Amplifiers - A blind test or a dumb test?

[STC]

I made some recordings of different amplifiers where the loudness level was arbitrarily judged to be level matched using a SPL for the highest peak of my favorite track. I didn't use test tones.

 

What surprised me was the volume level varied very much between tracks and the difference in SQ was obvious when compared side by side of the recordings.

 

I then took a short sample of each recording below the clipping level and normalize them using Audacity. I expected the sound quality difference would be eliminated. Sadly, that wasn't the case.

 

These amps were measured and their FR were found to be flat. They were driven well below their maximum rating of 250W for a 89dB speakers. Their input sensitivity is well within the preamp DAC's max output.

 

What is causing the difference?

 

 

[17629v2]

"What is causing the difference?"

 

The design and parts quality.

[semente]

There is more to amplifier performance than plain old frequency response.

 

Check Stereophile or Soundstage for an idea of a comprehensive set of measurements.

 

R

[esldude]

Well Soundlabs have one of the more uneven impedance curves of any speakers you will find. Even small output impedance differences in the amps will cause audible frequency response changes. A quick listen and keeping in mind it is a youtube quality audio signal it sounds like at least one amp is distorting in the highs or possibly near oscillation. Depending on your Soundlabs the upper frequencies are seeing an impedance of merely one ohm possibly even less. Transformer coupled amps and Class D amps will both be very effected by this. The xfmr can interact with the reactive low impedance to have a response that nearly rings near 20 khz. Class D amps have output filters that allow response variations (and sometimes distortion) due to the low impedance in the treble region. Sometimes you get lucky. I have Wyred4Sound amps which alter the response just a bit more in the direction I want it. It would sound different than a good class AB amp however.

[esldude]

Here is the Soundlab A1 impedance curve. Most of the Soundlab panel speakers are more or less this curve with minor differences.

 

[STC]

There is more to amplifier performance than plain old frequency response.

 

Check Stereophile or Soundstage for an idea of a comprehensive set of measurements.

 

R

 

It is probably true but I am being just like a normal person who would judge sound quality by listening.

 

From the samples, I find it hard to pick one clear winner.

[STC]

Well Soundlabs have one of the more uneven impedance curves of any speakers you will find. Even small output impedance differences in the amps will cause audible frequency response changes. A quick listen and keeping in mind it is a youtube quality audio signal it sounds like at least one amp is distorting in the highs or possibly near oscillation. Depending on your Soundlabs the upper frequencies are seeing an impedance of merely one ohm possibly even less. Transformer coupled amps and Class D amps will both be very effected by this. The xfmr can interact with the reactive low impedance to have a response that nearly rings near 20 khz. Class D amps have output filters that allow response variations (and sometimes distortion) due to the low impedance in the treble region. Sometimes you get lucky. I have Wyred4Sound amps which alter the response just a bit more in the direction I want it. It would sound different than a good class AB amp however.

 

Thank you, Esldude. Would you mind telling the samples sounded distorted? Sample 5B did exhibit some distortion but that was microphone related.

 

Thanks.

[esldude]

There really is too much level difference. Looking at the RMS levels, 1A is around .8-1.0 db louder than sample 1B. 1C is close to A in the left channel and .5 or .6 db louder than A in the right channel. If you listen a few times you can hear this. I heard it from listening a few times, and then tried to measure it. I could hear there was a loudness difference and an image shift between 1A versus 1B and 1C. These things will sound like more than what they are and that isn't even seeing if there is a frequency response difference.

 

It would have been better to use a test tone and match 440 hz or 1 khz for level by measuring the voltage at the speaker inputs. If I had the original files, it might be possible to compensate for these factors well enough to work. With the conversion to MP3 quality over youtube doing so is iffy.

 

One reason normalizing might not work is very low frequency noise. Something not heard all that much happening at 60 hz and below could corrupt the normalizing. A distant rumble from a truck or lawnmower at some low frequency you wouldn't notice yet gets picked up in one recording and not another. You might try to EQ out everything below 200 hz, then normalizing, and then doing reverse EQ. The match might be better that way.

