ABX Audio

This is an Android app available at Google Play:

ABX Audio

QUICK START GUIDE

Let's cut to the chase: how to use this app.

Load 2 audio clips onto your device (in any directory). Start the app. Select the clips as A and B. Start an A/B/X test.

Usage Tips

By default, this app uses the device's default storage location. This varies across devices and versions of Android, but is almost always in internal storage. For example on my Galaxy Note 4 phone running LineageOS 16 (Android 9), it is:
/storage/emulated/0/Download
You can see this location in app "Settings".

The simplest way to use the app is to copy your music files to this location. Android has wonky permission rules for external SD cards. Because of this, the external SD card can be used, but it takes some extra steps. Sorry about that - I didn't make up those rules, but I can help you work around them! If you want to do this, keep reading:

Using the external SD card requires 3 steps:

  • Find the path to the external SD
  • Create a directory in the right path on the external SD
  • Enter the full path to this directory in ABX Audio settings

    • Step 1

    On my phone, this is the path to the external SD card:
    /storage/A9BC-0E00
    This varies from one device to another. Use your device's file explorer to find the path to the external SD card.

    Step 2

    On the external SD card, each Android app can only see data under its "home" directory, which Android defines as:
    ./Android/data/appname
    For ABX Audio, this is:
    ./Android/data/com.mrc.abxaudiofree
    Add this path suffix to the above path to the external SD card. So on my phone, if I want to use the external SD for clips, they must be in this directory:
    /storage/A9BC-0E00/Android/data/com.mrc.abxaudiofree
    Use your device's file explorer to find this location (create directories as needed). You can also make a subdirectory here. For example, I create a subdirectory called "abxClip". Copy your audio clip files to this directory.

    Step 3

    Now even though the app can use this directory, you can't navigate to it in the app's file explorer, because that requires going up through parent directories with restricted permissions. The only way to get there is directly. To do this:

  • Run ABX Audio and go to settings
  • Set "Clip Dir" to the path to your external SD card (manually type in the path, in my case /storage/A9BC-0E00/Android/data/com.mrc.abxaudiofree/abxClip
  • Hit BACK and click the Clip A button
  • Click the "Set Default Path" button - the default path you entered will appear above
  • Click the "Path" button - now it will go straight to that path

    • You only need to make these settings changes to the default path once. The ABX Audio app automatically saves settings for future use.

    What Is an A/B/X Audio Test?

    Click here to read more about what A/B/X, also known as double-blind audio testing, is all about, and how to interpret the results.

    An A/B/X test consists of a set of trials. You decide how many trials you want to do. The score is based on how many trials you get correct. The ideal confidence threshold can be as low as 51% or as high as 99%, depending on the purpose of your test.

  • Low threshold (51%): high recall, low precision. Detects the most subtle differences, but has false positives.

  • High threshold (99%): low recall, high precision. Reduces false positives, but only detects obvious differences.

    • With more trials, your results have less variance and more consistency. But too many trials in one sitting causes fatigue, impairing results.

    An A/B/X test is not timed. It is not a race. It is designed to be done in a calm, quiet environment, with no distractions, when you have all the time you need to listen carefully.

    Each trial: The sounds you want to differentiate are called "A" and "B". You decide in advance what A and B are, and you know which is which. In each trial, the computer designates an X, always either A or B. The computer picks X randomly for each trial. In each trial, you listen to all 3 clips and decide: X is A, or X is B.

    The app has a setting that randomizes (swaps) A and B for each trial. This is not necessary, but some people prefer to test this way.

    A/B/X Testing Tips:

  • Find a quiet environment
  • Take as long as you want
  • Listen to the clips as much as you want
  • Switch back and forth as much as you want, as quickly or as slowly as you want.
  • Start by listening only to A and B, not X, until you can tell A and B apart.
  • Then, listen to X and decide if it's A or B.

    • Differentiating A from B is more about the brain than the ears. Your ears are already picking up some of the differences, but your brain is ignoring them so you don't "hear" them. What to listen for:

  • Extreme low and high frequencies
  • Fast transients (castanets, cymbals, the sudden thwack of a stick hit, plucked strings)
  • Background sounds (can be masked at low bit rates)
  • Spatial - image breadth and depth

    • During the test, audio clips A, B and X are started together and kept in sync with each other. Normally, clips A and B are the same musical excerpt except that one has been modified in some way (compressed, equalized, etc.). Keeping them in sync during the test means when you switch, the music keeps playing (your position in the musical selection doesn't change).

    Equipment and Recordings

    Make sure you have good equipment and recordings. You won't hear differences if your equipment is masking them, or if your sound clips don't use the full frequency bandwidth. Most compression algorithms - AAC, MP3, OGG - save space by masking sounds that people don't usually notice. This is usually in the extreme high and low frequency ranges. Most phones and tablets have very good sound quality. Usually the limiting factor is the headphones.

