Message ID
DKIM signature
Download raw message
Hi John, I haven't forgotten about your last message but I've been running into a lot of system and or speech recognition trouble so I've been trying to figure that out. A lot of my dictation has been misspelled such as 'sometihng' or 'suspenda nd hibernate' and I've been getting a lot of random words in front of what I say such as 'the', 'and the', 'a', etc. I'm reinstalling numen to see if that fixes the issue and noticed go from apt package manager is version 1.18 and numen now requires 1.19. So I deleted apt go and downloaded go from the official website to get the newest version but when I run ./install-dotool.sh inside of numen directory I get 'you need go (sometimes packaged as golang)'. What directory does go need to be installed in for this installer to work properly? command -v go gives me the output '/usr/local/go/bin/go'.

Sent with Proton Mail secure email.
Message ID
<Fs0K7vSI67QbICQw7LmoaX5xYD4ZsIXQHgi_0l8dhn805zInlfih03IMcZY7b1FK0HzxvuB0lK_YCk4AQd7uYBL9qQdaGY4vJJzHxgRNomM=@protonmail.com> (view parent)
DKIM signature
Download raw message
Hi sublime.

I think the go problem is /usr/local/go/bin isn't in the root user's path.
You could try making a symlink to it in /usr/local/bin like:
	sudo ln -s /usr/local/go/bin /usr/local/bin
Or straight up copying the file:
	sudo cp /usr/local/go/bin/go /usr/local/bin

That said, I wouldn't worry about the version if still compiles with the
older version.

I'm not sure why transcriptions would be mistyped like that.  Do you
think it's specific programs? Has it been happening in tiles-terminal
and the like?  I have reinstalled everything using the scripts to see if
I notice anything. When I run `numen --version` (a new addition) I get:

Getting 'the' and 'a' before your transcription has always been an issue.
I think the language model is trying too hard to turn what your snippet of
words into a sentence.  I see the vosk team is working on support a fancy new model
so we'll see if that helps.  One day I might try tweaking a model for numen
but not soon.  I guess as a hack we could also have a phrase that strips
any articles at the start, but not ideal.  I'm also planning on improving
the implementation of "scribe" because I've recently been having issues with
"sun" starting transcription.
Reply to thread Export thread (mbox)