Espeak output module checked into CVS

[Lukas Loehrer]

Hynek Hanke writes ("Re: Espeak output module checked into CVS"):
> > * Listing and setting the synthesis voice works only partially. The
> > > problem is that the listed names do not identify voices uniquely, so
> > > not all voices can be selected at the moment.
> 
> Could someone please explain what is the problem here?

I am not sure if the mapping used in the output module  from espeak voices to
speech-dispatcher VoiceDescriptions is sensible. Currently, each
espeak_VOICE is mapped like this to one VoiceDescription:

espeak_VOICE.name -> VoiceDescription.name

espeak_VOICE.languages: the first language in the list is split at the first 
"-".
The First part is used as "language, the second part as "dialect" in
the VoiceDescription. If there is no "-", "dialect" is NULL.

One issue with the above mapping is taht it assumes the
espeak_VOICE.name property to be unique. I do not know if this
assumption is sensible. Also, I am not sure if we get all selectable
voices in this way.

> > * Setting voices like "FEMALE2"
> 
> I believe this is supported in espeak API. TTS API, which espeak API
> conforms to, however treats this in a little bit different way. One
> should ask for a female voice with variant=2. The set of symbolic voice
> identifier strings in SSIP is limited, so it should not be difficult to
> setup a mapping to the second more general communication code.

It does not currently work because of what I believe is a bug in espeak.
Basically, using espeak_SetVoiceByProperties() and SSML currently do
not mix well. 

The output module currently implements the mapping
somewhat different from what you describe. I will change it to match
your description if necessary.

Also, I am a bit confused about the "variant" property in the speak_VOICE
type and things like "en+12". The latter seems to select some kind of
overlay for the specified voice.

Best regards, Lukas