[ Date Index ] [ Thread Index ] [ <= Previous by date / thread ] [ Next by date / thread => ]
That gets the Archaeology/forensic response:
Store the original recording as it will be done better in a while.
And yes a first rough cut for keywords etc.
From Dr Adrian Midgley's hand
On 31 Jan 2014 15:23, "Julian Hall" <linux@xxxxxxxxxxxx> wrote:One other project in the pipeline is to transcribe voice recordings from local residents of their wartime memories. Of course that /does/ run head-on into the 'many voices/many words' scenario and training would be impossible as the recordings are already done. In this case I think the best way forward will be to listen to the recordings on headphones and repeat what they are saying myself for DNS to pick up.
Julian
On 31/01/14 07:01, Adrian Midgley wrote:
Sphinx.
Not production.
I played with ViaVoice a while back, didn't find it useful enough.
Difficult problem.
Recognising a small set of words said by anyone is fairly well solved. Wonky but workable.
Recognising many many words said by one person sort of solved requiring a lot of training. (Palm solved handwriting by deciding how people would write letters, I've not seen voice recognition approached that way but ...)
Recognising many words said by many people - not solved.
>From Dr Adrian Midgley's hand
--
The Mailing List for the Devon & Cornwall LUG
http://mailman.dclug.org.uk/listinfo/list
FAQ: http://www.dcglug.org.uk/listfaq
--
The Mailing List for the Devon & Cornwall LUG
http://mailman.dclug.org.uk/listinfo/list
FAQ: http://www.dcglug.org.uk/listfaq
-- The Mailing List for the Devon & Cornwall LUG http://mailman.dclug.org.uk/listinfo/list FAQ: http://www.dcglug.org.uk/listfaq