I use the AT&T Natural Voice engine for most of my general fiction* conversion.. fairly resource intensive, but one of the better sounding voices. I keep a list of standard substitutions as I notice them. The engine does poorly on abbreviations and foreign loan words, and of course on heteronyms. Lead, axes, alternate, etc. You can specify alternate pronunciations in a phonetic language. Concatenated engines like Natural Voices, Cepstral, Neospeech and RealSpeak are limited in how much you can alter speed and timber before they get unusable.. NV tends to clip syllables at anything above roughly +1 or +2. Most of these engines are available via Nextup and other online retailers.
Freeware engines such Festival tend to have somewhat lower out-of-the-box quality, but are more flexible (at least if you can tolerate LISP). In particular, in a synthesized TTS engine, you can turn up the speech speed much further before it becomes unintelligible, but it sometimes requires practice to understand.
Synthesized speech compresses quite well with voice codecs.. if I'm not using an external MP3 player, I'll compress it with Speex at quality 4 or 5.
R C
*(I generate audiobooks from Webscriptions and Gutenberg for commute and other relative downtimes.)
On 3/14/06, Jeroen Hellingman (Mailing List Account) <jeroen.mailinglist@bohol.ph> wrote:
Hi All,
I am studying the options for preparing ebooks for text-to-speech. Does
anybody have experience with that and willing to share experience.
I am looking at things like SSML, aural-CSS, and text-to-speech
software. Any software that can support this? My intention is to add the
relevant tags to my TEI master, and generate SSML from that, feed that
to TTS software to obtain audio files (Ideally, I would only post the
SSML, and let people regenerate the speech when needed). Any tools that
can be advised?
Things to consider are additional tags to disambiguate words with
identical spelling (read and read; record and record, for example), and
to help pronouncing dates, currency amounts, measures, abbreviations, etc.
Issues I found is lack of support for things like aural CSS, expensive
software, etc.
Jeroen.
_______________________________________________
gutvol-d mailing list
gutvol-d@lists.pglaf.org
http://lists.pglaf.org/listinfo.cgi/gutvol-d