Jonathan Duddington, разработчик новой системы для синтезирования речи написал мне письмо. Публикую его здесь, тк надеюсь на помощь сообщества. Если кто-нибудь заинтересовался и может помочь разработчику с реализацией поддержки русского языка, пишите мне на почту igor4u@gmail.com и я дам вам его контакты.
I found your details as the administrator of the Ubuntu Russian translators list and wondered whether such a project might be of interest.
I have written "eSpeak", an open source text-to-speech synthesizer which has recently been included for Ubuntu "Edgy". In addition to English, I have started the implementation a few other languages which I hope to improve with assistance from native speakers. The project is at: http://espeak.sourceforge.net/
Russian would also seem to be a good choice to add. I know nothing of Russian, except what I can read at places like http://en.wikipedia.org/wiki/Russian_...
I have a question. Russian spelling-to-sound seems regular, which is good, but it seems that which syllable of a word has the main stress varies, and cannot be found from the spelling. Some languages have regular stress position (eg, next-to-last syllable in Polish) or they mark stressed syllables with accent marks. But not Russian.
In Russian, must the stress position of every word be learned separately?
In English also, the stress position can vary, but the default is the first syllable of a word, and there are some rules (some prefixes are unstressed, some suffixes take the stress or put the stress on the previous syllable). Nevertheless there are many exceptions.
What is the situation with Russian? Does a text-to-speech engine need a dictionary of all the words in the language in order to look up the stress position? If so, does a free machine-readable list exist? Or are there some rules (or at least generalizations) which would limit the number of exceptions?
If you know anyone who would be interested in helping with a Russian speech synthesizer, please let me know. I can set up an initial attempt at Russian which others could work on to improve, see: http://espeak.sourceforge.net/add_lan...
.png)
