated: Mp3Enc31 -qual 9 -br 192000
Not rated: Lame 3.88b -h --abr 192
Digital Ear 1. (quality rating = 1) - Lame 3.88b -h -b256
2. (quality rating = 4) - mpegplus 1.7.9 -standard
3. (quality rating = 5) - Lame 3.88b --r3mix -b128
4. (quality rating = 8) - Lame 3.88b --r3mix -b128 --lowpass 16
5. (quality rating = 10) - Lame 3.88b -h --abr 192 --nspsytune --athtype 2

Not rated: Mp3Enc31 -qual 9 -br 192000
Not rated: Lame 3.88b -h --abr 192

 

Encoder/Settings Spectral View Listening comments by bAdDuDeX; other notes by ff123
Original sample No ringing

There is nothing wrong with bAdDuDeX's soundcard or his headphones (Grado SR 225).
MPEGPlus 1.7.9, -standard No ringing

mpegplus sets max bandwidth no higher than 17.5 to 18 kHz in this particlar portion of the sample. Most of the high-frequency energy within the cymbal pulses are captured, and the peak gradually rises up out of the background sounds instead of poking up suddenly. There are few or no dropouts in the region from 10 to 16 kHz.
Lame 3.88b -h --abr 192 Ringing. The ringing is just as loud as any other artifact (to me). It's not like I have to listen carefully or blast the volume in order to hear it. In fact, I tried using some old headphones I had and could still hear it.

max bandwidth goes up to about 19 to 19.5 kHz. However, although the initial portion of the cymbal pulses are captured, a significant portion of the pulses are cut off. I believe that these partially-encoded high-frequency pulses are one of two types of artifacts described as "ringing."

Also, from 10 to 16 kHz, there are dropouts. I believe these can also be heard as "ringing," for example, if they are present in music encoded with mp3 at 128 kbs, where the default lowpass filter is usually 16 kHz.
Lame 3.88b -h -b256 Rings

The dropouts between 10 and 16 kHz have filled in. Therefore, any remaining ringing is probably not caused by problems in this region. Peaks still appear suddenly from nowhere (compare with mpegplus), and the high-frequency energy is still only partially encoded.
Lame 3.88b --r3mix -b128 Rings

No dropouts to speak of between 10 and 16 kHz. But the characteristic Lame narrow spiking above 16 kHz remains, and in fact is highlighted.
Lame 3.88b --r3mix -b128 --lowpass 16 That sounds much better [than regular --r3mix -b128]

Another indication that the narrow spiking above 16 kHz is causing audible problems.
Lame 3.88b -h --abr 192 --nspsytune --athtype 2 There's not too much ringing. More than MP3Enc but less than LAME with -h -b256.

Right in line with what the spectral view looks like. Not much, if anything, in the way of dropouts below 16 kHz either.
Mp3Enc31 -qual 9 -br 192000 No ringing

MP3Enc never allows any energy to be encoded above 16 kHz. There are some dropouts between 9 and 16 kHz, and one low frequency dropout at 4.92 seconds (see my page: Dropouts with MP3Enc/Alternate Codecs). Some people say that the sound of MP3Enc is muffled, and this spectral plot is certainly consistent with that assessment. But bAdDuDeX much prefers this type of sound to the ringing distortion he hears in Lame without nspsytune.
Lame 3.88b -V1 -mj --nspsytune --athtype 2 -b128 --lowpass 19.5 Sounds fine

Spikes are wider and now capture most of the energy in the cymbals.
FhG/Radium ACM codec, CBR 192, regular stereo No listening evaluation

Again, this FhG codec never allows anything above 16 kHz to appear for this sample.
FhG FastEnc CBR 192 (using Cool Edit Pro with MP3 ME plugin) No listening evaluation

This FhG codec also does not allow anything above 16 kHz to be encoded. There are a significant number of dropouts between 10 and 16 kHz, though. bAdDuDeX has said the only codecs he's heard ringing on are Blade, Lame (without nspsytune), and FastEnc, although FastEnc seems to be least problematic in this area, and Blade seems to be the worst.
Lame 3.88b -q0 --abr 192 --nspsytune --athtype 2 No listening evaluation

Almost all traces of narrow spiking in the sample have been eliminated.

Listening test by the Digital Ear:

Looks like the d-ear either doesn't hear the ringing, or other artifacts skew the results as kennedyb4 suggests. Here is how the d-ear ranked the five with 1 being best and 10 being worst:

1 - Lame 3.88b -h -b256
4 - mpegplus 1.7.9 -standard
5 - Lame 3.88b --r3mix -b128
8 - Lame 3.88b --r3mix -b128 --lowpass 16
10 - Lame 3.88b -h --abr 192 --nspsytune --athtype 2

Listening test by Wombat:

ff123 wrote:

> encode with Lame 3.88b with the following settings:
>
> 1) Lame 3.88b --r3mix -b128 --lowpass 16
> 2) Lame 3.88b -h --abr 192 --nspsytune --athtype 2
> 3) Lame 3.88b --r3mix -b128
> 4) Lame 3.88b -h -b256
>
> How would you rank these from best-sounding to worst-sounding, and why?

First:
I donīt have exceptional hearing. I had the luck to build speaker systems together with very good people who are still in the business with their work. This way i educated my hearing to be more sensitive for overall tonality and pictureness of recordings than others maybe. I canīt hear up to 20khz. Also i am not sensitive on preecho or small artifacts. There are sure a lot of better with their headphones on. I donīt like listening under headphones. I want to see and feel the stage in front of me.

Here is my rating.

Best -> Worst

4:) Hard to tell, maybe the singer shouts more harmless ;)

1:) Not the dynamic as the original but the middlepart is clearly more transparent than with 2 and 3

I am not so sure about 2 and 3, 3 is more "rough" but nearer to the original and 4 is not this "rough" but with over all tonality faults.

rough -> must be the "ringing"

Listening test by Hans Heijden:

I'm pretty sure now what you hear. These are bursts at 16 kHz only. The original wav has tones at this frequency. Lame usually does not encode these, but sometimes it does momentarily, even increasing it a few dB in strength.

It's all easy to hear with resampling to 32 kHz. At normal 44,1 kHz playback, I hear it once, at 4,3 sec in the left channel of the 256 CBR sample. Had to reverse my headphones, because my right ear is a little better.

Your high-frequency hearing must be way above average Baddudex!

 

Return to ff123's home page