Skip to main content ↑ ↓

Notice

Please note that most of the software linked on this forum is likely to be safe to use. If you are unsure, feel free to ask in the relevant topics, or send a private message to an administrator or moderator. To help curb the problems of false positives, or in the event that you do find actual malware, you can contribute through the article linked here.
Topic: ABC/HR questions (Read 14277 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

ABC/HR questions

I have a few questions regarding the ABC/HR java prog.
When setting up a test, are the offset and gain automatic calculations reliable?

ABC/HR questions

Reply #1 –
Offset correction worked very reliable for the samples I've tested (within its limits; it won't detect offsets greater than 400 milliseconds).
Volume correction simply scales the samples so they have equal RMS. It doesn't do anything fancy (like ReplayGain), but this should work fine, since the samples in a listening test are highly similar anyway.

ABC/HR questions

Reply #2 –
I believe I can safely say that I have a good experience with Java ABC/HR, and for offset detection I never encountered any problem.
For volume normalisation, I can't say if it's accurate but I'm sure that my ears haven't detected any problem either.

@schnofler: would you consider an option in preference as:
- [] enable auto-gain per default
- [] enable auto-offset correction per default
?

When a test consist in an evaluation of 200 samples, an automated way to perform these operation would be highly appreciated

ABC/HR questions

Reply #3 –
I personally use(d) wavegain to pre-gain all my samples at the same gain (92 db) before encoding for testing.  I think it's the safest method, but I never really tested the auto-gain.  Only thing is, it probably takes processor time, so it may be unreliable.

As a second note, the ABC/HR bug I mentionned during the 128kbps test was still found in the last edition of tests (48kbps) -- it regards file caching;  I'm not sure to what extent it can be fixed, but you can email me at [EMAIL REMOVED] if you need to reproduce it.

Peace,
Tristan.

ABC/HR questions

Reply #4 –
Quote
,Feb 13 2006, 06:58 PM]I personally use(d) wavegain to pre-gain all my samples at the same gain (92 db) before encoding for testing.
[{POST_SNAPBACK}][/a]

Before encoding? So if you have a sample at ~70 dB, you're increasing it by 20 dB before encoding? In this case, I fear that the test is flawed: for a same sample, quality could differ a lot if you modify (by a wide margin at least) the gain. For example, LAME VBR and [a href="http://www.hydrogenaudio.org/forums/index.php?showtopic=35030]Musepack[/url] have both problems with low volume samples, and your method would completely hide these flaws.

ABC/HR questions

Reply #5 –
Well, it matches the way that I encode my music anyways -- wavegain before encoding, so I rather not have those flaws.  In any case, most people don't play around with their volume knob so much when listening to music, so this attitude fits the scenario best.

ABC/HR questions

Reply #6 –
Question: how to restrict playback range in the ABC/HR config?
I tryed fooling it with a 88000 offset for original sample, but it did not liked it (frozen when loading the config file)

ABC/HR questions

Reply #7 –
Quote
Question: how to restrict playback range in the ABC/HR config?
I tryed fooling it with a 88000 offset for original sample, but it did not liked it (frozen when loading the config file)
[a href="index.php?act=findpost&pid=365508"][{POST_SNAPBACK}][/a]

The offset is stored in miliseconds, not samples.
A value of "88000" would mean that ABCHR have to start the playback after 88 seconds. That would be damn long samples for a listening test 
Add 1000 or 2000 to the computed and real offset of each encoding.

ABC/HR questions

Reply #8 –
Thanks.
I hope it will also help getting rid of the "click on start" bug.

ABC/HR questions

Reply #9 –
Are you using "java sound audio engine"? Go to Options\Settings\Playback and be sure you're using "périphérique audio principal". Clicking issues may be avoided with it.

ABC/HR questions

Reply #10 –
It's not a matter of clicking during playback.
Some samples have an audible click on start, making identification easy.

ABC/HR questions

Reply #11 –
Quote
It's not a matter of clicking during playback.
Some samples have an audible click on start, making identification easy.
[{POST_SNAPBACK}][/a]


That problem should be fixed in the [a href="http://user.uni-frankfurt.de/~bkuckuck/abchr-java-0.51b.zip]latest version[/url] (sources).

ABC/HR questions

Reply #12 –
I wouldn't trust automatic gain, whether it is RMS normalization or replaygain. These techniques were not designed for level matching two pieces out audio files at test. I can think of situations where any of them would fail in doing a proper level match.
 

ABC/HR questions

Reply #13 –
note regarding automatic offset detection: it fails on sample3 of the current 48kbps aac listening test.