Use email button above to contact.

Introduction

The Switchboard-1 Telephone Speech Corpus (LDC97S62) consists of approximately 260 hours of speech and was originally collected by Texas Instruments in 1990-1, under DARPA sponsorship. The first release of the corpus was published by NIST and distributed by the LDC in 1992-3. Since that release, a number of corrections have been made to the data files as presented on the original CD-ROM set and all copies of the first pressing have been distributed.

Switchboard is a collection of about 2,400 two-sided telephone conversations among 543 speakers (302 male, 241 female) from all areas of the United States. A computer-driven robot operator system handled the calls, giving the caller appropriate recorded prompts, selecting and dialing another person (the callee) to take part in a conversation, introducing a topic for discussion and recording the speech from the two subjects into separate channels until the conversation was finished. About 70 topics were provided, of which about 50 were used frequently. Selection of topics and callees was constrained so that: (1) no two speakers would converse together more than once and (2) no one spoke more than once on a given topic.

Data

In this release, assembled and published by the LDC, all known errors affecting the original publication of speech files were corrected. In addition, modifications have been made to the contents of the NIST Sphere headers of all speech files, to identify each file as being part of the new release and to make the usage of the sample_count header field consistent with standard Sphere usage. (In particular, the sample_count field should reflect the number of samples on each channel in the file. In the initial release, this field was improperly set to be the total number of samples in both channels of the file this has been corrected in the new release.)

Since the 1997 release, the Switchboard transcripts have been carefully revised at The Institute for Signal and Information Processing (ISIP) and additional problems have been discovered and patched. Three speech files, part of the original release, were inadvertently left off the 1997 revision. After corpus users noted some problems in the original speaker attribution table, LDC audited the problem calls and corrected the attributions. The latest version of ISIP transcriptions, the ISIP update of the ICSI phonetic transcriptions, and corrected word alignments are all available at ISIP. The LDC makes the transcript summaries available via in the online docs folder. Researchers have used SWB-1 data for various annotation projects including discourse annotation/speech acts, part-of-speech tagging and parsing, up-to-date orthographic transcriptions, and phonetic transcriptions. This summary documents which files have been used for the various annotations. In addition to the index of these file characteristics, there is also a table detailing speaker attributes.