Voice recognition software for translation/long-form writing

April 21, 2018 9:29 AM

So, I'm sort of an all purpose translator/transcriber/sometimes writer/language person, bottom line being I do a lot of typing in lots of different formats, typing that involves a lot of formatting, editing, and stopping. The recent advances in microphones and voice recognition (Siri/Alexa, Google Translate voice recognition, dayum!) have me wondering...would it be economical to use it to replace or supplement typing? What are your experiences with this? Snowflake & hardware details inside.

I like my tech dumb, reliable, and extensible, but I'm comfortable wading into exotic waters (TRADOS anyone?). My daily drivers are an i7 16g RAM Win10 Yoga900 and a couple random Chinese-brand Androids I've mind-melded Google Services Framework onto+fleet of older laptops I use for smart TV/music/projector terminals & backup machines & video/audio CPU processing set & forget type stuff.

I hate wearing headphones and usually have music & podcasts on (hell, sometimes at the same time!), but I usually work from home, so I'll adjust if I have to, and when I don't work from home, I'm definitely in headphones. I also have like...50 tabs and 6 documents on two monitors open right now. I know I can cut down on that if memory requires.

I will pay for software, even hardware, even a new computer! But, given how so many of the voice recognition solutions out there seem to be cloud-based, is that necessary? Up to now I haven't even used voice commands or search on my gadgets because I'm so comfortable in text, plus all the spooky spyware out there, plus I just said "Hey Cortana" and it tried to open my webcam, buggy little bastard, but Microsoft, Google, and Amazon already know everything about me anyway, and app-specific permissions are a thing now.

I'm toying around with Voice Typing in Google Docs tonight, but usually I use Word (autocorrect ftw).

Are we in the voice-capable future yet? How does speech recognition and voice dictation fit into your life?

posted by saysthis to Computers & Internet (9 answers total)

I gather John Birmingham has written a number of books using Dragon Dictation, so it's definitely A Thing, and I don't think it's cloud-based.
posted by pompomtom at 10:03 AM on April 21, 2018

My experience is that unless you're an extremely poor typist or have a disability that prevents you from typing, speech recognition software is going to be much less efficient and much more frustrating than typing once you factor in all the time you'll spend correcting the results (and all the subtle errors that both you and spell-check will miss).
posted by trig at 10:45 AM on April 21, 2018 [2 favorites]

It's not my personal cup of tea, but I have an acquaintance who does a lot of writing and editing and has switched over almost exclusively to Google voicetyping. That's mostly for text generation, though -- the modern equivalent of dictating a book to a stenographer or using a dictaphone for memos.

It's been a minute since I've used good ol' Trados, but I can't imagine it playing well with... well, much of anything else, haha. A quick google suggests that it does support Dragon (at least, some versions of Trados support some versions of Dragon -- looks like 2017 may not support NaturallySpeaking, boo). Depending on how you work, you could potentially try speaking your translations into a Google doc and pasting the result into Trados to edit. If your jobs involve a lot of custom dictionary use, that might be a pain, though.
posted by halation at 10:50 AM on April 21, 2018

I help setup transcripts for the Mefi Podcast, and you can see the results of the automated attempts. You could get better results by working interactively, repeating what those get wrong, and it's likely better as you won't talk overlapping someone else, but it's definitely not 100% flawless right now.
posted by Pronoiac at 11:55 AM on April 21, 2018

Translator here; I use a combination of a translation environment tool (Déjà Vu X3) and Dragon Naturally Speaking voice recognition software. Both are highly recommended for boosting translation productivity. I dislike headphones as well, but use a Logitech H360 Headset resting on top of my shoulders with the microphone shoved up to within 2 to 3 inches of my mouth. The accuracy of the Dragon software is astonishing, even though I am not using the very latest version. You can get a lot more information on software and hardware for translators on the proz.com website.
posted by delphic at 12:23 PM on April 21, 2018 [1 favorite]

Dragon has been amazing for almost 10 years now, since about version 4. Windows computers were crap at the time so you can imagine the technology has advanced since then. I've found that it was highly dependent on good microphones when I last used it a few years ago.

You have to train your brain to use it, and voice recognition historically disliked some accents (American South for instance) but for direct dictation without a ton of formatting commands, its great. I used it for stream of consciousness writing - no sure of the exact requirements of your translation work, but it may be worth trying.
posted by twoplussix at 2:14 PM on April 21, 2018

Hilarious side effect for me of training myself to dictate to a robot for hours on end was that I sometimes had to then suppress the urge to speak the punctuation when talking to people, especially when leaving voicemail or otherwise speaking deliberately.
posted by twoplussix at 2:17 PM on April 21, 2018 [2 favorites]

Have used the Dragon dictation software and was surprised how well a $30 product (I did not buy the newest version) could work.
posted by forkisbetter at 3:43 PM on April 21, 2018

For transcription, Trint works well, though it is cloud-based. It automatically transcribes a recording. I've had some interviews come back flawless, with no mistakes, and it even does surprisingly well with some strong accents, though it is not perfect. Very useful when I need something done quickly (generally, I get a transcript back within a few minutes).
posted by pinochiette at 5:51 PM on April 21, 2018