Voice Recognition Software - Is it any good?

[police state]

I want to use voice recognition software for an application I’m involved in; the catch is that most of the words are non-standard English; think medical, scientific type stuff. Has anyone any experience of this kind of software? In an ideal world, I would want it to be 100% reliable, I realise this is likely impossible, but how close can I get, and what package would people recommend?

[Muncher]

It used to be quite popular 4-5 years back and I tried several different versions but ultimately they were never any quicker or easier than using a keyboard. They seem to have become less prominent lately so I guess they either never improved them sufficiently or decided it just wasn't a good idea.

[groomi]

When I was commuting in the car for 2 hours each way, I used to think about setting up a laptop on the passenger seat with a headset on to dictate memos or to-do lists.

Never got round to trying it out, but I did use the voice recorder on my mobile which I then typed up at home.

Still wonder if it would work/be useful...

[cjs]

I just don't understand why we are not further down the road with Voice Recognition, to me it seams the obvious route to take, not just for dictating letters but with the complete computer operation. Why we are still using a keyboard that was invented decades ago? They had voice recognition on Star Trek back in the sixties.....why haven't we got it yet!!

[Plotloss]

Bit of experience from the telephony side.

You can tune it to exceptional levels of recognition if you look at the high end commercial grade speech to text engines such as those used by large automated telephone systems.

Not at all cheap but exceptionally clever.

[pdV6]

The bottom line is (as ever): you get what you pay for.

It can be done (badly) on the cheap or rather well (for a lot more).

[SGirl]

I used Dragon NaturallySpeaking for a time a couple of years ago when I had loads of text to translate from hard copy. It's not bad at all. An ITI guy reviewed all the various packages out there and came to the conclusion that the Dragon package was the best.

It's certainly easier than typing it all out! And you can play back what you've said to double-check that what's on the page is what you said in the first place. It's not 100% perfect, obviously, but the more you train it, the better it gets. I'd estimate that by the time I'd finished working with it, I was getting accuracy of around 90%.

[Plotloss]

Max english recognition level for high end STT apps I think is around 96-98% depending on the content of the interaction.

German interesting (or not, depending on your leanings) can be tuned to 100% efficiency and TTS is practically bang on as well.

[morningside]

The dragon one is quite good but like most voice software it is the training of it that is most important and one of the hardest thing to do is KEEP speaking naturally when setting it up. This is easier said than done as you tend to start talking like a stinted dalek and keep looking at what is 'typed' instead of a continuous flow of speech, also they do NOT seem to like any thick accent.
I have never umbrella treetop goldfish problems with them

Edited by morningside on Wednesday 4th April 14:18

[irm]

use naturally speaking 9

ok, setting up was difficult (reading Kennedy’s inaugural speech) as i speak with an accent

ok if you accept its limitations, doesn’t do Yorkshire yet

AND PROOF READ CAREFULLY

[davidn]

Another vote for Dragons naturally Speaking. My staff use it to generate property inventory taken from a dictafone. Training it isn't too bad and it learns as you go so after a few translations it becomes around 80 to 90% accurate. Saves us stacks of time.
Agree you do need to proof read, especially if it's intended for a customer. I inadvertantly managed to get it to learn and correct 'brick built out house' to 'brick built shit house', don't know what I was thinking about at the time. Once rumbled I was embarassed and secretly amused to see how many times it had come up on previous documents. Caused a few queries amongst clients but then it's easy to blame the computer.
Cheers
David

Edited by davidn on Thursday 5th April 20:50