Society of Robots - Robot Forum
Software => Software => Topic started by: airman00 on January 21, 2008, 10:37:14 PM
-
Hi ,
So as you know I've been working on making Chives speak the weather and speak the time .
I have been using this text to speech engine http://www.research.att.com/~ttsweb/tts/demo.php (http://www.research.att.com/~ttsweb/tts/demo.php), but I have to prerecord the voices .
So I have two approaches
One I record the sounds of the first digit numbers , 1 ,2 ,3 ,4, 5, 6 ,7 ,8 ,9 and the sounds of ten, twenty, thirty, etc. then for a number I would play sounds , like for 31 it would be the sound for thirty and the sound for one , two files playing right after each other . Now I can do this sort of engine in Visual Basic.
My problem is that sometimes the temperature comes out all wierd and not natural. But when I record the osund as one , like I text to speech 31 it sounds better than thirty played and then one played. Is it me or does it really get messed up ,can someone please jsut do the speech to text a few times to see which way is good .
In that scenario when i record full numbers I would need 100 sound files( which isnt a problem , since i have a 60 gb hardrive) but it would take a long long time . So perhaps some of you *wink wink * can help me out a bit and do 5 or 10 , but contact me before you do any , so I shouldn't have doubles. But that is only in the case where it is necessary to record the numbers as a whole.
And of course, this will be entirely open source! Hooray!
,Eric
-
My problem is that sometimes the temperature comes out all wierd and not natural. But when I record the osund as one , like I text to speech 31 it sounds better than thirty played and then one played. Is it me or does it really get messed up ,can someone please jsut do the speech to text a few times to see which way is good .
can you post mp3/wav examples of it?
What you might want to do is that 30 then 1 method, but for any numbers that don't sound right, do a 31 special exception sound.
Also, when one says 30, and then 31, that 'thirty' part of the sound is actually different! You may want two different recordings for 30, where one is alone (30), and the other expects a second number (31,32,33).
-
OK so I finished the code for this thing, ill post it up after i get back from vacation. However, its in visual basic
It does sound a bit awkward when he says the time and temperature , but I am thinking of recording every sound individually instead of the system I have now. Only thing in my way is laziness.
So ill post the code and the video of it online , probably next week ( maybe even tonight )
By the way http://www.research.att.com/~ttsweb/tts/demo.php
(http://www.research.att.com/~ttsweb/tts/demo.php) is a very cool text to speech program , but you have to record each sound separately .
By the way this program will go on chives and on a new robot I am working for your car ( kind of like r2d2)
thats why I added speech controlled mapping and find nearest gas station , convenence , store , ect.
If you have any other tips I can do to improve the naturallness of the sound , please tell me. thank you
,Eric