speech to text v2
BlitzMax Forums/BlitzMax Programming/speech to text v2
| ||
is now using wav's for allophones, and should be x-platform Where I have the allophone wavs right the words sound *really* good HOWEVER, I'm no sound editor and I need YOUR help If you have a good ear for sounds, a nice voice and can chop up wavs (is the any latency with oggs?) Please contact me via the email address in my profile Thanks |
| ||
getting there http://homepage.ntlworld.com/chris.camacho/example.ogg still need help... |
| ||
Hi Chris, Concatonating wavs will improve the quality of the speech however the phonemes will not blead well. Have a listen to my ChipTalk software. It uses .wavs from the original SP0256-AL2. It would sound better if all the wavs were re-recorded with a human voice but since I was trying to recreate the SP0256 sound, I never bothered. http://www.speechchips.com/shop/item.asp?itemid=6 If you were going to do a PC speech synth you would have the luxury of recording 100 or 200 common words and only resorting to phonemes when you encounter an unknown word. The best speech synthesizer uses thousands of pre-canned words and phrases and only build a small percentage of their words on the fly. Also, when building words, they sometimes store the transitions between phonemes so that there is some smoothing between sounds. |
| ||
recording 100 or 200 common words seems a bit(lot) like cheating to me! Its comming along okay, I'll probably have a bash at redoing the allophones some time now I have a working framwork what allophone's did the SP0256-AL2 use? |
| ||
Check the spo256-al2 datasheet for a list of phonemes. Also, there is a sub dir under the chip talk installation with all the phoneme wavs. |