Callum StonemanI've noticed that when using the #Microsoft OneCore or SAPI5 voices with #NVDA, dates in dd.mm.yy format are read incorrectly. For example, 01.01.26 would be read as "the first of January 1926". If the year is written as 2026 in this example, it will obviously be read out correctly, but if it's only two digits it will be read a century out. I'm using the Microsoft Hazel voice primarily, but it also happens with the SAPI5 voices from my (admittedly limited) testing. Doesn't happen with eSpeak. Anyone else experienced this? @NVAccess
@CallumStoneman @NVAccess Don't you just hate hate hate speech synths thinking they know best. Could somebody tell me who doctor congo is please? I really do think screen reader makers should take far more responsibiity for working on how their products sound, and try working with the speech synth companies rather than just saying it's an SEP!
James Scholes@CallumStoneman @NVAccess It's almost as if speech synthesises shouldn't be trying to interpret text that is itself ambiguous...
NV Access@jscholes @CallumStoneman We do report issue like this to Microsoft for their OneCore and SAPI 5 voices (and yes, we're still dealing with 2 Cuban Pesos of flour in cake recipes from time to time). It is frustrating and it's exactly why NVDA does NOT make those assumptions unless you tell it to (with the speech dictionary). You CAN use the speech dictionary to work around it but of course that is less than ideal do. Do keep writing to Microsoft and feel free to loop us in as well.
Seedy!@CallumStoneman @NVAccess Someone obviously didn't Y2K-test those voices lol