Registering and logging in removes this ad.
Registering and logging in removes this ad.
Romanian Language Voice Recognition
Submitted by Neuronyx on Wed, 04/18/2007 - 18:00.
In the last two days I've played with Dragon Naturally Speaking 8.0 Italian version and found that training and voice recognition is almoust the same as Romanian from a phonetically point of view (both languages are Latin based, very similar in fact when it comes to vocals or other sounds) and even without too much training or a quality microphone, I have reached a very good speech recognition rate.
The only problems I've encountered were when I tried to add into the vocabulary some new words containing central european characters (iso8859-2 or iso8859-16 - ALT 0xBA) like "ş" and "ţ", the program simply refusing to add in the list any of the words formed with those characters. The dictionary also refused to delete or replace the italian "ce" conjuction words spelled like the romanian "şi" ("and" in english) and because of that, errors were everywhere.
It would be very nice to have those problems fixed because, except this, everything works just perfectly in romanian, even with a few words added and a little training. After adding a few thousands words from some documents I have reached a succes rate around 95% in just 1 hour of training, prooving that Italian recognition is pretty close to the Romanian (future:)) version that I hope it would be launched soon.
Regards
Radu O
Radu, Fantastic! I had
Radu,
Fantastic! I had heard the two languages were similar but didn't realize just how close they were.
I'm not sure you can add the special characters without the Professional version, but I won't say its impossible. If you have any skill with programming, you might want to look at Vocola or UniMacro, free programming add-ons for DNS. However, again maybe I'm promising too much: I don't know whether they will work with non-English versions.
http://vocola.net/
http://qh.antenna.nl/unimacro/aboutunimacro.html
Bruce
Dear Bruce Thank you for
Dear Bruce
Thank you for answering so quickly and thanks for the links. I'll try to make something but I'm not promising anything because it's more like a fun hobby for me not an everyday job. I am interested in this since 1997 when I found some articles about neural networks and voice/image recognition programs, and, because then I had build an romanian database of words for my personal correct spelling in microsoft word (it didn't exist at the time - not a very profesionall method, just took a few hundred official electronic documents and laws, replaced the space character with enter and made a list of thousands sorted by number of duplicates) I wanted to see if that list can be used in a voice recognition software. At that time (Dragon Dictate 4.0 if I remember well) it was too much time consuming to enter the sound file for every word so I decided to leave it for the following years.
Nowadays I don't think it will be very hard. Yesterday I've replaced the data0x.bin file in the training directory with some romanian texts and trained it for a while but didn't get very good results with this method. Better results were obtain in spelling each individual word from a smaller document but it was time consuming already. I'm wondering if the training module can be customised to be more sensitive to recognition or to give more control in learning (or deleting) certain words because if so, I think a romanian language extension (even for my only personal use) can be easily obtain.
Anyway, I'll put some results after more researching and playing with the programs from your links, and if they are promising, maybe someone will find time and resources to produce the first romanian voice recognition software on the market.
Regards
Radu O
Radu hello when you add new
Radu hello
when you add new training texts, first add any words they contain that aren't in the Vocabulary -- see here
http://www.synapseadaptive.com/Joel/changingaenrollmentatext.htm
Judy
Word List and DNS 9
Hi,
Could you please tell me, would it be possible to get the Romanian word list you created for DNS? I would like to purchase DNS, or may consider it if I can get it working in Romanian as you have.
Regards
Stefan Bordeianu