Several weeks ago, I recorded some brief audio demonstrations of my dictation tool which allows people to dictate directly into the Leasey Word application. People have particularly welcomed this concept and have asked a number of questions about it. I want to answer some of those questions here.
I am doing this initially using my personal Mastodon account because that was where the demos originated.
Let's deal with some basics first.
The dictation feature within Leasey word will be available within the next major update, hopefully at the end of this month.
Dictating text into an application is not something you can just pick up and work with. If you want to become proficient at it and avoid correction, it does take some practice. The documentation relating to the best use of Leasey word's dictation feature runs to just over 30 pages. That is how seriously we take this subject. It's not a few paragraphs which have been thrown together. It is a discussion of this subject which gives useful advice, and it is suggested people should read it when it is available. Purchasers of the Catching Up with Leasey training course to begin this month will obtain some tuition on this feature as well.
We have chosen to use an online service for interpreting dictated text. The reason for this is accuracy. We prefer that the text being transcribed is as accurate as possible. Anyone can create an application using freely available local transcription models, and if people want to use those and hope for the best, good luck to them. We however are interested in serious voice recognition, and that means exceptionally accurate transcription. I would go as far as saying that individuals creating such solutions have not had sufficient experience in working with people using voice recognition and just implemented it because it seemed like a good idea at the time.
Having dealt with the basics, let's get down to some specifics.
Does the tool auto-punctuate?
Yes, at the moment, although that will change in the coming months. There are ways of controlling how it does punctuate though. Voice inflection certainly plays a large part in this process. However, you can force punctuation signs by speaking them, such as full-stop or period, comma, exclamation mark, etc.
Can you dictate lengthy passages without stopping?
Yes. That is not recommended when learning but it is certainly possible. There is also a feature called Live Transcription which is designed to capture lectures or meeting notes through transcribing what is being spoken on a continuous basis.
Does JAWS have to echo back each phrase as I dictate?
No. It can play a sound or have no feedback at all. If you do choose to have JAWS echo your text back, which if you cannot see is helpful in many situations, it will echo the punctuation signs even if the JAWS punctuation level is set to Some or None. This is because it is critical that you understand the punctuation signs which have been inserted, whether it has auto-punctuated or whether you have forced the recognition of punctuation signs.
Typically, how long does it take for a response to be processed if you speak a short phrase?
The average time on a good internet connection is 0.6 seconds.
It is possible to vary the processing time if you typically speak more slowly. For example, you can set it so that you could say, this is a very, then pause, and then say, good test. That would then be interpreted as a constructed sentence. So we're not talking here about the time interval between you speaking a phrase and JAWS echoing it back. What we are discussing however is how fast that audio processing is. If you speak relatively slowly, you do not want the system to interpret a part of speech as a complete sentence necessarily.
Can you speak commands to move the cursor to specific words or select them?
Yes you can do that. If you know the text you want to select or move to, you can say those words. The documentation describes those procedures.
What about dictating individual characters, including letters, numbers and punctuation signs?
You can do that too. In addition, with or without spell mode, you can indicate initial capitals or fully capitalised text. Spell Mode, like Dragon NaturallySpeaking, accepts letters, numbers and punctuation signs. Dictating signs used in Markdown is also possible even if Spell Mode is disabled.
What about inserting regularly used phrases and text passages stored as Leasey Clips or Leasey Texts?
Those too can be inserted with voice commands which can save an enormous amount of time. You can also use a voice command to copy the entire document to the clipboard, which brings us to the next important point.
Given that Leasey is a global application, why is dictation not available everywhere?
There will undoubtedly come a point where it will at least be available in Leasey applications themselves. There is an obvious next candidate within an app being worked on. But Leasey Word is a good starting point. Copying the whole document to the clipboard by voice does however mean it can be pasted elsewhere.
Can I Add custom words and phrases to improve the recognition accuracy?
The simple answer is yes, but we go a little further. One of the ways in which accuracy can be improved upon is if you give context as to what it is you are dictating about. There is a world of difference between the style you might use to dictate something to a friend as opposed to writing a formal document. So you can specify that on a document by document basis if needed. You do not have to do that but it can be helpful.
Can I dictate in Spanish?
Yes you can. The reason we chose Spanish was because one of our Beta Testers is a fluent Spanish speaker and there have been several requests for that. So you can speak Spanish, and then instantly switch to English if that is your preference.
In summary, accuracy is the name of the game here. Voice recognition is a serious business and recognition accuracy is by far the most important aspect of this whole area of working. There's no value in saving yourself time by dictating text only to have to spend an equal amount of time correcting misrecognition errors.
In all my years of working with Dragon, Windows Speech Recognition services, and other models, I have not seen anything which comes close to what Leasey Word is able to do. So when you get it, do try it out.