Text to speech complaint and Hello everyone

Boris Gaertner Boris.Gaertner at gmx.net
Sun Aug 22 15:46:22 UTC 2004


From: Joshua Scholar wrote:

>On further testing I find that anything with double quotes fails:
>Speaker man say: '"Hello," she said.'.
>Numbers blow it up, even single digits.
>Speaker man say: '7'.
>The sequence space, carrage-return, space blows it up even 
>in the middle of some text.
This is really helpful information. With this facts it is not too
difficult to try to do something.

After some debugging, I think that these methods should be
carefully revised:
Speaker >>clauseFromString:
Speaker >>phraseFromString:
Speaker >>wordFromString:

A small modification of  speaker>>clauseFromString:
solves some of your problems - but I am not entirely happy
with the result: When we add the double quotes to the set of
token delimiters,they are simply skipped. For a first attempt
to improve the situation this is acceptable, but double quotes
have a function in speech: They influence intonation and pauses.

The digits and numbers require some additional work: We need
a method that converts numbers into their written form:

 473 -> fourhundred and seventythree.

Please have a look at  Integer>>asWords. It is perhaps not
exactly what we need, but it is something that we can use.

In Speaker>>phraseFromString  I added code that converts integer
denotations into words. This is not a perfect solution; it does
not work for floats, but it works for integers. Also, we are
currently not able to pronounce mathematical expressions, which
is certainly a desireable feature.

Attached you find a change set with these changes. Please
try it. I did not label this as a fix, because it is still so very
incomplete, but I want to give it to you now for two reasons:

First, you should know that you will get support on this list
and second, I want to give you some encouragement: Bug
fixing is not so very difficult in Squeak. What I did is this:

>From the debugger, I saw that Speaker>>clauseFromString:
is a method where a 'self halt' could be helpful. With that
halt, I was able to follow string parsing and speech
generation step-by-step and I could quickly find where
some tentative changes might improve the situation.


>There are probably many other bugs.

>It's probably best to test the program by reading articles.
>If it can read a few web sites then its probably robust.
This is an excellent idea! I am really ashamed that I never
did that.


Greetings, Boris
-------------- next part --------------
A non-text attachment was scrubbed...
Name: SpeakerFixes.1.cs
Type: application/octet-stream
Size: 1439 bytes
Desc: not available
Url : http://lists.squeakfoundation.org/pipermail/squeak-dev/attachments/20040822/570c94fe/SpeakerFixes.1.obj


More information about the Squeak-dev mailing list