I just had a quick try of Talon on Windows, using wav2letter (I don't have Dragon) and the recommended scripts[0]. It... doesn't work well for me at all.
I was able to get a couple of simple commands to work in Chrome, sometimes, such as "reload" and "show history". In Visual Studio code, it just spouted a bunch of errors in the console [1], and in JetBrains Rider all it would do it type gobbledygook, like a cat had walked on the keyboard or something. Pretty dissapointing :(
The logs also fill up with "WARNING actions: skipped because they have no matching declaration: (user.select_next_token)".
It was a bit confusing to use too (apart from not really working, I mean!), as it wasn't clear if I had to use some kind of command to enable voice commands, or if it was litening all the time. Eventually I figured out that it seems to be the latter, but still, it's not clear what commands it has heard and understood - I found myself speaking and nothing was happening, and I had no idea what it had understood. Similarly, I'd say something like "close tab", and it would type some nonsense like "aa&" into the current file - again, no idea what command it was actually trying to use.
I recommend asking about hiccups on the Slack [1]. My basic analysis is you're missing a vscode-side plugin, you might need to restart Talon once after dropping in knausj_talon, and that your statement of "it doesn't really work" comes mostly from you guessing command phrases that don't exist. It's a strict command system - you need to learn/know the commands, you can't say just anything and expect it to work.
It's a tool to be learned and practiced, it's not fully optimized for out of the box experience (yet), currently more optimized for customization and total control by people who have the time and motivation to go hands free (e.g. due to limited motor function).
This is what it can look like if you practice a bit: [2]
---
Some recommendations:
- say "say hello world"
- say "help alphabet"
- say "help context"
- say "command history"
- say "dictation mode" then speak freely, then say "command mode"
I wasn't attempting to use it as a support forum, more venting after trying it, as I was hoping for something that Just Works(TM) at least for the basic, without me having to spend weeks hacking around with it. I'm quite happy to spend some more time tuning and customising, but only if I get the basics working easily, I'm sure it understands me, and I'm sure it's going to worth the time.
I'm hope you don't think I'm trying to be mean, it was just a really disappointing first experience. Might be my expectations were misaligned.
I had actually already installed the VSCode plugin and restarted both VSCode and Talon (I don't remember if I saw it in a comment in a .talon file, or if I saw it in the console logs, but somewhere it told me to install a plugin). Similarly, I installed the Rider/Idea.
I wasn't quite just saying anything :) Tho it wasn't clear what commands were actually accepted; from looking in the .talon files, I didn't see anything like a string literal, more something code-like. I had to guess at what commands were supported, for example "reload" in Chrome (assuming that's actually a correct command, even that simple command only worked some of the time).
I'm willing to give it another go, but is there a getting starting guide/tutorial for how to get started using it, and how to see what it's actually trying to do when it does something? I used the getting starting guide on the Talon site, but that only tells me how to install it, not actually how to use it.
I edited the parent since you started writing this post, maybe re-read the parent as I preempted some of the questions upthread.
For browser stuff, I'm actually really disappointed with how the "actions implemented in talon files" feature turned out in practice (which is the code-like action(...): syntax you saw), and I'm planning to deprecate it, which should clear that up a bit. Browser commands come from places like generic_browser.talon and tabs.talon. Looks like reload is "reload it"
You can get significantly more insight into what's happening by both saying "command history", and opening the repl and running `events.tail()` in it.
There's also a much more accurate speech engine in beta right now, which will be released soon, but I suspect most of the confusion wasn't accuracy related.
Thanks, appreciate the extra information - I plan on giving it another go tomorrow, armed with this new info!
Something else I wanted to ask about - does the voice recognition engine (either wav2letter or the new one you mention) adapt/learn according to the individual using it? I have a fairly strong Scottish accent, and would prefer to speak naturally if possible.
First - online learning is generally not necessary with my Conformer model (which is generally >5x better at not making errors than the current public model). It's also quite good at accents. My overall goal is to ship speech models so good you won't actually wish for model training.
I view fully automatic online training as a sort of anti-pattern - Dragon does that and it will randomly forget entire words. Talon may eventually have some kind of process for self-serve model training. I do have some plans for what that might look like.
Even without automatic model training there's already a feature to automatically create a sort of "personal dataset" as you use Talon, which you can use to train speech models (Talon or otherwise) down the line, or even send me to improve the main model.
Yeah, my brief experience and impression with these voice assisted coding is they require $300 microphone and a quiet room to get acceptable level of accuracy.
It probably is worth for physically impaired people (but i fear what 6hrs daily of this will do to their vocal cord). I am more interested in BCI technology which is where i see the future.
I'm a non-native speaker and have been succesfully using Talon in an open-office environment with a $20 mic - just to offer a different point of view. I've been using it for coding, not so much dictation so YMMV, but expensive mics, library environments or American news anchor accents are absolutely not required.
Talon doesn't require a $300 mic or a completely silent room. It works fine on my macbook air's builtin mic, and explicitly handles some interesting background noise scenarios (e.g. loud music, sitting next to a running dryer).
The author of talon doesnt support homebrew installation. Other people try to add it, but it doesn't work to install that way. Have you tried the installer on the website?
I was able to get a couple of simple commands to work in Chrome, sometimes, such as "reload" and "show history". In Visual Studio code, it just spouted a bunch of errors in the console [1], and in JetBrains Rider all it would do it type gobbledygook, like a cat had walked on the keyboard or something. Pretty dissapointing :(
The logs also fill up with "WARNING actions: skipped because they have no matching declaration: (user.select_next_token)".
It was a bit confusing to use too (apart from not really working, I mean!), as it wasn't clear if I had to use some kind of command to enable voice commands, or if it was litening all the time. Eventually I figured out that it seems to be the latter, but still, it's not clear what commands it has heard and understood - I found myself speaking and nothing was happening, and I had no idea what it had understood. Similarly, I'd say something like "close tab", and it would type some nonsense like "aa&" into the current file - again, no idea what command it was actually trying to use.
[0] https://github.com/knausj85/knausj_talon [1] "No such file or directory: 'C:\\Users\\MyUser\\AppData\\Local\\Temp\\vscode-port'"