eSpeak in Norwegian, part 1

[Jonathan Duddington]

In article <45770DEF.7060807 at ubuntu.com>,
   Henrik Nilsen Omma <henrik at ubuntu.com> wrote:

> I then generated .wav files of each text file using the initial voice
> files. I first made files at the standard speed of -s160, but found
> that slower files were easier to analyse and settled on -s100.

> While I was at it I did the same for Spanish, Polish and Swedish
> (it's good to have some competition among neighbours!).

I wouldn't want to advertise eSpeak's Norwegian, Swedish, or Polish
voices at this stage!  Anyone who might be interested enough to help to
improve a language will download eSpeak to try it.

I'll put an English demo file on the espeak.sourceforge website.

> eSpeak was already installed on my system and I didn't want to play
> too much with that. I just put in a quick hack to use the old
> application with the new data files. I unzipped the new espeak in my
> home directory and placed the data files in ~/espeak/

eSpeak will pick up the espeak-data directory in the user's home
directory (if it exists) in preference to the one in /usr/share.

It's important to use the correct version of the espeak-data with the
compatible version of the program.  Sometimes there are changes in the
format of the compiled data files.

> Which raises the next question: What is the most useful form I can
> provide feedback in? I've made some comments on individual words
> below, mostly vowel sounds, but I suspect a more informed comment
> about the phenomes might be better. I guess having the native
> listener tweak the language files directly would be ideal but I'll
> need to grok more of the eSpeak toolchain to do that.

Your feedback is useful, and I've commented in detail below.  I'll make
changes once I see your answers to my questions.

It's useful if you can determine whether an error is in the
spelling-to-phoneme conversion (i.e. in the no_rules and no_list
files), or something else (such as a bad implementation of a phoneme
sound).

So it's useful to understand:

1.  The no_rules and no_list files (see eSpeak's docs/dictionary.html
file).

2.  The -x and -X command line options.

3.  How to test a pronunciation using explicit phoneme mnemonics eg:
      speak -vno "[[spr'o:kvi:t at nskA:p]]"

4.  speak --compile=no  to re-compile the espeak-data/no-dict file after
    making changes to no_rules and no_list.

> Språk (ubestemt) betegner menneskenes[1] måter[2] å kommunisere[3]
> på. Bevisst[4] kommunikasjon skjer[5] ved hjelp av lydspråk[6],
> tegnspråk og skriftspråk, ubevisst kommunikasjon for eksempel ved
> kroppsspråk. Språkvitenskap[7] betegnes som lingvistikk[8].

> [*] http://frodo.bruderhof.com/norskklassen/sounds-g.htm

[1] "menneskenes". The 3rd 'e' is too long

Because it's followed by a single consonant.  Perhaps a "enes" word
ending is an exception, [En at s] or [@n at s] rather than [e:n at s] ?

[2] 'r' needs to be more pronounced
Yes (deferred).

[3] The last 'e' has the wrong tone/flavour Sounds like an å, should
be like 'Long E' on [*]

Yes, I made final "e" an open-schwa sound (for which I used phoneme
mnemonic [@2]), as spoken in the Introduksjon sections at
http://www.languageonthe.net/norsk/  (aatte, sproeyte, hjerte, krone).
Is "kommunisere" an exception?
[[k'Ommu-:n,i:sare:]] or [[k'Ommu-:n,i:sar@]] rather than
[[k'Ommu-:n,i:sar at 2]].

[4] "Bevisst". The 'e' is too long/too much emphasis, and the i should
be very short (double consonant rule).

It translates as [[b'e:vIsst]], with the short [I]. It sounds like the
"short i" should be shorter, perhaps shorter than other short vowels?
 
Is "be" in "bevist" an unstressed prefix (like in German)?

[5] needs a longer 'e' More like 'Long E' on [*]

I added a rule according to the comment in [*]:
"Short e before r is usually pronounced much more like å. e.g. hver
(every)". Without this rule it would be  [[S'e:r]]

[6] the 'y' sounds like the 'ee' in Leeds, but should be like 'Long Y' 
in [*]

The example word in [*] for long-Y ("ny") does indeed sound like the
"ee" in Leeds.
Compare the sounds which I've used for long Y and I
  speak -vno "[[y:]]"
  speak -vno "[[i:]]"
also short Y and I
  speak -vno "[[y]]"
  speak -vno "[[I]]"

[7] 'å' needs to be longer like 'Long Å' on [*]

It's followed by two consonants, so eSpeak makes it short.  I can see
that Språkvitenskap is a compound word with Språk as the first element,
but eSpeak doesn't know that. This is a general problem with compound
words in Germanic languages which use the convention of a short vowel
before two consonants.  no_rules could have a rule so that Språk is
always [[spr'o:k]].

[8] needs a shorter 'i'
See [4] above.