NVDA 64-bit migration breaking older voices, very long (2 posts)
The 64-bit NVDA migration is obviously going to mean a lot of older speech synthesizers (such as SAPI4 voices) and braille displays relying on drivers that use 32-bit libraries will stop working. I was really happy to see the issue was being taken so seriously that there were/still are plans to make a compatibility layer, and this was considered a blocking issue, meaning NVDA 2026.1 would not be released without this being addressed. You can read that issue here:github.com/nvaccess/nvda/issues/19030#issuecomment-3588231868
And then it was "addressed"… By removing any mention of SAPI 4 and other legacy drivers. There are still plans to return to this, and I sincerely hope this happens, as part of work on a different somewhat controversial topic - a new addon environment, with a more restricted API access which will be running in a separate process.
Now, anyone who’s already using Espeak or Microsoft voices with a modern braille display is probably going to just shrug, but this change is going to have some very serious implications that I suspect may cause some people to skip these newer NVDA releases. From my comment on the GitHub issue:
Many people still rely on older speech synthesizers for various reasons. There are languages where the only good speech options are older legacy synthesizers, and many people who use specific voices due to a hearing impairment, because they are more intelligible. For these people using Espeak, OneCore or purchasing a commercial speech addon may simply not be an option.
The only way out will be for addon developers to come up with their own bridging solutions, assuming an addon for the synthesizer is even still being supported. Eloquence already has at least 1 project to do this, and I’d be really curious to hear from people who use it if performance is worse as a result, whether the supposed performance gains are lost due to using a bridged synthesizer. It’s possible to make a fast screen reader with bridged synths — Take ZDSR, which is a 64-bit app but with support for SAPI 4 and 32-bit voices, and it’s blazing fast, faster than NVDA. But no amount of bridged addons will help anyone using a SAPI 4 voice which currently doesn’t have an addon. I’m lucky enough that for Polish, an amazing community sprung up around RHVoice, an open-source synthesizer that now has a number of fast and intelligible voices, but other languages are not so lucky. (1/2)
