[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Emacspeak] Swiftmac 2: pre-PR request for testers



Voice variations are definitely an issue with the so-called "modern
voices".

1. In my experience, using completely different voices, vs variations
   on a voice is too distracting.
2. I've compared  that  in the past to the "ransom note phenomenon"
   where one uses glyphs from different fonts on the same page.
3. needless to say, the above is known to be distracting in the visual world.
4. Switching voices like that mid-stream will also break intonation
   and likely add latency.
5. Piper (neural-net synthesis ) has "multi-voice models" but they
   tehmselves acknowledge that this reduces quality.

   But all in all I'm really glad to see you investigating this issue
   -- rather than getting stuck in issues of setup.

Robert Melton writes:
 > Yep, that is a good idea. I been focusing on trying to just make it work, now
 > I have to dig into the specific usage of it in emacspeak. But with this and 
 > hopefully my beepcaps fix, I think that will make the implementation feature
 > complete. 
 > 
 > Will still have to work on voices some as the only tools I have to really work 
 > with are:
 > - literal voice, as of Swiftmac 2.0 the cost of switching is a non-issue
 > - pitch multiplier, between 0.5 and 2.0 
 > - rate between .1 and 1 (which is a MacOS value of 0% to 100%)
 > 
 > I want to experiment with like using a completely different voice for things 
 > in quotes (like Alex for regular, Samatha for in quotes) and have lots of 
 > other things I would like to build up on the lisp side once I lock down this 
 > final implementation, so I can work on the more fun tweaky bits. 
 > 
 > > On Mar 27, 2024, at 14:16, T.V Raman <raman@xxxxxxxxxx> wrote:
 > > 
 > > Also, re spatialization and latency:
 > > 
 > > 1. You need only spatialize the Notification Stream.
 > > 2. So as with Linux, run two server processes,
 > >   A. The main seaker process, no spatialization, no extra latency
 > >   B. The notification process, spatialized, some extra latency but wont
 > >      be noticed, since it's not used much.
 > > --
 > 
 > --
 > Robert "robertmeta" Melton
 > lists@xxxxxxxxxxxxxxxx

-- 


|Full archive May 1995 - present by Year|Search the archive|


If you have questions about this archive or had problems using it, please contact us.

Contact Info Page