21–27 of 27 posts
Assuming this does not yet exist: If anyone is interested in building something like this (that works offline) please let me know, I'd like to join. I have some experience building web extensions but unfortunately not the time to develop & maintain this on my own right now.
Assuming this does not yet exist: If anyone is interested in building something like this (that works offline) please let me know, I'd like to join. I have some experience building web extensions but unfortunately not the time to develop & maintain this on my own right now.
All you need is a CSS selector for the aria label and a click action. Once the voice transcription is done it’s 1 line of JavaScript. You’d have to rely on the aria labels though.
Earlier quoted context omitted.
Dragon (dictate/NaturallySpeaking/professional/whatever they are calling it this year) has the ability to do this in conjunction with their official browser add-on, but it is an absolutely miserable product otherwise and I would not recommend it. I can say the words "click reply" and this message will be sent.
as a multi-decade user of the various forms of Dragon, it's not a miserable product. The company is a bunch of ass holes now owned by another bunch of ass holes. But for day-to-day use, it's not bad assuming a Windows environment doesn't mess with it. Most people go south with Dragon when they expect to be of the speak the keyboard or expected to be a perfect transcription. The reality is, it's like dictating to a go…
- Running the accuracy tuning tool more than once actually degrades accuracy
- User profiles will degrade over time, necessitating a export/new profile/import process every few months or else the accuracy becomes unacceptable
- Does not work with electron applications that apparently fucking everything has to be nowadays
- The browser plug-in… Dear God. It outright does not function properly on many top websites (Text input on twitter and twitch come to mind immediately), and it occasionally causes Dragon to slow down to an infuriating degree. Sometimes it actually causes certain sites to break!
- Unstable. Crashes to desktop a couple of times per week, and leaves some appendage running in such a way that it can't be directly restarted. I had to write a PowerShell script (https://gist.github.com/Karunamon/f2b684d083fc341ee79aab1fb2...) and put it on my start menu to deal with this.
Earlier quoted context omitted.
as a multi-decade user of the various forms of Dragon, it's not a miserable product. The company is a bunch of ass holes now owned by another bunch of ass holes. But for day-to-day use, it's not bad assuming a Windows environment doesn't mess with it. Most people go south with Dragon when they expect to be of the speak the keyboard or expected to be a perfect transcription. The reality is, it's like dictating to a go…
On the contrary, my issues with Dragon are not surrounding its recognition accuracy (though I would kill to be able to hook it up to whisper), but literally everything else. - Running the accuracy tuning tool more than once actually degrades accuracy - User profiles will degrade over time, necessitating a export/new profile/import process every few months or else the accuracy becomes unacceptable - Does not work with…
1) anyone who knows anything never runs accuracy tuning tool. 2) apparently user profile degrading is greatly reduced in version 16. Don't really know because I haven't migrated yet. 3) electron is a flocking abomination. I blame JavaScript and the HTML Dom. 4) I would not entirely blame the browser plug-ins. There is a boatload of JavaScript out there that does everything it can to fark up dragons need to look into a text area so can decide how to make things work. 5) that sounds more like a Windows, may be hardware problem. I haven't had a Dragon related crash (15.61) for more than a couple of years. Windows is a piece of shit but it's crash rate is also dropped.
re: wisper oh I would love to have a good whisper environment on Linux that's a reasonable approximation of Dragon. What that means is ability to dictate into any window anywhere, the ability to edit the content in the window, and an extension that lets you add a grammar for commands (preferably based on Python)
then I would be able to ditch Windows and tell Nuance to fark off
Earlier quoted context omitted.
All you need is a CSS selector for the aria label and a click action. Once the voice transcription is done it’s 1 line of JavaScript. You’d have to rely on the aria labels though.
It's not quite that simple. For example the button doesn't necessarily have to have an aria label. Could be the (nested!) content. There could be multiple candidates (buttons). The speech recognition thing is also not trivial. Yes there are WebAssembly based solutions but those have to be properly integrated and tested (not sure how good they actually are)
1. look in a object we had in which a bunch of DOM objects have been identified via CSS selectors of XPaths to see if there was an object named 'home' if that not exist
2. look for an element with the id home if that not exist
3. look for an element with the class home - take the first one. If that not exist
4. look for an element in which the text node was 'home'
There were a few other heuristics to determine what to do also dependent on if you said I select home, I click home, I move the mouse over home etc.