• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!

Review and Measurements of Benchmark AHB2 Amp

Sal1950

Grand Contributor
The Chicago Crusher
Forum Donor
Joined
Mar 1, 2016
Messages
14,073
Likes
16,609
Location
Central Fl
It took me about 6 months to get trained to hear the smallest amount of lossy compression artifacts.
Quick OT question, after the 6 months (or lifetime) of training, what codec and speed in a lossy format have you found transparent? Or can you detect all lossy compression artifacts? Not trying to put you on the spot, just interested on your opinion.
 

Blumlein 88

Grand Contributor
Forum Donor
Joined
Feb 23, 2016
Messages
20,524
Likes
37,057
I just read the paper. It doesn't say much about sighted listening being any good. It does start with this telling graph which I have mentioned countless times:

View attachment 28294

Out of megabits/sec of of audio data, the brain only captures a few bits/second. No way, no how does the brain capture for the long term minute details in music. You would go crazy in real life if you had to remember such nuance in my voice as we were talking for example. Or the background noise in your home or environment.

So anyone who says they listened to something for weeks, then switched to something else for weeks, is just wrong.

As to the thesis of the paper, the conclusion does it in: that most of what they are talking about is good practices summarized in ITU BS1116, "Methods for the subjective assessment of small impairments in audio systems." That is, familiarity and training matters when it comes to testing.

And yes, "slow learning" is super important. But it has another name: training. It took me about 6 months to get trained to hear the smallest amount of lossy compression artifacts. This does NOT at all say that when conducting listening tests, I need 6 months, or even 6 minutes to hear differences. Indeed, switching times over 1 second severely limit my ability to detect small artifacts.

The long process for hearing small artifacts may be needed to find the right artifacts. Sometimes it takes me half of hour of comparisons to figure out what to focus on. But once there, the actual comparisons occur in milliseconds, not "long term."

And again, there is nothing in there about sighted listening being more correct. Yes, you can do the training sighted, but the final test better be blind to mask other factors getting into the test.

Yes, other work indicates total perceptual bandwidth of our 5 senses is about 11 or 12 mbps. About 10 mbps of that is for vision. The hearing mechanism supplies our brains with around 90 kbps to 110 kbps (various methods reach different conclusions). We know from the difference in echoic and longer term memory the brain doesn't keep even that much long term.

All of that makes it hard to square it with the brain working at 40 bps. If you read this book:
https://www.amazon.com/Reading-Brain-New-Science-Read/dp/0143118056

It will make it clear how that happens. Though about reading and mostly vision it will show you how pattern matching on multiple levels works all this out to not overload the 40 bps processor which is our conscious brain. Of course 40-50 bps is for only the conscious work of the brain. The brain runs at 200 hz roughly, but is a massively parallel processor. Much of it is not conscious. The brain's total processing is 400 gbps. With only 40-50 bps conscious.

This is as good a short description of the book as you'll find:
"Brings together the cognitive, the cultural, and the neurological in an elegant, compelling narrative. A revelatory work."
-Oliver Sacks, M.D.
 

Blumlein 88

Grand Contributor
Forum Donor
Joined
Feb 23, 2016
Messages
20,524
Likes
37,057
Quick OT question, after the 6 months (or lifetime) of training, what codec and speed in a lossy format have you found transparent? Or can you detect all lossy compression artifacts? Not trying to put you on the spot, just interested on your opinion.
I'll bet he can hear the soon to be announced proprietary multi-lossless FLAC. You encode a file into it, and the size of the file grows by 400%, and yet will decode into the same original bits. But due to the added density of bits they are better quality bits. 1's are more purely 1. And zeroes are more purely nothing. I think it will be called MQALAC.
 

Rusty Shackleford

Active Member
Joined
May 16, 2018
Messages
255
Likes
550
And yes, "slow learning" is super important. But it has another name: training. It took me about 6 months to get trained to hear the smallest amount of lossy compression artifacts.

Could you explain more about this training?
 

digitalfrost

Major Contributor
Joined
Jul 22, 2018
Messages
1,521
Likes
3,086
Location
Palatinate, Germany
It's pretty warm here. atm almost 30°C in the room. AHB2 is warm to touch, but not uncomfortably hot. Definitely warmer than my hand. I tought I'd take a look in the manual about operating environment.

