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

Emacspeak phrase breaks

The timing on this question  is uncanny. 
I installed Festival earlier this week and just started up a
dialog with Alan Black on using it productively with
1) Festival is definitely the right way to go --as opposed
to mbrola --MBROLA's reliance on the buggy freephone as well
as its closed source does not make it an acceptable

2) The Festival API does not currently provide a clean API
   for stopping speech and finding out where the speech
   stopped-- this is one of the things I  put in my wish
   list to Alan.

Returning to Brian's specific question about chunking:

Emacspeak's chunking algorithm is fairly complex --you do it
a disservice in speculating that it merely breaks things on
commas. It is also not Dectalk specific in anyway, and as
everything else in Emacs is completely customizable.  Look
in the source code --module dtk-tcl.el --and see how
dtk-chunk-separator-syntax is used-- basically, speech is
broken up into chunks where each chunk is a "clause" --the
meaning of what a "clause" is is mode-specific.  So for
instance, in programming modes, a clause is a line of code--
but not in English text.

Festival does have a chunking algorithm of its own --but
that is English prose specific and the output when you hand
it programming source, HTML or LaTeX markup etc is more than

----- Original Message -----

From: "Bryan R. Smart" <bsmart@pobox.com>
To: emacspeak@cs.vassar.edu
Subject: Emacspeak phrase breaks
Date: Sat, 20 Feb 1999 02:30:05 -0500

I have written a speech server to interface Emacspeak to the Festival speech system.  While Festival is a more CPU intensive speech option than Freespeech/Mbrola, the Festival option seems more stable.  The Festival source is also freely available, and works on systems for which Mbrola binaries are not available (my workstation running Alpha Redhat Linux, for example).

The point of that is to introduce all of you to the situation before I pick your brains for tips :].

Raman, I'd like to know the thinking behind your strategy for breaking up chunks to be sent to the speech server.  Usually, I receive entire lines from Emacspeak, but I've noticed that chunks are separated into separate chunks when a "," is encountered.  Is this the only character?  Is there a variable that contains all of these characters?

I assumed that the reasoning behind this was that, as the DEC-talk pauses slightly between chunks, breaking chunks at commas would allow Emacspeak to produce smoother speech than would be had by passing chunks as lines (and having a break at the end of each line).  It seems, however, that Emacspeak usually passes entire lines as chunks, so I must not have the right idea regarding chunk breaking.

Also, is their a way that I can disable this feature?  Festival automagically examines a sentence's clause structure and punctuation in order to separate the material into reasonably sized chunks.  When Emacspeak is doing something similar, it creates more utterances (chunks of spoken material) than is reasonable.

If the behavior of the Emacsspeak chunking is modifiable, it might be more reasonable to have long blocks of material sent as sentences.  That would allow Festival's advanced modules to do a more efficient job of TTS than they are able to do at the moment.

Thanks in advance for any help.  Some small modifications to Festival and organizing text into usable chunks are the only obstacles that remain.  Alan Black is helping with the speed tweaking, and I hope to come up with some solutions to the chunking problem here.


Best Regards,

      Adobe Systems                 Tel: 1 408 536 3945   (W14-128)
      Advanced Technology Group     Fax: 1 408 537 4042 
      W14-128 345 Park Avenue     Email: raman@adobe.com 
      San Jose , CA 95110 -2704     Email:  raman@cs.cornell.edu
      http://labrador.corp.adobe.com/~raman/        (Adobe Intranet)
      http://cs.cornell.edu/home/raman/    (Cornell)
    Disclaimer: The opinions expressed are my own and in no way should be taken
as representative of my employer, Adobe Systems Inc.

       To unsubscribe or change your address send mail to
"emacspeak-request@cs.vassar.edu" with a subject of "unsubscribe" or "help"

Emacspeak Files | Subscribe | Unsubscribe