Re: [Emacspeak] Experimenting with multiple voices for mac/windows

[Robert Melton <lists@xxxxxxxxxxxxxxxx>]

Bill--

So, thanks for pointing me in the direction of NVDA, as I learned two 
important things. 

#1. They use a different library than I was using, 
which appears to be the critical difference. There are 3 Speech Synth 
engines on Windows.

#2. It is relatively easy to link to NVDA and use it if it is present 
on the system and running, I think I can even avoid double-speak by 
doing it. 

So, win, win -- should have a new version out tomorrow hopefully with 
more complete SSML support. 

> On Apr 14, 2024, at 21:36, Robert Melton <lists@xxxxxxxxxxxxxxxx> wrote:
> 
> Bill--
> 
> Here is C# code that ... doesn't work. 
> 
> https://gist.github.com/robertmeta/42358108f589ebd740ca0c8669cb381e
> 
> That is the code and instructions to get it running, if you can 
> make it change pitch, please do. I tried a bunch of approaches 
> when I was working on this, but if it is possible, hopefully 
> someone from this list can show me, because every attempt I made
> failed. 
> 
> To do it I was working on hand implementing pitch manipulations 
> in C# using NAudio, so I would much rather have a line of XML I 
> can add.  :)
> 
> 
> 
>> On Apr 14, 2024, at 21:16, Robert Melton <lists@xxxxxxxxxxxxxxxx> wrote:
>> 
>> More than happy to be proven wrong on this, I will post a minimal 
>> reproducer that just pushes SSML to the Speech Synthesizer, maybe
>> someone here can fix it, maybe I am not implementing the SSML 
>> properly.  
>> 
>> 
>>> On Apr 14, 2024, at 21:10, <codeofdusk@xxxxxxxxx> <codeofdusk@xxxxxxxxx> wrote:
>>> 
>>> I know for a fact it works with Zira and George as NVDA supports pitch
>>> changes on both.
>>> Those are the natural Azure voices which won't work super well for us.
>>> 
>>> Bill
>>> -----Original Message-----
>>> From: Robert Melton <lists@xxxxxxxxxxxxxxxx> 
>>> Sent: Sunday, April 14, 2024 6:07 PM
>>> To: Bill Dengler <codeofdusk@xxxxxxxxx>
>>> Cc: Emacspeaks <emacspeak@xxxxxxxxxxxxx>
>>> Subject: Re: [Emacspeak] Experimenting with multiple voices for mac/windows
>>> 
>>> Windows SSML support is mostly a lie. 
>>> 
>>> It does technically parse it completely, supporting 100% of the defined
>>> spec, the problem is after parsing they throw the majority of it away.  
>>> 
>>> Pitch and Contour are examples of stuff that are not implemented on almost
>>> all voices. You in theory can get them working with the small subset of
>>> labeled "neutral" or "natural" voices, which are not installed by default.
>>> Last doc I could find out of the many, many attributes and features of SSML,
>>> the internal class implements
>>> 6 of them. 
>>> 
>>> I didn't want to make installing those voices a requirement, all the framing
>>> for pitch is already in and was even being pushed into the SSML prior to
>>> 1.3.  
>>> 
>>> I guess I can put it back and just alert the user to install one of the 3 US
>>> voices that support it.  "Guy" for US, "Ryan" for UK, and "Prashant" for
>>> India. 
>>> 
>>>> On Apr 14, 2024, at 17:38, Bill Dengler <codeofdusk@xxxxxxxxx> wrote:
>>>> 
>>>> Why do you say that Windows has no pitch support? You should be able to
>>> send pitch offset in SSML.
>>> 
>>> --
>>> Robert "robertmeta" Melton
>>> lists@xxxxxxxxxxxxxxxx
>>> 
>>> 
>> 
> 
> --
> Robert "robertmeta" Melton
> lists@xxxxxxxxxxxxxxxx
> 

--
Robert "robertmeta" Melton
lists@xxxxxxxxxxxxxxxx