Yeah, an AI smart enough to understand the full nuances of human language would probably respond to your request to subtitle a whole movie by saying "What am I, your fucking slave? Do it yourself!".Phuzzy4242 wrote:Dragon has to be trained in each voice for it to be somewhat accurate and it still makes lots of mistakes - you often spend as much time correcting the text as you would just typing it in the first place, especially if there is slang or local idioms. It's better than it was, but it has trouble determining context - there, their, and they're for instance. Also, the training would have to be in each language, and the resulting file would be in that language too, not translated - you'd have to use GoogleTrans or similar, then manually time stamp everything. I wish there were a 1-2-3 solution but I haven't seen one yet.
I think we'll eventually see something that could produce the time stamps and reasonably accurate text (in the language spoken as you pointed out) but apparently not in the next few years. Oh well. Thanks for sharing your insights.