    Examples:

  • If your headphones don't reproduce extreme low or high frequencies, or reproduce them with distortion, it makes it harder to discern differences.
  • If your original recording doesn't have much energy in the extreme high or low frequencies, then the compressed file will sound like the original.

    • Don't confuse wide frequency response with exaggerated bass and treble. A headphone can boom out a ton of bass as 60 Hz, yet fail to reproduce sounds below 40 Hz. A headphone can sound bright with lots of treble by boosting 2 kHz to 12 kHz, yet fail to reproduce sounds above 14 kHz. These headphones would not be effective for ABX testing.

    Some headphones are designed to "sound good", not to be accurate. The euphonic distortion designed into them makes them unsuitable for ABX testing. Ideal headphones for ABX testing have wide, flat frequency response with low distortion and fast, smooth spectral decay.

    ABX Audio App Options

    Some people like to know how they're doing during the test. Others don't want to know whether their guesses are right or wrong until the end of the test. ABX Audio can run a test either way; use the settings screen to set your preference.

    At any time during a test (even mid-trial), click Results to check your results so far. On results screen, click Back to continue the test in progress, Quit to end the test. On the test screen, click Back to end the test.

    Sample Clips

    When using this program, you provide your own clips. But I created a few to help get you started. Both are from original recordings with excellent sound quality. Sharing brief excerpts for the purpose of testing constitutes educational use which doesn't violate copyright. Copy these to your device and try them out with ABX Audio. I suggest starting with the 64k MP3 because it is the worst quality with obvious distortion. Use that to train what to listen for, then try each higher bit rate as the distortion becomes more subtle. The VBR are the hardest to differentiate in an ABX test. Even though their average rate is about the same as 192 kbps, they use up to 320 when needed for complex waveforms.

    Here's a flute quartet with castanets. The castanets have extended HF transients and the flute tones have complex timbres, both of which can reveal lossy compression. It is from: Tour de France by the Quintessenz flute ensemble, on Geniun GUIN 87108.

    Flute original
    64k MP3
    96k MP3
    128k MP3
    192k MP3
    VBR V2 MP3 (112k - 320k, average 196k)

    Here's a small orchestral ensemble. There are several string instruments and a harpsichord, a complex waveform with extended HF content which can reveal lossy compression. It is from: Vivaldi for Diverse Instruments, Philharmonia Baroque, Nicholas McGegan, on Reference Recordings RR-77.

    Vivaldi original
    64k MP3
    96k MP3
    128k MP3
    192k MP3
    VBR V2 MP3 (128k - 320k, average 191k)

    Confidence Percentile

    This expresses how confident you can be in the test result. Another way to say this: Given that you got X out of Y trials correct, and each trial has 2 choices (X is A or X is B), what is the probability that you could get that many correct by random guessing? Confidence is the inverse of that probability. For example, if the likelihood of guessing is 5% then confidence is 95%.

    For example, if you get 1 trial correct your confidence is 50%. If you get 2 of 2, confidence is 75%.

    This is discussed in more detail in the article linked above.

    The app computes your confidence percentile as you go through the test and shows it on the Results screen.

    Musings

    Human perception gets interesting near the threshold of audibility. When two sounds are so close you can barely tell them apart, the actual differences may be perceived as something different from what they really are. For example, try the same clip, with one 0.2 dB quieter than the other. The only difference is volume, but it's often perceived not as a difference in volume, but in space or richness. An ABX test simply determines whether you can tell them apart. It doesn't mean you can qualify exactly what the difference is.

    This sometimes reveals itself in subjective reviews of audio equipment. For example, a company makes their CD player output 0.5 dB louder than standard. When reviewers A/B compare it with another, the reviewer perceives this small difference in volume as sounding slightly more rich, more full, and glows about a "veil lifted from the music". It's not a placebo effect - the reviewer is definitely hearing a real difference. But what he's hearing is not what he thinks it is.

    Another point: the human ear perceives the frequency and time domains differently. We can sometimes discriminate timing more precisely than one would expect from frequency response. Put differently for engineers: Take a musical signal with fast transients and do a DFT ( discrete fourier transform). From the DFT, remove frequencies too high for the listener to hear as pure tones. Now reconstruct the signal from the DFT's remaining frequencies. The listener will sometimes be able to hear the difference as slightly less crisp or smeared transients. How is that possible, when you only removed frequencies he couldn't hear? That's what is meant by saying the human ear has greater acuity in the time domain, than one would expect from its frequency response. As frequency increases toward one's limits of perception, there's a small range of frequencies you can't hear as pure tones, but you can hear their contribution to the rise time of an impulse. Given how non-linear the ear and perception is, this shouldn't be surprising.

    Have fun! And do not be dismayed - differences can be subtle, and it takes time and experience to train the brain to make the most of what the ears are telling it.