|
From: Silas S. B. <ss...@ca...> - 2010-05-14 15:43:52
|
Hi Fergus, that's fantastic. If you want to improve the Chinese (Mandarin) voice then I'd suggest using Yali Cheng's sampled syllables, and I have Python code that (1) calls eSpeak to get a "transliteration" and (2) uses eSpeak's transliteration to sequence Yali's syllables. (It also has its own code to handle things like multiple 3rd-tone sequences and tone3+tone5 sandhi better than eSpeak currently does.) Go to http://people.pwf.cam.ac.uk/ssb22/gradint download Gradint and Yali voice (the .exe file is actually a zip so you can just unzip it on Linux, or put it in the same directory as gradint and gradint will unzip it for you when you first run gradint). Then run Gradint like this: python gradint.py 'justSynthesize = "zh "+sys.stdin.read() ; outputFile="-.mp3" ' That will read UTF-8 from standard input and write MP3 to standard output provided you have LAME on the system. Alternatively, you can try Cameron Wong's "Ekho" http://e-guidedog.sourceforge.net/ which also uses Yali's syllables and the same Mandarin dictionary as eSpeak. Might be a bit more awkward to set this up though. (Ekho also has Cantonese and Korean voices but the Korean is not very good because it doesn't correctly handle the way syllables are merged into each other. At least that's a problem you can get away without doing in Mandarin.) Silas -- Silas S Brown http://people.pwf.cam.ac.uk/ssb22 |