Environmental
0° C to 50° C (32° F to 122° F) Ambient Temperature
Up to 80% Humidity - Non-condensing

Are you kidding me? I run hardware in data centers, these are insane numbers. They're basically saying I can run this thing in the jungle? (Asking for a friend)
 
OP
amirm

amirm

Founder/Admin
Staff Member
CFO (Chief Fun Officer)
Joined
Feb 13, 2016
Messages
44,376
Likes
234,556
Location
Seattle Area
Quick OT question, after the 6 months (or lifetime) of training, what codec and speed in a lossy format have you found transparent? Or can you detect all lossy compression artifacts? Not trying to put you on the spot, just interested on your opinion.
AAC is the most transparent codec. If I spend enough time, I can usually tell it apart from the original but may take me a long time to find the right spot in the song that does it. I am confident for anyone else, it is fully transparent above 256 kbps.
 

reza

Active Member
Joined
Mar 27, 2019
Messages
110
Likes
131
I just read the paper. It doesn't say much about sighted listening being any good. It does start with this telling graph which I have mentioned countless times:

View attachment 28294

Out of megabits/sec of of audio data, the brain only captures a few bits/second. No way, no how does the brain capture for the long term minute details in music. You would go crazy in real life if you had to remember such nuance in my voice as we were talking for example. Or the background noise in your home or environment.

So anyone who says they listened to something for weeks, then switched to something else for weeks, is just wrong.

As to the thesis of the paper, the conclusion does it in: that most of what they are talking about is good practices summarized in ITU BS1116, "Methods for the subjective assessment of small impairments in audio systems." That is, familiarity and training matters when it comes to testing.

And yes, "slow learning" is super important. But it has another name: training. It took me about 6 months to get trained to hear the smallest amount of lossy compression artifacts. This does NOT at all say that when conducting listening tests, I need 6 months, or even 6 minutes to hear differences. Indeed, switching times over 1 second severely limit my ability to detect small artifacts.

The long process for hearing small artifacts may be needed to find the right artifacts. Sometimes it takes me half of hour of comparisons to figure out what to focus on. But once there, the actual comparisons occur in milliseconds, not "long term."

And again, there is nothing in there about sighted listening being more correct. Yes, you can do the training sighted, but the final test better be blind to mask other factors getting into the test.

Explicit memory is a misleading measure of what you take in. There is no way you can explicitly remember the details of songs that you know and have listened to many times. But if I play for you an altered version of the song, you'll immediately notice that something is off.

At any rate, bits and cycles and other computer metaphors have little relevance to the wet machine, unless by bit they refer to the information content of the music. But that's another can of worms.

The mind builds its representations using a mechanism known as hierarchic abstraction. A tiny bit of information in a higher level corresponds to a massive amount of information at the lower levels. Conscious perception takes place at the highest level of that hierarchy, sensation at the lowest.
 
OP
amirm

amirm

Founder/Admin
Staff Member
CFO (Chief Fun Officer)
Joined
Feb 13, 2016
Messages
44,376
Likes
234,556
Location
Seattle Area
Could you explain more about this training?
Sure. It had three parts:

1. Understanding the signal processing so that you know what is hard to encode, and what is not.

2. Content that is very revealing of compression artifacts.

3. Listening at high compression ratios first to learn the artifacts and then gradually going higher and higher bit rates.

Despite having been an audiophile for three decades, I had little ability hear compression artifacts when I started. When I finished, no one could touch me. :)

An interesting side-effect was that I learned how to truly listen and find corner cases in fidelity. As such, I have very high success rate in many other public blind tests. It is like people who speak multiple languages and how much easier it is for them to learn new ones.
 

Biblob

Addicted to Fun and Learning
Joined
Sep 13, 2018
Messages
633
Likes
600
AAC is the most transparent codec. If I spend enough time, I can usually tell it apart from the original but may take me a long time to find the right spot in the song that does it. I am confident for anyone else, it is fully transparent above 256 kbps.
How would you rate OGG Vorbis used by Spotify? Comparably? Or the difference too small to be really noteworthy for an untrained listener as me.
 

