Numbers Corpus Release 1.3 Center for Spoken Language Understanding UPDATED: 23 August 2002 This document describes the file naming conventions used for this distribution and gives a brief description of the various file formats used. File Naming Convention ---------------------- Each filename in the corpus (other than the documentation) encodes information about the call number, utterance type, and file type. A typical filename will look like: NU-1214.firstname.wav The "NU" prefix indicates the corpus name, i.e. Numbers. The number between the "-" and the first "." is the call number. Call numbers are described below. The string between the first and second "." is the utterance type. The following utterance types are in this corpus: phone Phone number streetaddr Street number as taken from an address zipcode Zipcode utterance other1 Any other number that is not one of the above. There may be several "other?" utterances. Additional ones will be identified as "other2", "other3", etc. The final, three letter, extention indicates the file type. The following types are in this distribution: wav The speech data txt The orthographic transcription phn The phonetic transcription File Formats ------------ There are three file formats used in this corpus (other than the documentation). The .wav file format used is the RIFF standard file format. This file format is 16-bit linearly encoded. The "txt" file is simply an ascii text file. The "phn" file contains a time aligned phonetic transcription. The first two lines of the file are considered the header. The first line defines the "frame rate", which is the number of milliseconds per frame. Each line after the header looks like this: 0 203 i: The two numbers indicate the starting and stopping frame, respectively, that corresponds to the label, which is the third field. By using the "frame rate", the correspondence between the phonetic labels and the waveform can be determined.