Introduction

Overview

SPPAS provides automatic annotations using language-independent algorithms. This means that adding a new language to SPPAS simply requires adding the necessary linguistic resources for annotation:

  • Text normalization requires lexicons;
  • Phonetization requires a pronunciation dictionary;
  • Alignment requires an acoustic model;
  • Syllabification requires a configuration file with the rules.
Since SPPAS is a language-independent annotation tool, you can annotate any language as long as the necessary linguistic resources are available. The quality of the annotations depends on the accuracy of these resources.

This document includes tables listing the phones available in the resources used for phonetization, alignment, and syllabification of a given language. The first column shows the symbols used by SPPAS, while the other columns help users better understand their meaning.

SPPAS encodes phones using X-SAMPA, a machine-readable phonetic alphabet based on 7-bit printable ASCII characters. It is a language-independent notation that covers the entire International Phonetic Alphabet (IPA). A SPPAS plugin allows converting time-aligned phones from X-SAMPA to IPA.

The acoustic models created by Brigitte Bigi were trained using the HTK toolbox, version 3.4.1. HTK was developed by the Machine Intelligence Laboratory (formerly known as the Speech Vision Robotics Group) at the Cambridge University Engineering Department (CUED) and Entropic Ltd. Microsoft has since licensed HTK back to CUED and is supporting its redistribution and development via the HTK3 website. (Source: http://htk.eng.cam.ac.uk/). Please note that HTK is available for free download after registration, and users must agree to its license terms. Section 2.2 of the license states that HTK cannot be distributed or sub-licensed to any third party, either in whole or in part, in any form.

Download and Install

To automatically install the resources in SPPAS, open the graphical user interface using the sppas command (either sppas.bat or sppas.command). Then, click the Annotate button in the menu, followed by Add languages or Add annotations in the toolbar. A dialog will open where you can select the languages or annotations you want. Finally, click the Install button.

Alternatively, you can install linguistic resources using the command-line interface with the script sppas/bin/preinstall.py.

Available Resources

The Ortolang repository allows to download the linguistic resources. It contains two types of data, organized into separate folders:

  • lang folder: This folder contains language-specific resources required for automatic annotation. Each language may include lexicons, pronunciation dictionaries, acoustic models, and/or syllabification rules. These resources are provided as ZIP files, named using the ISO 639-3 language code (e.g., fra for French, eng for English, cmn for Mandarin Chinese). For a complete list of language codes, visit http://www-01.sil.org/iso639-3/.
  • annot folder: This folder contains additional linguistic resources needed for certain types of automatic annotation, such as statistical models. These resources are also available as ZIP files, with one file per supported annotation type.

The available resources were initially created for use with SPPAS), but since they are open source, they can be freely downloaded, used with other annotation tools, and even modified or redistributed in most cases.

Contribute

Each provided resource is not perfect… and can always be improved. How?

  1. Edit the file, apply your changes, and email us your modified version.
  2. Send new audio recordings and transcriptions to help retrain the acoustic model.

You can also create new resource files and share them! Altruism contributes to the quality of life… you can make a difference.

Licenses

Most of the distributed files are released under the terms of the GNU General Public License.

However, some resources may be subject to more restrictive licenses. Please check carefully, especially before redistributing them.