The Kestrel TTS text normalization system. (May 2015)

• Record Type: Journal Article
• Authors: EBDEN, PETER; SPROAT, RICHARD
• Is Part Of: Natural language engineering. Volume 21:Part 3(2015)
• Page Start: 333
• Page End: 353
• Publication Date: 2015-05
• Subjects: Natural language processing (Computer science) -- Periodicals; Software engineering -- Periodicals; 006.35
• DOI: 10.1017/S1351324914000175
• Languages: English
• ISSNs: 1351-3249

Abstract

<abstract abstract-type="normal"> <title>Abstract</title> <p>This paper describes the Kestrel text normalization system, a component of the Google text-to-speech synthesis (TTS) system. At the core of Kestrel are text-normalization grammars that are compiled into libraries of weighted finite-state transducers (WFSTs). While the use of WFSTs for text normalization is itself not new, Kestrel differs from previous systems in its separation of the initial <italic>tokenization and classification</italic> phase of analysis from <italic>verbalization</italic>. Input text is first tokenized and different tokens classified using WFSTs. As part of the classification, detected <italic>semiotic classes</italic> – expressions such as currency amounts, dates, times, measure phases, are parsed into protocol buffers (<ext-link ext-link-type="uri" xlink:href="https://code.google.com/p/protobuf/" xlink:type="simple" xmlns:xlink="http://www.w3.org/1999/xlink">https://code.google.com/p/protobuf/</ext-link>). The protocol buffers are then verbalized, with possible reordering of the elements, again using WFSTs. This paper describes the architecture of Kestrel, the protocol buffer representations of semiotic classes, and presents some examples of grammars for various languages. We also discuss applications and deployments of Kestrel as part of the Google TTS system, which runs on both server and client side on multiple devices, and is used daily by millions of people in nineteen languages and counting.</p> </abstract>