Almost all the software in the repositories only does text to speech
but a search on transcription turned up "transcribe" which claims to
do what you want. Disclaimer: I've never used that or anything similar
so can't vouch for it's usefulness.

I used a piece of software called 'Viavoice' which came with an early version
of Suse linux back in the 90's it worked ok then but i haven't tried it since,

There is another you may take a look at called Sphinx-4 a Java application
used in robotics for speech recognition.

I am
investigating Sphinx2 albeit fairly slowly. There is a genuine
reason for this query, have my right hand/wrist in a splint.
A damaged nerve in forearm that could take months
to come right -if ever. One handed typing and mousing
is very slow.

I believe Viavioce was first put out by IBM for their 'OS/2" os. I
use it back in the day. It was
great. Didn't know it was made for Linux. If it is as good now as it was
then, I will have to try it.

I have not looked at Viavoice yet but a search turned up a
mass of Sphinx material. While I investigate this and try to
get something running I have reactivated Dragon on the
MS half of the laptop so I can at least get some editing done.

IBM sold Viavoice to Scansoft about eight years ago. Under
their new name of Nuance, this organisation now markets the voice
recognition software 'Dragon Naturally Speaking".

I have been pestering their local agent (she agrees with me) for Nuance
to produce a version for Linux, but so far no success.

I thought IBM had sold it with the OS to the company that markets it
now. The name begins with an S. I
think Ser.... some thing. Last time I use OS/2 was around 2000. Still got my
disk of OS/2 Warp 4. The blue box, cd-rom.
Got the OS with the three 3.5 disk, the bounus disk, books and all. Had an
computer that was able to run it but it died.
May look for another.

I investigated Voice to text software back in 1983. It was 98% Accurate they
claimed, but actually got about one in ten words wrong. So, always required
heavy editing.

I never saw anyone say it has improved. I'd be interested to know if they've got
the accuracy up on that stuff and if so to what? I had an idea how to improve it
but never got the resources to develop my theory.

The copy I have of Dragon is several years old but does
much better than 10% error. Can do complete paragraphs with
line feeds and punctuation and no mistakes.

It does guess on new words but I'm teaching it to recognise those
and in my present one handed situation is well worth using.

Thanks for the report, good to know. I may end up implementing my idea anyway
for a new application because of economic issues and speed. How fast is Dargon?