Post by Verbivore on Feb 26, 2022 5:02:36 GMT
Much of the available voice-to-text software is American, and the Americans don’t say “varze”, of course. The software may well have correctly written vase if you’d said “vayze”.
Whenever I update software that might include a resetting of defaults, thereby making it incompatible with my typographic preferences and pedantry (wordprocessing, page-layout, etc.), I get a fit of Control Freak during which I check and reset all my preferences, stylesheets, and dictionaries.
Features that always seem to revert to a factory default of ON can include spellcheck-on-the-go; autocorrect; language selection ….
I can select an Australian spelling dictionary and grammar checker yet manage thereby to confuse the host software (e.g. MS Word – horrid PoS!): it doesn't know if/when I need ~ise or ~ize; practice or practise; it completely fails to deal well with punctuation of any compound sentences … .
It does give me the irrits when, even after I've selected the appropriate geo-linguistic-metrics from those offered, the damn machine will throw up a weird glyph from another alphabet; it will fail to recognise a legitimate alternative spelling; it can change alphabets. It will change measures from one's preferred units to the factory settings; and reset those damn green and red squiggles highlighting perfectly legitimate words, phrasings, and punctuations. It's no wonder the English-speaking world is increasingly confused when digital squiggles question a writer's correct work and sabotage their self-confidence.
Before there was voice-to-text (V2T) I did a lot of work with early-ish OCR software. When I started (ca '94), the successful recognition / interpretation rate wouldn't better about 65–70 per cent (assuming good, clean, original). By the time I quit using OCR (ca 2016) it had advanced to upward of 85 per cent, occasionally hitting 90%+.
One can only hope for similar progress with V2T, given its already widespread employment by professional agencies – including, perhaps especially, news services. (No Aunties mentioned by name. lol)
Automatic subtitle setters / translators based on V2T also manage to make some impenetrable or risible content – to one's occasional amusement.