Advanced ABX, New cross-platform ABX application Options

[googlebot]

Advanced ABX

No currently available ABX application is able to switch without audible transients between tracks. The resulting clicks and pops can both increase the number of false positive and false negative results. The former happens, when the type of clicking correlates to the type of switch: A/A or A/B. False negatives become more probable, when loud clicks desensitize your hearing or keep you from turning up the volume to more revealing listening levels.

Advanced ABX provides transient-free switching!


Other exclusive features:

• Impedes "fishing" for good results by design by enforcing a chosen number of rounds.
• Customizable transition length to adapt to different material, for example in- and out-of-phase audio data.
• Easily extendable transition mechanism: A very simple API allows scientists and developers to drop-in their own custom transition modules, for example to allow completely seamless switching.

Non-exclusive features:

• Computation of statistical significance (p-value), also known as "probability of guessing".
• Multi-threaded, low resource consumption.
• Cross platform Java binary. Custom Windows and Mac executables.
• Nice and simplistic GUI.
• Textual output for easy publication of results.

Screenshot:



Notes:

I could only test on a Mac, yet, where it works flawlessly. Please report wether it also works on other platforms. Input is currently WAV only, and the files must have identical format and length. The program is basically able to work with all PCM integer bit depths up to 32 bit, but in practice one is limited by ones specific JRE's capabilities. Java on the Mac, for example, is limited to 16 bit. High sample rates should not be a problem anywhere. The back-end already supports looping, it just isn't implemented in the GUI code, yet.

The source code is going to be released later on. Please don't drown me in feature requests, actual code submits are much preferred.

Download:

Go here.

This post has been edited by googlebot: May 1 2010, 21:53

[C.R.Helmrich]

Hi googlebot,

just read this thread for the first time. Was away from HA for the last few weeks due to the same reasons as Alex.

Thanks very much for your work! I work in Fraunhofer's audio coding department, and I can tell you that artifact-free switching in a blind test software makes testing much easier. However:

AA is basically just a tool with feature completeness already in sight. One may add a couple of automatic decoding filters, maybe automatic track alignment (encoding delay correction) a more complete UI and that's about it.

Sorry, but that's incorrect. Think of all the things that could go wrong. For example, what if the compared WAV files have differing

• lengths, as mentioned? Solution: take length of shortest file
• offsets? Solution: automatic sample-accurate delay alignment, see Synchrotron
• sampling rates? Solution: error message, or automatic resampling of lower-sampling-rate files to highest sampling rate in the test (in combination with previous points, quite some work!)
• Bit depths? Solution: increase word length of lower-rate files to highest word length in the test
• Number of channels? Similar solution as above

These and Alex's feature description regarding segmenting and looping (which I strongly support) are only some of the features that would attract interest away from ABC-HR and the foobar plug-in to your software. Here is another thing which hardly anyone talks about on this forum but which I'd like you to consider.

The scientific community has been using the MUSHRA test methodology in the evaluation of audio codecs for years. It's also used for the majority of internal tests at Fraunhofer. I disagree that it's only useful for "intermediate audio quality". I successfully used it for AAC tests between 160 and 256 kbps stereo (for example, with less than 10 listeners, it revealed that a particular codec was not transparent at 256 kbps). Take a look at the screenshot in this thread on the RateIt tool written by Jean-Marc Valin. Fraunhofer uses a self-made MUSHRA tool for testing (no, sorry, can't share it), which looks very similar to RateIt but has the additional feature of looped playback, with hotkeys for setting loop start and end points for faster changing of loops during playback. Send me a PM if you'd like to know further details. Why don't you consider turning your AA tool into a MUSHRA tool? Artifact-free switching is just as important for MUSHRA as it is for ABX!

I'm not surprised that nobody contributed any code to your project so far. It's much less work asking for improvements than implementing them Same with me: I program for a living during day time, I don't want to continue doing it at night. But if you can use some software engineering support (requirements, planning, GUI design mock-ups, ...), I'd be willing to contribute. Last note: notice how nobody replied to Jean-Marc's RateIt thread! You're not the only one with a good idea but little (initial) support

Best,

Chris