There's a new module in CVS called 'masktest'. I spent the last few weeks writing this program. It can measure masking based on input from people. We need it to obtain better masking curves than we have right now (from Ehmer...). It's not finished yet, but you can get a feeling of what it's supposed to do. It will measure the masking between tone and tone, noise /tone, tone/noise and noise noise. When it's ready it's supposed to be downloaded by a lot of people who then take this test and send us a file back that contains the masking data. There are two directories: printmask is a tool for printing the contents of the files created during the test to stdout, masktest is the actual application. Before you do anything with it, read the READMEs. They tell you much more than I write here. There's also a huge todo list. The source as it is now lets you go very quickly through the test, with only 2 different frequencies, 2 masker amplitudes and 2 masker frequencies testet. This is useful for debugging. If you want to take the entire test you must comment something in and out in arrays.h, it's explained there. But be warned, that takes about 160 hours. We must reduce the number of frequencies and amplitudes tested. There's one known bug at the moment: The masker amplitude is not properly displayed in the coordinate system. I didn't fix it because the bug is hidden in a mess of code, namely Testwindow::dosomething() and Testwindow::whatsnext(). They will be replaced anyway, it wasn't worth improving. I need help for this and lots of other things. I'm at wit's end with some problems and I'll be in California for only one more week. I'm going on vacation after that and be back in Germany in February and can continue working on it. But perhaps someone has already finished it by then and thousands of people have taken the test :) And please, everybody, think of a better name for the project. I was not very creative with masktest. -- Sent through GMX FreeMail - http://www.gmx.net --- >8 ---- List archives: http://www.xiph.org/archives/ Ogg project homepage: http://www.xiph.org/ogg/ To unsubscribe from this list, send a message to 'vorbis-dev-request@xiph.org' containing only the word 'unsubscribe' in the body. No subject is needed. Unsubscribe messages sent to the list will be ignored/filtered.
On Sat, Jan 06, 2001 at 04:54:18AM +0100, Robert Voigt wrote:> There's a new module in CVS called 'masktest'. > I spent the last few weeks writing this program. It can measure masking > based on input from people. We need it to obtain better masking curves than we > have right now (from Ehmer...).You should administrate a hearing test at the begining, insert listening breaks, and perhaps collect for stats for analysis (i.e. at least age), also insert enough dummys for sample rejection. --- >8 ---- List archives: http://www.xiph.org/archives/ Ogg project homepage: http://www.xiph.org/ogg/ To unsubscribe from this list, send a message to 'vorbis-dev-request@xiph.org' containing only the word 'unsubscribe' in the body. No subject is needed. Unsubscribe messages sent to the list will be ignored/filtered.
> You should administrate a hearing test at the begining,It measures the threshold of hearing with tone and then with noise before the masking test.> insert listening breaks,It is planned to add the functionality to save and restore sessions. Let the user decide when he's had enough, or do you mean people should have a break after a few minutes, because hearing changes?> and perhaps collect for stats for analysis (i.e. at least age),This is already implemented. Try it and see if you like. It's just a proposal of mine and is meant to be changed.> also insert enough dummys for sample rejection.I don't understand. What samples do you mean? -- Sent through GMX FreeMail - http://www.gmx.net --- >8 ---- List archives: http://www.xiph.org/archives/ Ogg project homepage: http://www.xiph.org/ogg/ To unsubscribe from this list, send a message to 'vorbis-dev-request@xiph.org' containing only the word 'unsubscribe' in the body. No subject is needed. Unsubscribe messages sent to the list will be ignored/filtered.
Robert Voigt wrote:> > The source as it is now lets you go very quickly through the test, with > only 2 different frequencies, 2 masker amplitudes and 2 masker frequencies > testet. This is useful for debugging. If you want to take the entire test you > must comment something in and out in arrays.h, it's explained there. But be > warned, that takes about 160 hours. We must reduce the number of frequencies > and amplitudes tested. >You should make a shorter option, probably let the user quit whenever he wants, while making the order of tests random, thus still obtaining full coverage. Sorry if I'm proposing something already done, didn't have time to look at it. -- Beni Cherniavsky <cben@crosswinds.net> (also scben@t2,cben@tx in Technion) No, No! You're not thinking; you're just being logical. - Niels Bohr --- >8 ---- List archives: http://www.xiph.org/archives/ Ogg project homepage: http://www.xiph.org/ogg/ To unsubscribe from this list, send a message to 'vorbis-dev-request@xiph.org' containing only the word 'unsubscribe' in the body. No subject is needed. Unsubscribe messages sent to the list will be ignored/filtered.
