You can speak to a real person, but can you dictate to a person while they're typing?
I would imagine the weirder comparison is that I can type my thoughts into a box just fine, but speaking them makes them all jumbled. I guess it's because I'm used to comment boxes triggering my "let's write a comment" mode, so I would need some time to get used to dictating them.
>You can speak to a real person, but can you dictate to a person while they're typing?
Isn't that what secretaries and court-room clerks have been doing for over a century?
Besides businessmen, also lots of succesful (print) journalists and writers in the 30s to 80s had personal assistants that they dictated their articles etc to.
Court room clerks, assuming you mean stenographers, are trained in typing shorthand. Basically, this lets them type at upwards of 250 words/minute. Something far beyond untrained individuals.
For comparison, you are probably only able to type upwards of 80 to 90 words/minute using standard keyboards. And that is, again, with training.
There was a fun article recently that went over writing shorthand. Journalists used to be commonly trained in writing methods where they could get over 100 words/minute. I know of literally zero folks that are trained in that.
I do about 100wpm normally, but I also first laid hands on a computer keyboard at the age of six, and never really took them off again. I can manage 120wpm sustained with a negligible error rate, but it's a stretch. It's easy to imagine doing twice that with training in a restricted orthography that makes it possible to represent words with fewer glyphs - in other words, a shorthand.
The thought of 100wpm on paper, though, leaves me frankly amazed.
Not just shorthand - but they type with "chords" instead of letters. Typing "stop the press" is 3 chords instead of 14 key strokes which allows them to greatly increase the speed at which words are entered. As you mentioned though, it requires training.
Is it because you have to wait for an external cue to start your sentences? I recently had an experience where I had a whole SMS conversation with my wife (4-5 back and forth messages) without typing. And the thing that was jarring to me is that I had to press a button, wait for an interface cue, and then talk. All that waiting for external stimulus would often mess with my natural rhythm to speak.
A big part of this I think, is the one-way nature of just about every dictation system. I can talk to someone without having a concrete idea of what I actually want to convey, and figure it out mid-way into the conversation. Dictating to speech recognition is like carefully crafting a question to place on stack overflow.
Creating a document (memo or whatever) with dictation definitely has something of a learning curve. I haven't done it much for years but dictating something for someone else to type up (yes, I used to do that) wasn't something that you just did naturally. It has to be more linear than writing--especially creating a document on a computer where you routinely change text and rearrange elements.
Same here. I think the biggest problem is that you can't pause for an extended period of time to gather your thoughts, and then continue dictating to the device. On the iPhone it stops listening after a beat or two. Then you have to go through the whole process again.
I haven't had this problem when using the dictation functionality via the iPhone keyboard's button for it. Can you elaborate? If this is something I'm liable to run into, I would like to know how to avoid it, because I can see it would be extremely annoying.
> I guess dictation is a skill like anything else.
It is, and like any skill it improves with practice.
The primary difference I observe between typing and dictation is the latter requires more forethought and precision to produce a good result. When typing, it's easy to try out many lexical variations very quickly, because the friction of editing is extremely low. This is not true of dictation.
You may find it beneficial, as I have, to pause after each sentence and take a moment to formulate the next in general terms before continuing to speak. You need feel no urgency when dictating to a computer, and to have a general outline in mind of what you mean to say makes it much easier to do so without stumbling or hesitating partway through.
It also helps to treat dictation as a first draft, rather than a finished product. I gather that in the days when personal assistants still took dictation, they did so in order to type up a draft for review and potential correction by the person dictating the document. While we now largely dictate text to computers rather than other people, the method of treating dictation results as drafts for review still has much to recommend it.
A big part of this is that typing and dictation are very different from conversation. Conversation is free flow, very flexible to errors (since people can infer each other's true meaning) and interactive. However, in both typing and dictating, your end product is often meant to be read and so must be error free.
For dictating this means planning ahead to make sure that the sentence is well-formed. Have you watched yourself type? There's a lot of backtracking, erases and restarts. The feedback from dictating is also slower, which then affects your input rate. If you do manage to ignore this and speak free flow, even with perfect recognition (which is impossible even for humans), you will have to go back and edit grammatical and cohesion errors.
The advantages of typing are faster feedback (this is unavoidable as the computational needs will always be less), working memory augmentation and better nonlinearity for edits.
When you take into account all that, the time to correct grammatical and spelling errors, to edit for comprehensibility, I doubt that speaking will come out ahead because input bandwidth is only the bottleneck for trivial pieces.
As someone who hates typing on mobile, but also dislikes speaking on mobile even when it correctly and easily transcribes my voice, you're definitely on point with the "why".
