gnuspeech

What is gnuspeech?
Releases?
- Development & “Coming Soon”
Why is it called gnuspeech?
Obtaining gnuspeech
Getting help with gnuspeech
- Manuals
Finding additional packages for gnuspeech
Further information
If you want to help with gnuspeech

What is gnuspeech?

Gnuspeech is an extensible, text-to-speech package, based on real-time, articulatory, speech-synthesis-by-rules. That is, it converts text strings into phonetic descriptions, aided by a pronouncing dictionary, letter-to-sound rules, rhythm and intonation models; transforms the phonetic descriptions into parameters for a low-level articulatory synthesiser; and uses these to drive an articulatory model of the human vocal tract producing an output suitable for the normal sound output devices used by GNU/Linux. The research that provides the foundation of the system was carried out in research departments in France, Sweden, Poland, and Canada. Some of the features of gnuspeech and associated tools include:

A Tube Resonance Model (TRM) for the human vocal tract (also known as a transmission-line analog, or a waveguide model) that truly represents the physical properties of the tract, including energy balance between the nasal and oral cavities as well as the radiation impedance at lips and nose.
A control model for the TRM based on formant sensitivity analysis that allows accurate specification of the relevant vocal tract configurations for speech and comprising a low-level articulatory model having a small number of parameters and a low bit rate. The model is based on research at KTH in Stockholm, LCTI (ENST) in Paris, and The University of Calgary.
Databases specifying the articulatory postures and control dynamics required to produce English speech from an augmented phonemic input. Some French vowels are also included.
Models of English rhythm and intonation based on research at IPO in The Netherlands, the University of Essex (UK) and the University of Calgary.
“Monet”—a GUI-based database creation and editing system that allows the phonetic data and dynamic rules to be set up and modified for arbitrary languages. The MONET real-time engine also translates augmented phonetic strings into synthesiser parameters.
A text-to-augmented-phonetics module to convert arbitrary text, preferably with normal punctuation, into the input required by the MONET engine. This also provides the API for the text-to-speech system.
A 70,000+ word English pronouncing dictionary with rules for derivatives such as plurals, and adverbs. The dictionary also provides part-of-speech information for later addition of grammatical parsing and includes 6000 given names.
Sub-dictionaries that allow different user- or application-specific pronunciations to be substituted for the default pronunciations coming from the main dictionary.
Letter-to-sound rules to deal with spellings and words that are not in the dictionaries.
Tools for managing the dictionary and carrying out analysis of speech.
“Synthesiser”—a GUI-based application to allow experimentation with a stand-alone TRM. All parameters may be varied and the output monitored and analysed. It was an important component in the research needed to create the original English speech databases.

Overview of the main Articulatory Speech Synthesis System

Why is it called gnuspeech?

It is a play on words. This is a new approach to speech synthesis from text. It is also a GNU project, aimed at providing high quality text-to-speech output for GNU/Linux. In addition, it provides a comprehensive tool for psychophysical and linguistic experiments.

Releases

Gnuspeech is currently under development. It is being ported from an original NeXTSTEP 3.x version to run under GNU/Linux. No full GNU/Linux release is currently available, but a release of the interactive Monet system for Mac OS/X and GNUstep is available, with some work remaining to be completed for GNUstep version (as of 2005-03-20—see next section for details).

Development & “Coming Soon”
Gnuspeech is actively being developed/ported. Interested persons are invited to contact the authors/developers through the gnu project facilities gnuspeech-contact@gnu.org. To join this mailing list, please visit the subscription page.

