speechd-discuss
[Top][All Lists]
Advanced

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

Ommited index marks in SSML with eSpeak


From: James Simmons
Subject: Ommited index marks in SSML with eSpeak
Date: Thu, 05 Jun 2008 09:19:57 -0500

Hynek,

Using the content of the marks to control highlighting sounds like a 
good idea, but the mark is received *after* the word is spoken, not 
before, so I'd have to use a combination approach to get the 
highlighting working, unless there is a way to get callbacks before the 
word is spoken.  (Or maybe make the mark show the coordinate of the next 
word?  I'll have to think about that).

I'm glad to hear festival is working.  OLPC currently ships with espeak, 
but they are interested in doing speech with highlighting throughout the 
Sugar environment and if festival worked better its possible they might 
switch.  Speech-dispatcher does not ship with OLPC yet, but it will at 
some point, and they do want to use SD as the normal method to produce 
speech.

Thanks again,

James Simmons


Hynek Hanke wrote:

>
>> I was interested in your theory that punctuation in the SSML tags was 
>> messing up the callbacks so I changed my program to completely 
>> eliminate punctuation in words spoken (except for the apostrophe) and 
>> to replace the word in the SSML tag with a number representing the 
>> byte offset from the start of the page being read.
>
>
> Hello, this proved to be a workaround for the SSML index marks bug 
> related to punctuation that I observed with eSpeak. However, the index 
> marks still do not work well with eSpeak. With latest 
> festival-freebsoft-utils, your application works with Festival very 
> reliably.
>
>>   I find that this has no effect.  The highlighting falls behind at 
>> exactly the same spot.
>
>
> Well, the only problem is that certain index marks are never 
> delivered. Your application would be much more robust (and I think you 
> would hardly notice the described bugs) if you relied on the content 
> of the index marks rather than on their count. Now that you use text 
> position as the content of the index mark, I believe you should use 
> the returned position for highliting and that could practically fix 
> the problem.
>
> There is still the problem that eSpeak doesn't seem to return some 
> index marks (or perhaps there is some problem with the communication 
> with the output module in Speech Dispatcher). To be able to solve 
> this, I'd need assistence of Jonathann or a way to test index marks 
> directly in eSpeak. Both the punctuation problem reported and now also 
> this another problem which is unrelated to punctuation happen also 
> with espeak 1.37.
>
> Thanks for helping to track down these issues,
> Hynek Hanke
> ,






reply via email to

[Prev in Thread] Current Thread [Next in Thread]