eSpeak in Norwegian, part 1

Jonathan Duddington jsd at
Fri Dec 8 15:38:34 GMT 2006

In article <4579484C.7010808 at>,
   Henrik Nilsen Omma <henrik at> wrote:

> Perhaps we should find a way to distinguish voices that are ready for 
> use and those that are just ready for testing and tweaking? Call it 
> no-pre, no-test or something?

It really needs feedback from native speakers to say whether a voice is
good, adequate, just about usable, or rubbish. 

> Yes, so we should figure out a better way of installing the latest 
> version on ubuntu (or running the latest anyway).

I find it best to run the program and data in my home directory, so
that I don't need to update files in /usr

   /home/myuser/espeak-data    the data files
   /home/myuser/bin/speak      the program file

Actually my /home/myuser/bin/speak  is a link into my development
directory where I compile the speak program.

> Oddly, "[[m,e:nn at sk'En at s]]" is actually better here.

This is where native knowledge of the language is needed.  This
suggests a possible rule:
   enes (_  'En at s

i.e "enes" at the end of a word is a stressed suffix (like English
"-ation").  So now you need to determine whether this is a general rule
or whether this word is an exception (obviously, I don't know).  Look
at all the Norwegian words which end in "enes" and see whether it

For English, I have a list of about 100,000 words from a spell-checker
program.  I run rgrep to find all the words which match a rule which
I'm considering.

Finding and testing spelling-to-phoneme rules can be tedious, depending
on the language.  They must give both the correct phoneme translation
and the correct stress position.  But surely no language is as bad as
English :-)

> > Is "be" in "bevist" an unstressed prefix (like in German)?
> >   
> Yes "be" is unstressed, and the I is very short indeed.

Are there any other unstressed prefixes?

> > It's followed by two consonants, so eSpeak makes it short.  
> AFAIK that rule only applies when the consonants are the same.

OK, I've changed the short-vowel rules.

This update (and the phoneme changes mentioned below) is in version
1.17l at

> Hm, none of these really capture the Norwegian Y.

So the example word "ny" at
is not good Norwegian?  I've now abandoned most of the vowel sounds
which I made from those examples and used replacements from other
languages, including a [y:] which is more different from [i:], so I
hope that's better. 

Try out the vowel sounds, and tell me which are good, OK, and bad. 
Then I can give a selection of alternatives for you to choose from.
Hopefully we can get the vowels sorted out fairly quickly.

The Norwegian vowel phoneme mnemonics which I've used are:

schwa ([@2] is an open variant for end-of-word, if it sounds OK)
  [@] [@2]

vowels  ( : indicates a long vowel)
  [i:] [I]
  [y:] [y]
  [Y]  [W]
  [e:] [E:] [E]
  [a:] [a]
  [A:] [A]
  [o:] [O]
  [u:] [U]
  [u-:] [u-]

  [AI]  [aI]  [OI]  [u-I]  [Yy]  [aU]

These phonemes are defined (with their relative lengths) in:
in the espeakedit package.  This also contains a vowel chart
(phsource/vowelcharts/no.png)  which shows their positions in "vowel
space" (roughly the position of the tongue in the mouth), so you can
compare them with vowels from other languages and find possible

More information about the Ubuntu-accessibility mailing list