Wombat

Master Contributor
Joined
Nov 5, 2017
Messages
6,722
Likes
6,459
Location
Australia
Concluding about something one hasn’t read, is easy. Why not read before commenting?

The paper draws upon a wider range of scientific disciplines across decades than is usual in AES papers, and quotes 81 references.

The main theme of the paper is our general model of hearing and perception. One problem that may not be fully understood by today’s general model of human hearing is listener fatigue. So the paper is in the spirit of laying out an agenda for future research rather than having all the answers right now.

(Another example is our understanding of transparent audio codecs, which has evolved over time).

Science is about asking questions too. Some of those questions cannot be readily answered today. This is very frustrating for some, while it’s an inspiration for others.


I did not comment on the content of the paper. I could not access it. I did ask if the term 'slow listening' that you extracted from the paper was validated. I think that was a fair question.

Thank you Amirm for explaining the term and its application.
 
OP
amirm

amirm

Founder/Admin
Staff Member
CFO (Chief Fun Officer)
Joined
Feb 13, 2016
Messages
44,376
Likes
234,556
Location
Seattle Area
How would you rate OGG Vorbis used by Spotify? Comparably? Or the difference too small to be really noteworthy for an untrained listener as me.
I have no recent data. At the time Vorbis came out, it was not a good codec. It sounded better to people because it defaulted to variable bit rate encoding, rather than fixed rate.
 

Jim777

Active Member
Forum Donor
Joined
May 28, 2019
Messages
124
Likes
203
Location
Greater Boston
I've done my share of blind listening tests (i.e. MUSHRA) so I can echo what @amirm said. I'll just add that the quality of AAC will depend on the encoder implementation as well, since there is no mandatory encoder implementation. The idea is to only define the bitstream and the decoder to allow people to improve the codec even after it's standardized. If you have a better psycho-acoustic model, you'll do a better job. But in any case, 256 kbps is probably transparent ~99.9% of the time (it helps if you're trained to detect AAC artefacts, find the right material and do very critical listening).
 

Sal1950

Grand Contributor
The Chicago Crusher
Forum Donor
Joined
Mar 1, 2016
Messages
14,073
Likes
16,609
Location
Central Fl
This thread is hopelessly off the rails, again, and I can see why John_S bailed out of it.
I agree John, but how much is there left to say about a amp that is as close to perfect as can be designed and measured in todays world?
 
OP
amirm

amirm

Founder/Admin
Staff Member
CFO (Chief Fun Officer)
Joined
Feb 13, 2016
Messages
44,376
Likes
234,556
Location
Seattle Area
I agree John, but how much is there left to say about a amp that is as close to perfect as can be designed and measured in todays world?
That's why I am letting it be. John did find his way through the mess to answer a few questions just yesterday so we are good there.
 

restorer-john

Grand Contributor
Joined
Mar 1, 2018
Messages
12,588
Likes
38,291
Location
Gold Coast, Queensland, Australia
I agree John, but how much is there left to say about a amp that is as close to perfect as can be designed and measured in todays world?

I was going to say something similar, but didn't want to appear as a fanboy. Realistically, there is little to complain about and a whole lot to celebrate in terms of performance of the AHB-2.
 

Sal1950

Grand Contributor
The Chicago Crusher
Forum Donor
Joined
Mar 1, 2016
Messages
14,073
Likes
16,609
Location
Central Fl
I was going to say something similar, but didn't want to appear as a fanboy. Realistically, there is little to complain about and a whole lot to celebrate in terms of performance of the AHB-2.
The only thing he could really do better is offer the same product for 50% less. LOL
 

RayDunzl

Grand Contributor
Central Scrutinizer
Joined
Mar 9, 2016
Messages
13,204
Likes
16,986
Location
Riverview FL
The only thing he could really do better is offer the same product for 50% less. LOL

The stereo version is 50% less than the bridged-mono version for a two channel system...
 

Sal1950

Grand Contributor
The Chicago Crusher
Forum Donor
Joined
Mar 1, 2016
Messages
14,073
Likes
16,609
Location
Central Fl
The stereo version is 50% less than the bridged-mono version for a two channel system...
That's backwards thinking.
It's a $3k stereo amp that can be bridged to $6k mono blocks.
Don't go turning subjectivist on me. LOL
 
Top Bottom