With that said, i think voice can still be superior, but perhaps i'm biased. I think what we need is an improved UX. Voice can handle the free flow part, but imagine being able to easily modify the context of the conversation via a few simple buttons. Maybe an undo button based on some context aware value (ie, not just words), and etc. The goal being to augment our ability to correct mistakes, or change the flow of the conversation.
I'm a prime example of why voice "typing" is difficult, i tend to misspeak very frequently. When using my voice, i often need to pause - which then usually causes the voice program to stop transcribing /eyeroll. Regardless, i still want to see it succeed, and think it can overall.
The main problem with voice regognition for me is privacy. I mostly use my phone to write when I'm in public and wouldn't want everyone around me to hear what I'm doing.
I know a lot of people who used to get upset about kids ignoring conversations for their "texting machines" and those same people now will dictate a text message in the middle of a conversation and give me the look like IM the one ruining their text conversation.
Interesting. I wonder how that stacks up against retention. I mean, part of the weird inter-connetivity of the brain relates to vocabulary and speech but tying that together with motor skills (printing, cursive, or in some ways typing) are, I think, pretty established. I think dictation is useful, sure. However, if people end up writing the way they talk, I hope the software auto-crops "uhhhs" and some other errata.
Don't think that will work for me; when I try voice recognition it gets so many words wrong that you can't even guess what the original input was. I don't have a speech impediment but voice recognition neural nets sure think I do. Also I type faster than most people speak when they are not trying hard to speak as fast as possible. I bet that's the case for many people here.
At least when your typing, you get a word wrong and immediately fix it. With Voice Recognition, you have to parse back through the whole message and pick out the corrections one at a time, many corrections if it has a 20.4% error rate.
Humans don't do well with that, they look at something and their brain fills it in correctly for them, it's easy to miss typos in your own work. When I used to do helpdesk and someone had password problems I'd have them retype their usernames, because they would not notice it was wrong, but it was often the problem.
Actually, you are a better typer, generally, if you do not correct mistakes as you make them. Reasons being: a) the computer can often just fix it for you and b) you shouldn't be looking at where you are typing if you really want speed.
Dictating for voice recognition is a skill you can learn, just like typing.
More than a decade ago, I saw voice dictation working 100% perfectly. The guy who did that was being paid to do the demo by the software company that wrote the voice recognition program, and he had obviously spent a lot of hours doing demos.
He wasn't faking anything, he was just doing it really well.
Other people who have an incentive to learn it can usually do it pretty well too. They are mostly people who are using their hands for something else, eg doing autopsies, or who don't have hands.
Ok, but this comparison seems a bit unfair since there are tools to "type" words faster like Swipe-to-Type. I mean, it has been literally years since I typed continuously character by character in the old-fashion way (there are always words that are not part of the dictionary and you have to type them character by character).
I have even found that it's faster for me to switch the language of the keyboard and keep using Swipe-to-Type than typing in the classic way.
I feel like autocorrect has made me less accurate because it often picks the wrong word and I have to go back and fix it. I can't get used to swipe-to-type.
Do you know mechanography? I know some people that can't get used to it too and I think it's because they haven't learned by heart the distribution of the keys (something that someone capable of typing with both hands on a desktop keyboard knows)
"on a mobile device". Talk about tying your hands. I'm sure on a real keyboard with a competent typist, there's still no contest - the typist will win.
Also, let me throw out my standard response: "No one said..." that you have to use a keyboard or voice dictation, you have the option of using both solutions.
I'm really splitting hairs here, you have valid points but…I use a steno machine to write code all day. I know, I'm probably the only one. But steno was forever locked up in proprietary industry until only very, very recently when Plover came to be. There are more and more people learning it every day. I'm hopeful that it will be comparable to the number of people using Dragon and the like. Sure, it won't replace the QWERTY keyboard, but at least people won't say "a what?" when I tell them I write on a steno machine.
The chorded keyboard idea is definitely interesting but voice recognition is what most people will use.
People don't always want to simply type fast. Imagine other use case cases, like designers, for example. They might be using a Wacom tablet and need to reach for the keyboard to "create a new layer", "change font to San Francisco", "50x100 rectangle", etc. Wouldn't voice recognition be much easier?
Thanks for that. Yes, speech control seems really natural for certain tasks, like a painting program especially where you use few and distinct words. The VR painting programs come to mind.
That case study interests me, but it really glazes over stenotype and palantype. I'll email Bill and see if I can get a discussion going. Cheers
Hell, on my mobile device I type by writing down the first letter of each word and letting the autocomplete do the rest. I doubt you can get faster than that.