> You should make a shorter option, probably let the user quit whenever he > wants, while making the order of tests random, thus still obtaining full > coverage. Sorry if I'm proposing something already done, didn't havetime> to look at it.Saving and restoring sessions is planned, but not done. I need help for it, as well as for other improvements. I'm not sure if it makes sense to make the order of the tests random. -- Sent through GMX FreeMail - http://www.gmx.net --- >8 ---- List archives: http://www.xiph.org/archives/ Ogg project homepage: http://www.xiph.org/ogg/ To unsubscribe from this list, send a message to 'vorbis-dev-request@xiph.org' containing only the word 'unsubscribe' in the body. No subject is needed. Unsubscribe messages sent to the list will be ignored/filtered.
>Maybe make it fixed but let the user say that he wants only a small >portion and choose a random portion. The idea is that most people >wouldn't want to devote 160 hours to this, even split accross many >sessions.I quote from the todo list: - Determine what combinations of frequencies and amplitudes should be testet and what resolution should be used. At the moment all possible combinations are tested at a luxurious resolution. The whole thing would take about 160 hours this way. We can make assumptions about where no masking can occur, and don't test these portions of the curves at all. It's just a question of making correct (!) assumptions and implementing it in the program. That way and by losing resolution in the curves we could go down to say 10-20 hours. Then we can still divide the whole thing into 4 independend programs, separetely downloadable. That is masking between: tone-tone tone-noise noise-tone noise-noise. That makes 3-5 hours for each. Haven't calculated, just a guess. -- Sent through GMX FreeMail - http://www.gmx.net --- >8 ---- List archives: http://www.xiph.org/archives/ Ogg project homepage: http://www.xiph.org/ogg/ To unsubscribe from this list, send a message to 'vorbis-dev-request@xiph.org' containing only the word 'unsubscribe' in the body. No subject is needed. Unsubscribe messages sent to the list will be ignored/filtered.
>How about we do all four tests at low resolution (like 6 tests across the >whole spectrum), then compair the results to our current expectations, >focusing in on the areas that disagree with our current knoweldge the >most.You mean getting 6 values, then let the software compare to what we already have and decide where to go into detail? Otherwise I can't make sense out of it. We already have low resolution data for tone-tone masking, and something worse for noise-tone masking. The purpose of this whole thing is to get curves that don't have to be interpolated like hell to fit Vorbis's needs. That's why I started with the resolution used in Vorbis (just to have something): 1/2 octaves for the masker frequency 1/8 octaves for the signal frequency 10 dB for the masker amplitude>We can't expect too many people to put int 20hours of listening tests, >but some will. It's much more reasonable to expect an hour or so out of a >great number of people, we need to find the best way to make that hour >count.I'd still say better 3-5 hours from less people with the resolution we need than 1 hour from a lot of people with data that must be interpolated too much. But let's make assumptions about where no masking can occur and see how much we can eliminate. This could be done by someone taking the whole test (perhaps we can eliminate something before that by looking at the Ehmer curves, so it won't take 160 hours) and then we'll see. I would assume that if there's no masking in one person's ears with one particular set of sounds, there won't be in anyone else's ears. I'm fixing some serious bugs in the code right now. Expect them to be in CVS by the end of this week. -- Sent through GMX FreeMail - http://www.gmx.net --- >8 ---- List archives: http://www.xiph.org/archives/ Ogg project homepage: http://www.xiph.org/ogg/ To unsubscribe from this list, send a message to 'vorbis-dev-request@xiph.org' containing only the word 'unsubscribe' in the body. No subject is needed. Unsubscribe messages sent to the list will be ignored/filtered.