I was recently in a study skills workshop where speech to text software was discussed as helpful if you can say words more easily than you can type them.

Obviously being mainstream education, there were recommendations for Windows and macOS, but nothing else.

I want to go back to the accessibility team with suggestions, because if I complain they didn't have any they'll most likely say something insincere and then never do anything about that.

So, if you have #recommendations for free or low cost Speech to Text software for #Linux users, please could you share them with me?

@kittyboy77
cc @NatalyaD can you help?

@xanna @kittyboy77

Off the top of my head I don't know of any speech to text stuff for Linux. One issue is that many win/Mac OS systems just use the same back-end as Nuance's Dragon. Or they're some kind of cloudy thing where your speech data is going to ??where and almost certainly all involving AI-training now.

Speech recognition is hard to do well and I've not yet come across a Linux program which had the necessary £££ and time investment.

I will boost your OP tho in case that helps.

@NatalyaD @xanna @kittyboy77

I'm using Talon for speech to text on Linux now.

It can do a ton of other voice-control stuff too, potentially in conjunction with an eye tracker or pedals, so you don't necessarily need to use your hands whatsoever. In using it purely for dictation, I'm only really scratching the surface of its capabilities.

You have to be at least a little bit techie to get it up and running, and you pretty much have to be on the Slack if you need any help. I wouldn't be surprised if the documentation improved in future - it's early days yet and the developer is clearly working hard on the software itself. Meanwhile he does answer questions impressively fast on the Slack, as do some power users.

(There _is_ a fair amount of documentation already, inc for abundant customisation options - but it's not at all beginner-level. My heart sinks at the amount of background reading I'd need to do to understand it! So I've done very little customising so far.)

The speech model is imported, and unlike Dragon, it doesn't try to train itself from you correcting your text: it only changes interpretation via being explicitly told stuff, to the effect of e.g. "this name exists" or "what you've been guessing as bleep should be blorp". After about a month of adding new correction-instructions to it, and me getting used to it, I was finding it reasonably comparable to Dragon in terms of dictation accuracy.

I actually kind of prefer Talon to Dragon now, because my Dragon would crash a lot, which outweighed some of its bells and whistles, and Talon's been really stable. And I like the clarity of the correction-instructions (even if they don't _always_ catch what I meant) versus the trying to "train" Dragon on things it might or might not learn to get.

Talon's not compatible with Wayland, and, I gather, won't ever be, because Wayland intrinsically doesn't support the way it works. So you need a Linux distro which can do the older one, ISTR it's called X or X-something. As I understand it, this factor could be a question-mark over Talon's long-term future on Linux, depending on how long the older one remains supported (or substitutable). Can't forecast that aspect because I don't fully understand the relevant Linux landscape.

I'm still on the free version of Talon, and there's a paid beta with more things and integrations, including being able to teach it a variety of pops and clicks and suchlike!

HTH

#Talon

@unchartedworlds @NatalyaD @xanna Thank you. That doesn't sound like an option for me to try, but it does seem like it could work well for someone with more tech skills. I will definitely pass it on

@kittyboy77 @NatalyaD @xanna

I might've given the wrong impression then - I think anyone with reasonable learning abilities could use it to dictate!

(Not saying anyone at all, because that would include e.g. people with dementia or learning disabilities who potentially could not, but like you're posting on Fedi and it's not much harder than that.)

It's just they'd want someone with a bit of techie knowledge to get it set up for them and do any troubleshooting.