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

RE: [kidsgames] Word

[ Whatever happened to putting kidsgames in the reply-to? not enough
votes, or was it voted down? ]

First, I did listen to some of the Festival synthesis examples
from their page, and I'm surprised how bad it sounds.  I guess the
high-end synthesis system I was listening to four years ago really
was high-end.  They do have a sample near the bottom which uses
a better synthesis technique, and it sounds pretty reasonable
I understand they are still using sub-word samples, but they have
different samples for each prosodic context.  I think you will be hard
pressed to do better using full words to synthesize connected speech.

And, if you are talking about reading books as you are in the following,
then you indeed want to synthesize connected speech.  Now, since you are
talking about the same material in multiple languages you are clearly
not talking about arbitrary text (translation is in an even sorrier state
than synthesis).  The text will have to be translated.  Getting a reading
of the translated text is even easier than getting it translated in the
first place.  So instead of looking for good speech synthesizers, you
should be looking for high quality speech compression.  Perhaps some of
the internet phone applications have it built in?  Note that speech
compressors will do horribly on things like music or environmental
sounds, but you want to keep these separate from the speech stream anyway
for a multiligual story so that you only have to do them once.

	- Paul

>Do you think that such a synthtesis would be useful for a multilingual word
>base? I'm talking here about making a drop in word base so that a program
>could, assuming you could do the grammatical conversion, be used to teach
>the same material in different languages.  A SDK starter is what I'm talking
>about I work on the english base and corresspond it to other language
>meanings and use the same word references to teach the same material,
>assuming a matching connotation and grammar can be aligned.  And regardless
>of what book is being read to the child they can hear the correct words, and
>translated because they have the asci word displayed as it is read by the
>synthesizer.  One story many languages!!
>														Brian Thompson
>> -----Original Message-----
>> From: Paul Kienzle [mailto:pkienzle@kienzle.powernet.co.uk]
>> Sent: Thursday, December 09, 1999 11:29 PM
>> To: briandthompson@home.com
>> Subject: RE: [kidsgames] Word
>> Festival is indeed a synthesis program.  The most recent distribution
>> includes LPC resynthesis (that is, they take a phoneme and chop it
>> into slices, smooth the power spectrum of the slices, and resynthesize
>> with the appropriate pitch) and diphone synthesis (that is, they take
>> actual recordings of phoneme pairs and concatenate them together,
>> while correcting for pitch) as the construction techniques.  No, I
>> don't know how they sound since I haven't yet taken the opportunity to
>> download them and try them out.  I do know, however, that upon hearing
>> the output of what was then a top of the line synthesizer (circa '95)
>> I thought it was crap, but when we did appropriate markup for prosody
>> (pitch and lengthening) it sounded really good.
>> What you are proposing (word-based concatenative synthesis) will still
>> sound bad unless you correct the prosody.  Unfortunately, this is
>> much harder to do on the word level then it is on the diphone level.
>> Consider the intonation pattern on 'interested' in the exchange "Are
>> you interested?  Yes, I am in interested."  To put the intonation rise on
>> the end of the question, you will have to detect the start of the final
>> syllable of the final word.  My personal bias is that the half-syllable
>> is the proper domain for synthesis, but I haven't had a chance to prove
>> it yet.
>> Now, to download and try out:
>> Festival version 1.4.1 sources, voices, and Linux binaries are
>> available from
>>     http://www.cstr.ed.ac.uk/projects/festival/download.html
>> and also from the US mirror at CMU
>>     http://www.speech.cs.cmu.edu/festival/download.html
>> I'll let you know how it sounds.
>>  - Paul
>> >What is estival and how do I get it? And what are the uses for it is it a
>> >synthesis program most of them sound poor still.
>> >
>> >> -----Original Message-----
>> >> From: Paul Kienzle [mailto:pkienzle@kienzle.powernet.co.uk]
>> >> Sent: Thursday, December 09, 1999 1:20 AM
>> >> To: briandthompson@home.com
>> >> Subject: Re: [kidsgames] Word
>> >>
>> >>
>> >>
>> >> I don't understand the point.  Word concatenation will not get
>> >> you anything
>> >> approaching natural speech.  Remember the scene from "Sneakers",
>> >> where they
>> >> fed the concatenated recording of "My voice is my password.
>> >> Verify"?  It will
>> >> sound as bad as that unless you go to a lot of effort to correct pitch
>> >> and time problems.  Surely festival or some such will have much
>> >> better output
>> >> with much less work?  And if festival isn't installed, then no audio.
>> >>
>> >> Paul Kienzle
>> >> pkienzle@kienzle.powernet.co.uk
>> >>
>> >> >I have the yawl and it is fairly large and was wondering how
>> to go about
>> >> >turning it into a database for text to speech conversion.  I was
>> >> going to go
>> >> >simple and clumsy like wav files and the ascii word association,
>> >> large but I
>> >> >could record each word, well maybe not all of them, and if its
>> >> done right I
>> >> >could do multilingual associations easily.  What are some
>> suggestions for
>> >> >making the database, what to use, I have an idea of the
>> structure I want,
>> >> >but I'd like to make it easy to add new words and languages for each
>> >> >'meaning'  such as home=casa and thesaurus.
>> >> >
>> >> >
>> >> 						Brian Thompson
>> >> >
>> >> >-
>> >> >kidgames@smluc.org  -- To get off this list send "unsubscribe" in the
>> >> >body of a message to majordomo@smluc.org
>> >> >
>> >> >
>> >>
>> >
>> >
>> >
kidgames@smluc.org  -- To get off this list send "unsubscribe" in the
body of a message to majordomo@smluc.org