CHATR was developed as a generic speech synthesis system. Rather than building a single synthesis system, CHATR was designed to be modular in a way that easily allows a large number of different modules to interact in a user defined way. Multiple modules, performing equivalent tasks such as waveform synthesis, duration prediction or intonation, are incorporated within the same environment, allowing close direct comparison.
CHATR is still under development but already offers a number of synthesis methods (non-uniform unit concatenation, diphone concatenation and formant synthesis) and is easily configurable. To realize this means there are many parameters that can be set, giving much power but resulting in an apparently bewilderingly large command set.
Specifically CHATR is not a text-to-speech system, though it does support a simple text input module. CHATR's main use in ATR will be within a speech translation system. To turn translated speech into plain text and then expect a text-to-speech system to recreate all the information that has just been thrown away is a waste of resources. The language generation system already has a labeled form of the utterance and it is this far richer form of the utterance that is intended as the input to the CHATR system--though many other types of input are possible.
The core system is written in C, but basic input and output are through a very simple Lisp system. This allows a uniform (simple) syntax for data and commands. There is a command line interface that aids the user in typing commands, with command, argument, file and variable name completion. Command line editing and history are also available. These are provided via the GNU readline library.
This system may also be used in a batch mode, when files of commands can be processed. It may also be used in a UNIX pipeline, reading commands and data from `standard in'. Also, a UNIX socket based client/server interface has recently been added.
This document consists of several sections. The next is an introduction for the impatient, showing what to do just to make the system run. Don't expect many details, and you will still require a further basic understanding of the system before you can use it fully. A more detailed description of the system then follows. Each section of the system is discussed, stating its function, related commands, and data formats.
A number of appendices give listings of all the current CHATR commands, variables and modules. These are automatically generated from descriptions within the code itself and may even be up to date.
Go to the first, previous, next, last section, table of contents.