Current work in progress:
- The original “Monet” articulatory speech database development system has been ported to Mac OS/X 10.3, with improvements from the original NeXT version, thanks to Stephen Nygard, and the files are being modified to compile under GNUstep—work that is almost complete thanks to Greg Casamento.
- The manual for “Monet” has been rewritten and completed. An additional manual reflecting the OS/X version should now be produced. This is mainly a matter of changing the screen shots, but includes changes to reflect the improvements in interaction introduced in the OS/X port.
- A manual for “Synthesizer”, the GUI-based application that allows direct interactive access to all the tube model (articulatory model) facilities/parameters has been written.
- A second dictionary is under way to produce a British English pronunciation more suited to the underlying speech data used for the original language definition. The current dictionary uses phonetic transcriptions representing General American, but the system speech data (postures, rhythm and intonation) are derived from British English. The resulting hybrid is entirely acceptable, and pleasant to listen to, but a consistent system would be better.
- The mirror exercise of creating posture, rhythm and intonation data representing General American should be undertaken at some point, though is of lower priority.
- The rhythm model and speed factor application require some generalisation to increase flexibility. Longer term research to improve the intonation model would be worth pursuing, though the rhythm and intonation are universally regarded as important strengths of the system even now. The question of rhythm and intonation for General American would be the first goal. As in British English, there is plenty of literature, but considerable debate!
- The transformation of the event list derived from the sequence of postures (phones) to parameters should be generalised to allow two forms: binary, and ascii-editable, plus an XML header to specify what is coming. The protocol should allow arbirtrary parameter sets to be defined to drive different or modified synthesisers. This is really a technical/implementation issue.
- The compromises included in the tube model (Tube Resonance Model, or TRM) to cope with the limited computer power of the NeXT computer can now be removed. At the same time the noise balances require correction, and a socket interface provided to generalise the communication method, along with direction of the output to the standard GNU/Linux sound output.
- The aforementioned changes are part of a more general effort needed to take the ported version of “Monet”, strip out all the interactive parts, and isolate the essentials needed for a “real-time Monet” that can form the basis of a service or daemon that provides direct text-to-speech translation without involving the development and interactive parts of the existing “Monet”. An initial “real-time Monet” would likely still run under GNUstep, but a pure C-version port should follow as a means of providing high quality text-to-speech for users who wish to run plain vanilla GNU/Linux without GNUstep.
- The dictionary editing and letter-to-sound rules should use the same phone/posture database as MONET to avoid the possibility of inconsistencies. Additionally, the letter-to-sound rules should be simplified and updated. Such simplification is possible because it is common words that are most likely to have irregular pronunciation, and these are now looked up in the dictionary. The updating is needed to make the pronunication consistent with the dictionary used (e.g. the rules should reflect British English if used with a British English dictionary).
- An ongoing review of the overall system is called for during and after the port to see what can be done better, check for any unsuspected bugs, etc.
Obtaining gnuspeech

Gnuspeech is currently fully available as a NextStep 3.x version, and partly available as both a Mac OS/X version and as a GNUstep version. These files are available in the CVS repository.

Getting Help with gnuspeech

Developers should contact the authors through the gnuspeech-contact@gnu.org mailing list. To join this mailing list, please visit the subscription page. Papers and manuals are available on-line (see below)

Manuals and papers

A number of papers and manuals relevant to gnuspeech exist:
- The “Monet” manual provides a detailed view of the facilities and screens associsated with the MONET subsystem, but does not describe the MONET engine that is used for real-time construction of parameters.
- The “Synthesizer” manual provides a detailed view of the interactive application that allows access to all the parameters and facilities of the Tube Resonance Model (TRM) synthesiser as a learning tool and a development tool.
- A paper presented at the American Voice I/O Society conference in 1995 provides a reasonably detailed explanation of the theory underlying the tube resonance model.
- A heavily cross-referenced “conceptionary” is available to provide access to some of the background terms and research in the relevant scientific fields.
- A guide to the pronunication notation used in the text-to-speech work showing the relationship between standard forms (IPA, Websters) and the ASCII-friendly form used in gnuspeech, with examples of actual pronunciations.
Finding packages for gnuspeech

There is the original NeXTSTEP Developer Package, which is available under a GPL, but does not run under GNU/Linux. There is also now a version of the full “Monet” system for Mac OS/X and GNUstep that provides the core of the text-to-speech development facilities and allows arbitrary text to be changed to speech. Note that further work is needed to strip this version to make a daemon-like module for incorporation within applications, or as a service, as noted above. Check out the Savannah CVS repository and search for “gnuspeech”. Current work is under the current directory.
Further information?

See the section on Manuals and papers

How to help with gnuspeech

To contact the maintainers of gnuspeech, to report a bug, or to contribute fixes or improvements, or to join the development team, send mail to gnuspeech-contact@gnu.org. To join this mailing list, please visit the subscription page.

Return to GNU's home page.
Please send FSF & GNU inquiries & questions to gnu@gnu.org.

We thank David Hill for writing this page.
Please send comments on these web pages to webmasters@www.gnu.org, send other questions to gnu@gnu.org.
Copyright (C) 1998, 2001 Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111, USA
Verbatim copying and distribution of this entire article is permitted in any medium, provided this copyright notice is preserved.

Page last updated 05-03-21 @ 17:45 pm PST