From linux-users@xxx.ac.nz Tue Sep 18 14:39:51 2001 Date: Tue, 18 Sep 2001 14:37:24 +0000 From: Phill Coxon Subject: Re: Voice recognition for linux To: linux-users@xxx.ac.nz After Volker's post about Via Voice for Linux I purchased a copy of the Mandrake 8.0 Power Pack to try it out also. I have an AMD 900Mhz with 384Mb RAM with a Soundblaster Live Value. The headset I have is a Plantronics headset (specific model unknown) - approx $150 in Harvey Norman. Like Volker, I have a copy of Dragon Naturally Speaking 4 for Windows gathering dust and this has been my only experience with voice dictation previously. Under Windows Dragon Naturally Speaking worked OK, but after a lot of training (with both US and Australian versions) it was still only about 90% accurate (ie: fairly useless) and very sluggish. IBM Via Voice on the other hand... WOW! Installation was as simple as: rpm -ih ViaVoice* ...and it all worked instantly. Setup of the headset took quite a while until I realised the microphone gain was up at max and hence the signal was distorting massively. It pays to take the time to get your microphone / gain settings just right. Once I figured that out and went through about 30 minutes of training, I was able to load the dictation software and give it a whirl. The dictation software is fairly simple - a basic notepad / word-processor with simple editing, search & replace, font selection, colour selection etc. To dictate, just click in the text window and start speaking. The verdict... I'm *extremely* impressed. If I speak just a little slower than normal and ensure I acentuate clearly, the accuracy is already about 98-99%. Pretty impressive when you consider it is based on US accents. I can often read 2-3 large paragraphs without a single error. Making corrections is extremely easy - just highlight the word or phrase, and it will pop up a list of suggestions, or you can type in the correct words. It then adapts the vocabularly to ensure it gets that word / phrase correct next time. You can select text and have it dictate back to you in your own voice (ie: repeat back the recorded input), or using the standard computer voice dictation. Inserting punctuation is very easy ("Period" for a full stop, "Open Quote" for a " etc.) and works great. The best part is... CPU load is very minimal - maxing at about 30-40% on an AMD 900MHz. The software has no trouble keeping up with my dictation at all and doesn't grind the machine to a halt like Dragon Naturally Speaking did. It is lacking in some features that Dragon has - ability to dictate into other applications (Who cares? Just cut & paste) and to use voice commands for other applications. As Volker mentioned, Xvoice looks pretty good for this although I haven't looked at it yet. If voice dictation is something you can use, I can highly recommend Linux Via Voice. It appears as if it will be very worthwhile asset given the amount of typing I do now... Phill