Many people believe that high-resolution audio (HRA) offers the potential for better sound quality than CDs by using higher sampling rates and greater bit depths—typically 96 kHz or more and 24 bits compared with CD's 44.1 kHz and 16 bits. A higher sampling rate can represent a wider range of frequencies extending into the ultrasonic region, while a greater bit depth can represent a dynamic range beyond the 90 dB practical range of CDs.
The big question is, can humans actually perceive the higher frequencies and wider dynamic range offered by HRA? Bob Schulein, a long-time audio-industry consultant, and Dan Mapes-Riordan, an expert in psychoacoustics—both founding partners at ImmersAV Technology—have developed three sets of audio files that allow anyone to explore this question for themselves.
One set of files is used to determine the dynamic range you can perceive. They are based on a clip from "Straighten Up and Fly Right," performed by Linda Ronstadt and the Nelson Riddle Orchestra. It was imported from a CD into the editing program Audacity at 44.1 kHz/24 bits.
Since the purpose of these files is to test dynamic range, not frequency response, it was unnecessary to use a recording at a higher sampling rate. And although the original source file did not have 24-bit precision, adding eight more bits allows the level to be adjusted with greater precision than the original file. The final sample was also normalized to the maximum amplitude without clipping.
To test high-frequency perception, they recorded one swipe across a set of Latin Percussion chimes using a GRAS 46BL measurement microphone and 26CS preamp with an on-axis frequency response out to 40 kHz (+3 dB). The mic signal was recorded on a Tascam DR-100 at 96 kHz/24 bits and normalized in Audacity. The chimes have plenty of energy above 20 kHz, making the sound ideal for testing high-frequency acuity.
In this graph, you can see the spectrum of the chimes recording as measured from the output of the Tascam recorder and the amplifier.
Each file in the dynamic-range test suite consists of a 20-second clip of the music played first at "reference" level, then at a reduced level, and finally again at reference level. This sequence is repeated in 19 separate files, and the level of the second clip is reduced in increments of 5 dB in each one, all the way down to -90 dB. As you listen to the files in order (put them in a playlist), note in which one the second clip becomes inaudible to you. Obviously, you'll need to listen at the highest tolerable reference level in order to measure the maximum dynamic range you can perceive.
The dynamic-range test files attenuate the clip in 5 dB increments from one to the next.
You might wonder why the attenuation stops at "only" 90 dB, which CDs are capable of. Why not go down 100 dB or more? According to Bob, "This was my original goal, but when I exported files with as much as 100 dB attenuation, I found a practical lower level, beyond which the analog output of the three players I measured could not faithfully track the 5 dB steps of attenuation.
"I used pink noise to take the measurements of the players' outputs. The players I measured were a Tascam DR-100, a Sony NWZ-A17, and a Pono Player. These devices have an analog output stage with a noise floor that eventually dominates the situation and doesn't let you go below -90 dB. There might be some players with a better noise floor, but I have yet to find one. And there is little chance that one would be quiet enough to track the full dynamic range of a 24-bit actual signal. From our testing to date, no one has been able to hear anything below 85 dB attenuation."
"The point is that dynamic-range specifications associated with 24-bit words are meaningful as long as you stay in the digital domain. But once you play the file back through a DAC [digital-to-analog converter], there is a practical noise floor that you can't go below. When someone takes the test, they start to become calibrated as to what a 60 or 70 dB signal-to-noise ratio sounds like—it gets pretty darn quiet, to the point that there is very little of interest down that low, even when listening with insert earphones in a quiet space."
To which Dan added, "It will be interesting if we get reports of people reaching 90 dB attenuation and still being able to hear the signal. This will require a very high-quality DAC analog stage and a quiet environment. We can easily amend the files to include more attenuation steps if necessary."
The other two sets of files test the limits of your ability to hear high frequencies using the recording of the chimes. In one set, a highpass filter with a slope of 192 dB/octave is applied to the recording—the sample is played at full bandwidth, then with the filter applied, and finally at full bandwidth again. The filter cutoff starts at 1 kHz and increases in 1 kHz increments in each subsequent file, all the way up to 30 kHz. As you play the files in order, note at what point you can no longer hear the second sample.
A highpass filter is applied to the second sample, increasing its cutoff frequency in 1 kHz increments from one file to the next.