The problem at least on iOS is that at least the last time I tried it it doesn't recognize when I'm switching languages for some words. This is extremely annoying when asking for a song with an english song title.
Another issue is that it doesn't recognize my dialect. I don't want to talk differently just so a computer understands me.
And last (this is also an issue with autocorrect!) it doesn't like at all when I'm using nonce words. I feel like the cause is that they seem to be not as common in English as in German, and compound words are usually written as two or more words instead of one.
Hopefully they can use this for closed captioning.
A few years ago, for some work-related thing, I was looking at closed captioning for various TV channels. One channel had the most bizarre gibberish in there. What the heck... was there an error in decoding? But the words seemed like english words. Upon closer inspection, I realized that if you pronounce the gibberish sequence of words, the sounds made sense. Turns out it was a bad job of automated speech recognition.
Something I've wondered about is if it would help if I could talk and type at the same time and it could use that to correct errors by cross referencing the two data sources.
"Beats humans at typing" is a super misleading title. I'm a hobbyist stenographer and I'm always interested in advances in voice recognition. But this is just a little test without much merit, in my opinion.
A couple points:
- "Typing" isn't what I'd call whatever people do while tapping away at a touchscreen keyboard. There are different actions, including swiping and tapping, but without hardware, "typing" is misleading.
- Human typing speed is variable. Many people type below 50 words per minute, but others break the 100 words per minute barrier (or more in the case of prodigies like Sean Wrona).
- To ever use this for captioning or court reporting, we'd have to overcome stenography, which is exceeding 200 words per minute. We're on a totally different level now.
If anyone wants to learn about stenography or try it out on their own keyboard for fun, check out our open source efforts at http://opensteno.org
I use voice recognition for taking notes whilst on the move. Although good for short sentences, a lot of improvement is needed before i could, for example, dictate a document. Punctuation is an example of a weak area.
There are also certain words and phrases that it gets wrong much of the time. I have learnt to not use these when making voice notes. There's definitely a learning curve to use voice effectively.
A friend of mine will repeatedly text back and forth using voice recognition and then gets frustrated when responses come slower or dont answer the question than they would like.
I'm just like stop texting and call the person then!
True, but you miss the searchability of text if you call someone. Plus, calling interrupts people, and usually my text chatter isn't important enough for a phone call.
55 comments
[ 1.2 ms ] story [ 61.2 ms ] threadI can speak to real person with ease, but talking to a device makes my ideas jumbled.
I guess dictation is a skill like anything else.
I would imagine the weirder comparison is that I can type my thoughts into a box just fine, but speaking them makes them all jumbled. I guess it's because I'm used to comment boxes triggering my "let's write a comment" mode, so I would need some time to get used to dictating them.
Isn't that what secretaries and court-room clerks have been doing for over a century?
Besides businessmen, also lots of succesful (print) journalists and writers in the 30s to 80s had personal assistants that they dictated their articles etc to.
For comparison, you are probably only able to type upwards of 80 to 90 words/minute using standard keyboards. And that is, again, with training.
There was a fun article recently that went over writing shorthand. Journalists used to be commonly trained in writing methods where they could get over 100 words/minute. I know of literally zero folks that are trained in that.
The thought of 100wpm on paper, though, leaves me frankly amazed.
Why is it easy to imagine doubling your speed on a keyboard with shorthand, but imagining doubling your speed on paper with shorthand amazes you?
This is what it looks like:
https://en.wikipedia.org/wiki/Shorthand#/media/File:Eclectic...
[0] http://qwertysteno.com/Basics/Introduction.php
It is, and like any skill it improves with practice.
The primary difference I observe between typing and dictation is the latter requires more forethought and precision to produce a good result. When typing, it's easy to try out many lexical variations very quickly, because the friction of editing is extremely low. This is not true of dictation.
You may find it beneficial, as I have, to pause after each sentence and take a moment to formulate the next in general terms before continuing to speak. You need feel no urgency when dictating to a computer, and to have a general outline in mind of what you mean to say makes it much easier to do so without stumbling or hesitating partway through.
It also helps to treat dictation as a first draft, rather than a finished product. I gather that in the days when personal assistants still took dictation, they did so in order to type up a draft for review and potential correction by the person dictating the document. While we now largely dictate text to computers rather than other people, the method of treating dictation results as drafts for review still has much to recommend it.
For dictating this means planning ahead to make sure that the sentence is well-formed. Have you watched yourself type? There's a lot of backtracking, erases and restarts. The feedback from dictating is also slower, which then affects your input rate. If you do manage to ignore this and speak free flow, even with perfect recognition (which is impossible even for humans), you will have to go back and edit grammatical and cohesion errors.
