[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
[gnuspeech-contact] Re: Hopi voice
From: |
Ken Beesley |
Subject: |
[gnuspeech-contact] Re: Hopi voice |
Date: |
Mon, 22 Aug 2005 20:24:21 +0200 |
User-agent: |
Mozilla Thunderbird 1.0.6 (X11/20050716) |
D.R. Hill replied 22 August:
Dear Prof. Hill.
Your message was send out some months ago, and I responded
to you then.
In case anyone else is interested, I insert a few reponses here:
Message: 3
Date: Mon, 6 Jun 2005 09:40:44 -0600 (MDT)
From: "D.R. Hill" <address@hidden>
Subject: Re: [gnuspeech-contact] Newbie requests mind-tuning
To: Ken Beesley <address@hidden>
Cc: gnuspeech <address@hidden>
Message-ID: <address@hidden>
Content-Type: text/plain; charset="iso-8859-1"
Hi Ken,
The important element that is missing from the gnuspeech suite of software
is "Synthesizer", the GUI front end to tube. The reason this is worth
having in constructing the databases for a new language is that, with a
new language, you need to know the articulatory postures related to the
sounds as precisely as possible. "Monet", which is perhaps more
important, is fully working under Mac OS/X and would allow the dynamic
composition rules for the postures to be developed. The other thing not
yet ported are the tools for creating dictionaries. Dictionaries can, in
principle, be produced using any editor. However, the dictionary tools we
used allowed each word to be heard, repeatedly modified and heard again
because they were tied into the "real-time Monet" component. It would be
a lot more laborious to work using Monet, but it could be done.
If these dictionaries are intended to record pronunciations
that cannot be derived automatically from the orthography,
as is often the case for English, then they will not be needed
for Hopi, where the modern orthography was well designed by
real phonologists.
So it may depend on just how different are the phonetic elements of Hopi
compared to the phonetic elements already in the database. I suspect
there may be speech postures, and therefore sounds, that are in Hopi and
have no equivalent in English. Also, I would guess there are differences
in the vowel qualities.
Yes, there are some different vowels, including a close back
spread vowel, and an open front rounded vowel. Vowels
have a significant long-short distinction, and there are a number
of diphthongs.
As for consonants, there's a velar nasal [ng], a ts affricate, and a
uvular stop. Several consonants with w- or y-release: spelled ky,
kw, qw, ngw, ngy. A glottal stop. A voiced bilabial fricative.
Nothing much worse than that.
In the best studied dialect (Third-Mesa Hopi) there is now
a distinctive falling-tone phenomenon. E.g. a long vowel,
a diphthong, or a closed syllable ending in a sonorant can
be either level-tone or falling-tone, and there are plenty
of distinct minimal pairs (i.e. the difference is phonemic).
I am intrigued by the possibility of trying this, hence this very quick
reply. I will give it some more thought. The missing elements I have
mentioned above only need porting. They were obviously all there under
NeXTSTEP. If you want a fast port of "Synthesizer", Steve Nygard would be
the best person to persuade, but he is busy with other things now. I'll
try and get more familiarity with the phonetics/phonology of Hopi.
I'm not ready to get into gnuspeech yet, but I am interested
in learning the status from time to time.
I've written a grapheme-to-phoneme converter, which also
identifies the stressed syllable in each word (no ambiguity).
More study will be needed to identify significant allophones.
Next year I hope to take a sabbatical and write a morphological
analyzer and perhaps a robust "chunking" parser for Hopi.
That will keep me busy. I might be ready to try to create a new
Hopi voice in gnuspeech by late 2006 or early 2007.
Then there is the question of intonation and rhythm.
These also need to be determined. The intonation will cause the bigger
problem, I think, because it was the one phonological component that was
not generalised when we created the system. However, Monet does allow
intonation contours to be put in there by hand, for investigative
purposes. I have no idea what the state of knowledge on Hopi intonation
might be. It is not particularly good even for English, and there are
competing theories out there.
I'd say the understanding of intonation and rhythm are
still very rudimentary for Hopi. It's on the list of things
to look into.
Thanks again for the update.
Ken Beesley
I look forward to hearing your reaction to the above.
Very interesting project you have!
More later.
All good wishes.
david
---
David Hill, Prof. Emeritus, Computer Science | Imagination is more |
U. Calgary, Calgary, AB, Canada T2N 1N4 | important than knowledge |
address@hidden OR address@hidden | (Albert Einstein) |
http://www.cpsc.ucalgary.ca/~hill | Kill your television |
On Mon, 6 Jun 2005, Ken Beesley wrote:
Mind-tuning: Using gnuspeech now for a new language?
I just discovered gnuspeech and am reading the available
documentation. I have a medium- to long-term goal
of creating a text-to-speech system for the Hopi language.
I had assumed that I would create a diphone or unit-
selection voice using a framework like Festival/Festvox,
but now I'm wondering if it might be possible or even
desirable to use gnuspeech in some way.
One problem in the audio recording of Hopi subjects (to
build a database for a diphone or unit-selection voice) would
be that few of them are acquainted with the orthography.
One possibility would be to present the prompts as
audio, perhaps generated by a program like gnuspeech.
Of course, a gnuspeech voice for Hopi could be very interesting
by itself.
My background: computational linguist, some background
in phonetics/phonology/IPA, specialist in finite-state
morphological analysis and generation. Competence in
Unicode, orthographies, input methods, XML. Programming
in Perl, Python, Java, C. Using Mac Tibook running OS X 10.3.9.
But I'm just getting into text-to-speech as a private interest.
Hopi Language Background:
1. There is a de facto standard or first-priority dialect now,
"Third Mesa Hopi", as documented in the excellent
"Hopi Dictionary/Hopìikwa Lavàytutuveni", 1997.
2. The phonology and orthography are well defined. I can map
reliably from orthographical text to phoneme strings, including
word stress and a falling-tone phonomenon, using a Python script;
no auxiliary pronunciation dictionary is required.
3. Phonetic details including allophonic variants, vowel lengths,
and the realization of the falling-tone phonomenon are still to
be investigated. Rhythm and intonation still need to be
investigated.
Big Question: Is the gnuspeech project currently at a state where I
could reasonably use it to create a text-to-speech system for
Hopi? Or should I concentrate on Festival/Festvox?
Thanks,
Ken
[Prev in Thread] |
Current Thread |
[Next in Thread] |
- [gnuspeech-contact] Re: Hopi voice,
Ken Beesley <=