randy (imported) wrote: Fri Oct 31, 2008 6:32 am
can we have an audio preview feature like youtube has? it is fun to listen to a computer say your message and people can tell when they mis spoke but didnt necessarily mis spell.
Voice synthesisys (have I spelled correctly?) takes lot's of cpu power (or results have such awfull quality, that it is unintelligible).
All of this - IMHO. I had hands-on experience with this - once tried to make a script which read email/weather. I tried festival (may be other engines had lesser requriements, but this one was free) - results were in:
1. too much processing power required.
2. was unable to tune it correctly - so to sound pleasantly.
3. Haven't found good female voice-file for russian language.
So I dropped the project.
Returning to this site - I can see 2 problems (may be more):
1. CPU time. - the one required to synthesize voice.
2. Bandwidth. - reason here same as for not hosting pictures - it is not cheap.
5 minutes mp3 (56kbit/s, mono) - 2Mb. - it's 300 seconds. About a 1 word per second = 300 words. let's say 5-7 letters per word = about 1500-2100 bytes of text not counting punctuation and whitespaces.
Hmm... some inefficiency here: 3k vs 2Mb.... To me it's too much.
PS. Joke from xkcd.com went a loong way (
http://xkcd.com/481/)