This is the download page for tools produced by LLSTI partners.
|
|
|
Not all the links on this page may be active still. |
|
|
|
Morphological Decomposition Shell (FLAN) |
|
A generic shell, FLAN, based on FSTs has been developed at LTRC, IIIT, Hyderabad. The purpose behind the design of this powerful shell is to create a free tool which can be used by people across the world to build morphological and other analyzers for their languages. It can be used as a stand-alone system or as a Festival module. The augmented FSTs can be put in a cascade where the output of one phase feeds as input to the next phase. Such a feed is so designed that even the selected features can be transferred across. Such phases in the cascade enable the shell to be used for handling chunking as well as morphology.
|
|
|
|
The FLAN Documentation explains how the shell is constructed and gives examples of how to use it. The shell itself, complete with example data for Hindi, is available for download here.
|
|
|
|
Optimal Text Selection |
|
HP Labs India have developed a tool which selects a subset of a large text corpus to provide a speech database with near-optimal coverage of all the language features specified. The tool and documentation are here. An example of how it has been used in the LLSTI project is in in our E-tech2004 paper, where the construction of the Kiswahili database is described.
|
|
|
|
INTSINT Intonation Modelling |
|
INTSINT (INternational Transcription System for INTonation) is a parametric phoentic model for intonation patterns, proposed by Daniel Hirst. It describes intonation with a limited set of abstract tonal symbols, which is designed such that the inventory of pitch patterns of a given language is not required. The input to the INTSINT system is a series of target points, which is estimated by a low-level F0 modeling technique called MOMEL.
|
|
|
|
CSIR in Pretoria have built and tested an INTSINT module for Festival. It comes with full documentation on how to build a module for a new voice.
|
|
Top
|
