OpenFst: A General and Efficient Weighted Finite-State Transducer Library - 2007

Explore the OpenFst library, an open-source solution for weighted finite-state transducers (WFSTs), in this informative talk by Michael Riley from Google. Delve into the C++ template library's efficient WFST representations and over twenty-five operations for constructing, combining, optimizing, and searching transducers. Learn about the shell-command level transducer file representations and programs, and discover how OpenFst is designed for both efficiency and scalability. Gain insights into key applications in speech, image, and natural language processing, pattern and string matching, and machine learning. Examine the library's overview, key algorithms, usage examples, and design details that allow customization of labels, states, and weights. Understand the lazy evaluation of many operations and find out how to access further information and download the OpenFst library. This talk, presented at the Center for Language & Speech Processing (CLSP) at JHU in 2007, also includes a brief biography of Michael Riley, highlighting his academic background and professional experience in the field of computer science and language processing.