ACL SIGLEX Resource Links

Special Interest Group on the Lexicon of the Association for Computational Linguistics

Phonetic Databases

Lists of words coded phonetically

• The English Pronouncing Dictionary is available for research or commercial development. See http://uk.cambridge.org/elt/reference/data.htm for licensing information. This is Cambridge's most popular data product, used by most leading players in the speech recognition software world. A new (16th) edition of the book is to be published later this year with for the first time a CD-ROM including sound recordings for all headwords. These sound recordings (approximately 70,000 in British English and 35,000 in American English) are also available for licensing.
• Moby Pronunciator: 175,000 entries fully International Phonetic Alphabet coded mpron.tar.Z [3.1MB]
• Carnegie Mellon pronunciation dictionary, phonemic transcriptions of 100,000 words with American English pronunciation, available for download by anonymous ftp.

Transcriptions of speech samples

• Available from ICAME Corpora: Corpora available from ICAME (International Computer Archive of Modern and Medieval English) in Bergen, Norway.
  • The London-Lund Corpus: Approximately 500,000 words of spoken British English. Various dates from 1960s to mid 1970s. Prosodically annotated version only.
• Treebanks available from the Lancaster University Centre for Computer Corpus Research on Language (UCREL):
  • The Lancaster/IBM Spoken English Corpus (SEC): Approximately 53,000 words of British spoken English, mainly taken from radio broadcasts dating from the mid 1980s. Available as orthographic text, tagged with the CLAWS2 part-of-speech tagging system, parsed, and prosodically annotated. There are also tapes of a standard suitable for the instrumental analysis of F$_{0}$ values.
• Available from the Linguistic Data Consortium: (A new service from the Linguistic Data Consortium: LDC-Online is a new search and retrieval service, offering convenient WWW access to the text and speech corpora of the Linguistic Data Consortium (LDC). For more detailed information, or to try it out, see the LDC-Online item on the LDC's home page (http://www.ldc.upenn.edu). All LDC text and speech resources are indexed for convenient online access, as long as no copyright or other restrictions prevent it. You can browse (where the copyright owner permits), or search by word, lemma or part-of-speech, or search with (limited) regular expressions combining these elements. Statistics such as word frequency and mutual information between words are also available. Retrieved speech can be displayed or played via a Java applet (for users with Java-aware browsers). A Netscape "helper application" is also available for transferring speech to other programs such as Entropic's waves+. LDC-Online is free to researchers at current LDC member institutions. An interactive tutorial is available to members and non-members alike, as is a guest account permitting access to the Brown text corpus and the TIMIT speech corpus.)
  • The Resource Management-Word Data Continuous Speech Database (RM1), Isolated and Spelled Word Data, This CD-ROM contains previously-unreleased isolated-word and spell-mode (spelled out words) speech data from the (D)ARPA Resource Management (RM1) Corpus. This data is based on a 600-word subset of the 991-word RM1 vocabulary and contains spoken and spelled words pertaining to the RM1 naval resource management task. This corpus was collected simultaneously as part of the RM1 Continuous Speech Corpus (NIST Speech Discs 2-1-2-4) and contains speech from the same sets of subjects used in RMI.
  • The Air Traffic Control Corpus (ATC0). This corpus is composed of approximately 70 hours of voice communication traffic between controllers and pilots.
  • Mandarin Chinese databases: Mandarin Chinese News Text, Mandarin Telephone Speech, Mandarin Lexicon
  • DCIEM SLEEP DEPRIVATION STUDY: MAP TASK DIALOGUES, This set of CD-ROMs contains the materials used to collect all 216 spoken dialogues digital audio, orthographic transcriptions, documentation, and source code for tools. The dialogues were selected to provide balanced representation at different points in a sleep deprivation experiment.

• The Trains Dialogue Corpus, which was released by the LDC last April, is now partially available online. Word transcriptions for each dialogue are available. The corpus contains 6 and a half hours worth of human-human dialogues, which includes 55,000 words and about 5500 speaker turns. Audio files for the dialogues are available on the CD-ROM.

To SIGLEX Resources Main Page