[STC]

There really is too much level difference. Looking at the RMS levels, 1A is around .8-1.0 db louder than sample 1B. 1C is close to A in the left channel and .5 or .6 db louder than A in the right channel. If you listen a few times you can hear this. I heard it from listening a few times, and then tried to measure it. I could hear there was a loudness difference and an image shift between 1A versus 1B and 1C. These things will sound like more than what they are and that isn't even seeing if there is a frequency response difference.

 

It would have been better to use a test tone and match 440 hz or 1 khz for level by measuring the voltage at the speaker inputs. If I had the original files, it might be possible to compensate for these factors well enough to work. With the conversion to MP3 quality over youtube doing so is iffy.

 

One reason normalizing might not work is very low frequency noise. Something not heard all that much happening at 60 hz and below could corrupt the normalizing. A distant rumble from a truck or lawnmower at some low frequency you wouldn't notice yet gets picked up in one recording and not another. You might try to EQ out everything below 200 hz, then normalizing, and then doing reverse EQ. The match might be better that way.

 

Thanks again for your input. I will PM you the direct link to the original files.

 

The image shift could be due to my inability to keep my head fixed. Anyway, I was avoiding the proper way to DBT and just wanted to feel how like most of the audiophiles judge sound by listening without proper level matching.

 

The arrangements of sample A, B and C were random. Sample A in track 1 may not be from the same amplifier in track 2. It could be sample C.

 

I don't think any of the tracks would drive the Amp into clipping as they were all recorded way below my usual listening level and yet they sounded different. The low impedance of SL shouldn't make a difference as they should able to deliver the power demand for the low playback volume as the peak volume never increased above 95dB.

 

It is so difficult to make up my mind as to which amplifier I should get. [emoji15]

[esldude]

Thanks again for your input. I will PM you the direct link to the original files.

 

The image shift could be due to my inability to keep my head fixed. Anyway, I was avoiding the proper way to DBT and just wanted to feel how like most of the audiophiles judge sound by listening without proper level matching.

 

The arrangements of sample A, B and C were random. Sample A in track 1 may not be from the same amplifier in track 2. It could be sample C.

 

I don't think any of the tracks would drive the Amp into clipping as they were all recorded way below my usual listening level and yet they sounded different. The low impedance of SL shouldn't make a difference as they should able to deliver the power demand for the low playback volume as the peak volume never increased above 95dB.

 

It is so difficult to make up my mind as to which amplifier I should get. [emoji15]

I thought based upon the one that was quieter that the order was random for each piece of music. Once I checked RMS levels I didn't go further.

 

Sent from my Nexus 6P using Computer Audiophile mobile app

[semente]

It is probably true but I am being just like a normal person who would judge sound quality by listening.

 

From the samples, I find it hard to pick one clear winner.

I would try listening long term instead of doing direct A/B comparisons and forget about the results of your test (which in my opinion are meaningless).

Try downloading Mario's (PlayClassics) files and use them as test tracks, together with some adequately recorded orchestral music or anything else you prefer that is complex and challenging.

 

Personally I would also rule out any amplifier which has a high output impedance.

 

R

[esldude]

Looks like I had a post go missing somehow.

 

I listened to the video again. Picking just from that I would go with the one which was quieter, the one used in the 1B sample. I would reiterate that level matching is a base level important step in these comparisons. Just on what I am hearing over youtube the amp for 1B seems a little nicer. I don't consider listening to youtube a good methodology.

[sandyk]

I don't consider listening to youtube a good methodology.

 

With a maximum of 187Kilobits .aac it isn't much better than .mp3 !

(That assumes that you download the video at 1280 x 720)

[STC]

I would try listening long term instead of doing direct A/B comparisons and forget about the results of your test (which in my opinion are meaningless).

Try downloading Mario's (PlayClassics) files and use them as test tracks, together with some adequately recorded orchestral music or anything else you prefer that is complex and challenging.

 

Personally I would also rule out any amplifier which has a high output impedance.

 

R

 

I have done long term listening without knowing which Amp and unfortunately I am not consistent with my choice.

 

Throw in a not so familiar song and my guess would be no better than flipping a coin even though I am listening for extra emphasis on the HF which I thought my original Amp was lacking.

 

So far out of the six subjects who took part, they are just split between the loudest of the two. No clear winner.

 

Maybe, I should try with the original files.