The RESPITE project aims to extend and apply two novel technologies - missing data theory and multi-stream theory - to the problem of robust automatic speech recognition (ASR), with particular application to cellular phones and in-car environments. Both these approaches rely on the successful identification of regions of the signal spectrum containing reliable information. Within the RESPITE project several contrasting approaches to this problem will be compared. One of these approaches is Computational Auditory Scene Analysis.
The RESPITE CASA Toolkit aims to provide a flexible, extensible and consistent framework for the development and testing of CASA systems within the project. This framework should facilitate the smooth integration of CASA software components contributed by the various RESPITE partners. It is also hoped that this toolkit may prove a useful resource to the CASA community as a whole, both as a reference framework for comparing different developments, and as a way to lower the barriers to entry into this research field.
To allow the desired flexibility and extensibility the CASA Toolkit has been designed around a block processing paradigm. Much of this design has been influenced by approaches taken in other signal processing systems, notably, the Speech Training and Recognition Unified Tool (STRUT) of the TCTS lab at Faculté Polytechnique de Mons, and the Ptolemy project, Department of EECS, UC Berkeley.
Chapter 2 explains how to download and install the CTK package on your system.
Chapter 3 describes the block orientated processing and online data flow model underlying the toolkit.
Chapter 4 provides a number of walk through tutorials which illustrate how to write CASA Toolkit scripts to run simple signal processing systems. These tutorials have been designed to be progressive and to illustrate as many of the principles of the CTK package as possible. Once you have worked through the tutorials you should be in a position to start building your own CTK systems.
Chapter 5 describes the CASA Toolkit graphical user interface (GUI). This interface has been built on top of the toolkit to aid the design of more complicated systems. The GUI allows systems to be designed in an intuitive manner by arranging widgets on a canvas. The graphical representation can then be saved as a CTK Script that can be run in the normal way. The GUI can also be used to visualise and edit existing scripts.
The toolkit has been designed to be extendable. Users with some C/C++ experience can write their own primitive blocks and compile them into the inbuilt library. Chapter 6 explains the necessary steps.