Dictation Software for Linux?

Yottabyte84 asks: "As a student, I often have the need to type up papers, however like most people I talk faster then I type. I've lately been looking to get a dictation program, but I don't wanna boot Windows every time I need to use it. IBM has a commercial version of ViaVoice for Linux and a free SDK. I'm unclear what the SDK can do, and would be willing to buy the commercial product if the SDK doesn't fit my needs. What I'd really like to be able to do is give spoken text input into the Linux apps I already use, but could live with being stuck with a simple included word processer. Have any of you tried ViaVoice or the SDK? How well did they work for you?"

Comments

I've used ViaVoice SDK and commercial versions

[slashkitty]

In a word, they are pretty good, but not perfect. The commercial version includes the teaching portion where you talk for like twenty minutes as it learns your voice. I've only tried it in it's own little app, which is not a real good word processor, but it's good for entering text. I would read paragraphs of Newspaper articles as fast as I could, and it was nearly perfect. It would miss names or things that you would expect. It could even play back the audio of what you just read in /Your Voice/ or it's own TTS engine. It wasn't as good when I tried to feed is other stuff. I guess it's geared to corporate / news speak.

I've also developed some test apps w/ the SDK. It's not as good for free text, but could handle special commands and vocabularies. Things like automating mp3 playing and turning on and off lights would be good for that. You should try it out if that's what you're looking for!

Re:I've used ViaVoice SDK and commercial versions

[slashkitty]

I haven't tried dragon's product, there is a review of it somewhere. I won the VV at LinuxWorld. I wouldn't have paid for it, when it comes down to it, I can type much better then I can speak. ;-) The SDK didn't work as well because of the learning, and maybe vocabulary? I think that if you have the commercial version, you can transfer the files to an SDK project, but I didn't try that.

The "ViaVoice Dictation Run Time Kit for for Linux V3" seems to be the same as the version you buy, minus the headset (which makes a hugh difference) ... it should at least have the free text dictation to try out.

The best long-term solution. . .

[foo fighter]

. . .is to learn how to type faster. Mavis Beacon Teaches Typing is a great tool for this, and it's also pretty fun. We use it to teach typing skills at the school where I work.

I've supported many administrative assistants and remember when voice to text software first came out. We tried a number of solutions, and for even moderate typing speeds (60-90wpm) we found they were much more efficient if they just typed, rather than tried to dictate.

I also remember recently the Director of our school was tired of dictating letters to tape to be transcribed by her assistant. (This was about 5-6 months ago). We tried Dragon Naturally Speaking and she didn't like it at all: typos, have to speak unnaturally, the weird feeling of talking to the computer (don't know why that was an issue since she talked to the tape. She's the boss).

Anyway. I'd invest some time in building your typing skills. It will have a higher payoff in the end.

problems?

[Roadmaster]

I had several issues while trying viavoice and dragon naturally speaking.

You definitely should give it a try before committing.

When you're typing, the kind of mistakes you make are of the kind known as "typos", where you mistype one letter for another. Those are hard to spot but in the end they don't matter that much.

With voice recognition, you have BIG mistakes; like, the engine mistaking one word for another totally different, or complete sentences. That is big, and actually requires you to go through the entire document and fix the errors.

Also, most voice recognizers i've seen aren't too adept at handling changes of mind; when you're speaking, you tend to make mistakes, because you changed your mind, or you meant something else, or whatnot. Recognizers aren't too forgiving of these situations. You'll end up with more crap on your document that you'll have to fix. In order to provide a coherent flow of speech for the recognizer, you basically have to be reading it from somewhere else. Which pretty much beats the purpose of the recognizer. If it's already printed, why not OCR it instead? it's bound to be much more accurate and faster. Also, if you have to write your stuff by hand before dictating it into the computer, you defeat the purpose of the voice recognition software. Because I guess you can type faster than you can handwrite :)

Re:problems?

[Rick the Red]

If it's already printed, why not OCR it instead?

It depends on the application. The *best* use I've yet seen for voice recognition (besides hands-free cellphone dialing, which is mostly just recognizing numbers) was on a loading dock. They got invoices and bills of lading from lots of sources, on lots of forms, with the information in lots of different places. They would just read the stuff they needed in the order their system needed it (shipper name, address, etc.; invoice number; part numbers; whatever). They got used to the various forms and could quickly find and read the desired info. I suppose you could teach an OCR program to recognize all the different forms, but what about the ones with coffee and jelly stains, handwritten forms, etc? Voice recognition was way cool for this app.

Re:speed speaker

[sydb]

How many mistypes do you make though? Doesn't matter during a game of quake if it coms out a byt wron gbut professional typists don't make mistakes. That's the 90wps.