The advantages of typing are faster feedback (this is unavoidable as the computational needs will always be less), working memory augmentation and better nonlinearity for edits.
When you take into account all that, the time to correct grammatical and spelling errors, to edit for comprehensibility, I doubt that speaking will come out ahead because input bandwidth is only the bottleneck for trivial pieces.
With that said, i think voice can still be superior, but perhaps i'm biased. I think what we need is an improved UX. Voice can handle the free flow part, but imagine being able to easily modify the context of the conversation via a few simple buttons. Maybe an undo button based on some context aware value (ie, not just words), and etc. The goal being to augment our ability to correct mistakes, or change the flow of the conversation.
I'm a prime example of why voice "typing" is difficult, i tend to misspeak very frequently. When using my voice, i often need to pause - which then usually causes the voice program to stop transcribing /eyeroll. Regardless, i still want to see it succeed, and think it can overall.
Humans don't do well with that, they look at something and their brain fills it in correctly for them, it's easy to miss typos in your own work. When I used to do helpdesk and someone had password problems I'd have them retype their usernames, because they would not notice it was wrong, but it was often the problem.
More than a decade ago, I saw voice dictation working 100% perfectly. The guy who did that was being paid to do the demo by the software company that wrote the voice recognition program, and he had obviously spent a lot of hours doing demos.
He wasn't faking anything, he was just doing it really well.
Other people who have an incentive to learn it can usually do it pretty well too. They are mostly people who are using their hands for something else, eg doing autopsies, or who don't have hands.
One that bugs me is proofreading. It seems like the tools in Microsoft word could do a lot more to catch and fix mistakes and with higher accuracy.
I have even found that it's faster for me to switch the language of the keyboard and keep using Swipe-to-Type than typing in the classic way.
How many people know steno? No one you know, but hey that's how we chew the fat on the Internet. How many of the 7 billion people can type?
People already effectively use dictation. John Siracusa used Dragon to write his 20,000 word Mac OS reviews:
http://arstechnica.com/apple/2013/10/os-x-10-9/23/
David Pogue has been using dictation for years:
https://youtu.be/x0GXX-SJuQM
Software developers have even cobbled together voice programming solutions:
http://ergoemacs.org/emacs/using_voice_to_code.html
https://www.extrahop.com/community/blog/2014/programming-by-...
http://web.stanford.edu/~ouster/cgi-bin/wrist.php - Creator of Tcl/Tk has had RSI for a couple decades.
Also, let me throw out my standard response: "No one said..." that you have to use a keyboard or voice dictation, you have the option of using both solutions.
People don't always want to simply type fast. Imagine other use case cases, like designers, for example. They might be using a Wacom tablet and need to reach for the keyboard to "create a new layer", "change font to San Francisco", "50x100 rectangle", etc. Wouldn't voice recognition be much easier?
Here's an interesting study on chorded keyboards:
http://www.billbuxton.com/input06.ChordKeyboards.pdf
That case study interests me, but it really glazes over stenotype and palantype. I'll email Bill and see if I can get a discussion going. Cheers
Another issue is that it doesn't recognize my dialect. I don't want to talk differently just so a computer understands me.
And last (this is also an issue with autocorrect!) it doesn't like at all when I'm using nonce words. I feel like the cause is that they seem to be not as common in English as in German, and compound words are usually written as two or more words instead of one.
I also have speakers with music playing
A few years ago, for some work-related thing, I was looking at closed captioning for various TV channels. One channel had the most bizarre gibberish in there. What the heck... was there an error in decoding? But the words seemed like english words. Upon closer inspection, I realized that if you pronounce the gibberish sequence of words, the sounds made sense. Turns out it was a bad job of automated speech recognition.
https://support.google.com/youtube/answer/6373554?hl=en-GB
It also produces amusing gibberish if the accents are strong enough.
A couple points:
- "Typing" isn't what I'd call whatever people do while tapping away at a touchscreen keyboard. There are different actions, including swiping and tapping, but without hardware, "typing" is misleading.
- Human typing speed is variable. Many people type below 50 words per minute, but others break the 100 words per minute barrier (or more in the case of prodigies like Sean Wrona).
- To ever use this for captioning or court reporting, we'd have to overcome stenography, which is exceeding 200 words per minute. We're on a totally different level now.
If anyone wants to learn about stenography or try it out on their own keyboard for fun, check out our open source efforts at http://opensteno.org
There are also certain words and phrases that it gets wrong much of the time. I have learnt to not use these when making voice notes. There's definitely a learning curve to use voice effectively.
I'm just like stop texting and call the person then!