In the other set of bandwidth-test files, a lowpass filter with a slope of 192 dB/octave is applied to the chimes recording. In this case, a full-bandwidth sample is followed by a lowpass-filtered sample, ending with the full-bandwidth sample again. The lowpass cutoff frequency starts at 2 kHz and increases in each successive file in 1 kHz increments up to 30 kHz. As you listen to the sequence of files, note at what point the full-bandwidth and lowpass-filtered samples sound the same. As the cutoff frequency continues to increase from that point, the three samples should all sound the same and you get no further benefit from the additional higher frequencies.
A lowpass filter is applied to the second sample, increasing its cutoff frequency in 1 kHz increments from one file to the next.
Because of a psychoacoustical phenomenon called masking, the high-frequency limit you note from the lowpass-filter test will likely be lower than the result from the highpass-filter test. The lower frequencies in the recording mask the higher frequencies, whereas in the highpass-filter test, there are no lower frequencies. As you might surmise, the lowpass-filter test is more closely related to real-world content, which typically has a wide range of frequencies.
The Test System
Of course, you can use whatever audio-playback system you have, and doing so will address the question, "What happens when I use my existing gear to play HRA files?" But to test the limits of your hearing acuity, you need equipment that can reproduce everything in these files. So you need to start with a high-resolution audio player, such as the Sony NWZ-A17 or Pono player that can play uncompressed 96 kHz/24-bit files; Bob has measured both and found them to be sufficient.
The dynamic-range test can be played on any reasonably high-quality audio system. Using headphones—especially inserted earbuds with excellent sealing capabilities—will give the best result because they block ambient noise from reaching your ears.
The high-frequency tests are more system-dependent, since they rely on the ability to reproduce frequencies above 20 kHz. If you have headphones with a useful frequency response out to 30 kHz, you can connect them directly to the player. However, most headphones and earbuds do not exhibit smooth response that high, as seen in the following measurements by Bob:
Bob Schulein measured the frequency response of circumaural and in-ear headphones using a tiny microphone placed next to his eardrum.
To use a speaker, you need an amplifier with bandwidth out to at least 30 kHz. Bob recommends the Dayton Audio DTA-120 class-T amp with 60 watts/channel, which is available from Parts Express for under $90. He also notes that this amp's signal-to-noise ratio (SNR) is not sufficient to conduct the dynamic-range test, but it's well-suited for the high-frequency tests.
To conduct the tests for yourself under the most optimum conditions, you need a high-res audio player, a wideband amplifier, a wideband speaker for the high-frequency tests, and insert earbuds for the dynamic-range test.
For the speaker, Bob built a custom rig using a Tymphany XT25SC90-04 tweeter, also from Parts Express, mounted in a 90° curved PVC joint. This tweeter has the necessary bandwidth and output level to conduct the high-frequency tests, and the whole thing costs very little.
Constructing a wideband tweeter is not difficult or expensive.
The two tweeters Bob built exhibit excellent frequency response out to 40 kHz.
In this graph, you can see the spectrum of the chimes recording as measured from the output of the Tascam recorder and the speaker; notice how similar the speaker graph is to the recorder's output above 1 kHz.
Bob recommends placing the speaker about eight inches from and pointed directly at your ear. Using an SPL meter, set the output level to about 90 dB SPL as measured from where your ear will be.
The Test Results
Bob brought the test files, player, amp, and tweeter rig to the 139th AES (Audio Engineering Society) convention and a meeting of the Chicago section of AES, both in November 2015. In total, about 30 people took the tests, with the following results:
No one was able to discern a dynamic range greater than 85 dB or high frequencies above 20 kHz, and the averages were less than that.
Here are links to the test files developed by Bob and Dan. You can download each set of files separately or combined into one large file.
Dynamic Range Tests
High-Frequency Tests Using Highpass Filter
High-Frequency Tests Using Lowpass Filter
Combined HRA Test Files
In addition to representing higher frequencies and wider dynamic range than CD, the sound quality of HRA is affected by the lowpass filters used during recording and playback, which can introduce audible artifacts within the normal hearing band. This aspect of HRA is not addressed in these tests. Bob and Dan are working on ways to test for such artifacts, and I am eager to see what they come up with.
For more about this project, check out the video made by Bob Schulein:
Bob and Dan recently appeared on the Home Theater Geeks podcast
to talk about the project. Also, the ImmersAV Technology website
is full of interesting info.
I encourage you to post your results in the comments! But please do not quote this entire article when you post a comment. If you want to respond to something specific in the article, feel free to quote just that portion, but the whole thing is way too long to wade through in the comments section. Thanks!