speakgoodchinese

[toads]

Thanks to your diagnosis, I've managed to get it working in ubuntu by replacing every occurence of g_thread_init(NULL); in the source with

 
if (!g_thread_supported()) 
   g_thread_init(NULL);

and it seems to have fixed the problem after recompiling (found this by googling). Everything now seems to be working great, except that I get

** (sgc:15204): WARNING **: No sound from praat!

whenever I try to record. Not really sure how praat works so I don't know how to approach this one.

[imron]

presumably the problem your software has with Ubuntu will be solved at some point.

I've managed to get it working in ubuntu

See, that didn't take too long.

[Skinkie]

http://www.fon.hum.uva.nl/praat/download_linux.html

This is praat, it has a static binary, see if you can get audio out of it by default We use this thing. I'll put your code into our git.

[roddy]

Crosses fingers

Presumably someone will fix the problems with XP soon

/Crosses fingers.

I'm attaching the error report that XP tries to send to Microsoft on the offchance it means anything to anyone. Greek to me.

f4f1_appcompat.txt

[toads]

When I do "record mono sound" in praat it works fine; I can play it back etc., but when I press record in sgc it looks like it doesn't even try to record anything, it just immediately gives me the above message, no 5 second wait or anything.

[Skinkie]

It could be this compile uses OSS and praat could be using portaudio by now. You can check it out by stracing the SGC app. (If it borks on a /dev/dsp)

Does playing audio work?

[toads]

SGC plays sounds perfectly, and I can't find anything related to /dev/dsp in the strace. For what it's worth I tried using both the current static praat binary from the website (4.6.12) and the one from ubuntu's praat package (4.5.06); both have the same effect.

[Skinkie]

Do you have a messenger like msn/icq/etc.? Otherwise mail me. There was one person on Linux having a general sound card problem. But no sound from praat is the error when NULL is returned as pointer.

[roddy]

Quick bit of extra info. Doctor Watson log says:

Application exception occurred:

App: C:Program FilesSpeakGoodChinesesgc.exe (pid=916)

When: 2007-8-6 @ 18:11:36.703

Exception number: c0000005 (access violation)

Event viewer says

Faulting application sgc.exe, version 0.0.0.0, faulting module ntdll.dll, version 5.1.2600.2180, fault address 0x00001016.

I tried to start SGC in safe mode, but it still crashed.

[newyorkeric]

roddy,

Did you get a chance to try the program? Any thoughts on its effectiveness?

Has anyone else tried it?

[mikevwilliams]

I may have found the problem people have been having, at least i've managed to get the program to work on my own computer when it didn't before. I think this may have to do with the Unicode settings for your computer. Mine is set to run non-unicode programs as Chinese. In the shortcut for SpeakGoodChinese, I changed the Current Code Page, under the Options tab in the shortcut properties to United States, and now the program works. By the way, I am using the latest beta from August, so I haven't tried any other versions but it could be the solution to everyone's Windows problems.

[Skinkie]

Mike if this is the problem, we really owe you something!

[imron]

If that's the case, you might also want to try using AppLocale, which should achieve the same results without needing to restart your machine, or change the default settings (which might be problematic for those users who need the default settings to be some other language).

[Skinkie]

Maybe it has something to do with me not providing all locales... I wonder if this is so.

[Czech Cara]

Hi all, is there anyone who tried a new tone recognition speech program on

http://www.mdbg.net/chindict/chindict.php?page=main , the link only appears when you first open the site.??? I failed to install it... Thanks

[newyorkeric]

There is a thread about the software and installation issues here:

http://www.chinese-forums.com/index.php?/topic/13775-speakgoodchinese

[roddy]

Merged the two topics.

The change suggested by mikevwilliams above seems to fix the problem - I haven't had a chance to actually use it yet (will wake people up and get into trouble if I start shouting into the computer now) but it's getting to the GUI and all looks well.

Detailed instructions for anyone who still hasn't got it working, and skinkie if he wants to copy and paste it to the programs webpage.

Find the /speakgoodchinese folder

Right click on the sgc.bat file

Properties

Options

Set current code page to United States.

Incidentally, I didn't need to restart.

[mikevwilliams]

Skinkie,

Is there any way of you providing a fix for this problem with the program itself. I ask this because I am running Vista now, and I can't change the Current Code Page like I could when I was using XP. I would really like to try this program out, but I can't make it work in Vista unless I change the non-unicode program setting back to english instead of chinese.

Thanks

[Skinkie]

I still did not (have the time to) find out which thing is broken (or better to say: missing) that causes this crash and is fixed with this locale change. A long shot would be getting this program to work with the GTK+ win32 installer.

...I don't know if you have the 'skills' to look how to do this.

[SGC_Rob]

Hi,

I am the one responsible for the recognizer and tone models.

Detailed instructions for anyone who still hasn't got it working, and skinkie if he wants to copy and paste it to the programs webpage.

Find the /speakgoodchinese folder

Right click on the sgc.bat file

Properties

Options

Set current code page to United States.

I have put this workaround on the download page.

http://www.speakgoodchinese.org/download.html

Thanks

Does the program test absolute levels as well as relative levels? That is, does it test that my 2nd tone rises high enough or that my 4th tone falls low enough? Or is it testing that my 4th tone falls enough relative to the starting point?

Yes. SGC compares the overall pitch (top) of the correct tone model to the pitch of the utterance. If the test utterance is more than 3 semitones below or 6 semitones above the model tones, it is discarded. Then the ranges of the tones (rises/falls) are compared. With the same cut-offs. Only then is the tone recognized. The recognizer too does use the shape and extend of the rises and falls.

I read somewhere in the documentation that the program is generous with regards to giving correct responses so as not to discourage new learners. In future versions it would be helpful if the user could choose the tolerance level.

Actually, you can by setting switches in the scripts. But you wouldn't want to.

The short reason is, the student should get the benefit of the doubt. A longer reason is that the recognizer type (Dynamic Time Warping on model contours) is much more likely to falsely reject a correct tone than to falsely accept an incorrect tone.

We managed to get both equal (the optimum solution) at around 15% for actual use data. This error rate is rather high for single bi-syllabic words, but building recognizers for beginning language learners is still an unsolved problem (for read "typical words" by native speakers we got below 5%, a meaningless statistics).

The tuning is done in two ways. First, SGC will pick the correct tone if it is close enough to the best (incorrect) solution. Second, many confusions which are part of Mandarin assimilations tend to remove the differences between tones in certain contexts (eg, tones 2 and 3 in the context of another 2 or 3). We just consider the utterance correct if the recognizer picks the wrong tone of a confusionable pair (eg, if 2-3 is recognized as 2-2).

These rules can be found in SpeakGoodChinese/SGC_ToneProt/SGC_ToneProt.praat

Look for:

# Special cases

However, if someone knows more about chinese tone assimilation/coarticulation phenomena, please feel free to point out our mistakes. We really are NOT well informed in the phonetics of Mandarin. Especially the neutral tone proved to be an enigma (however, we alread contacted Yiya Chen).

Btw, we also worked on a pronunciation recognizer. That proved to be much more difficult and the end result was so bad and the project time so short that we stopped that. It is still open to development, though.

Rob van Son