Resource: RATS Keyword Spotting
|Reference||RATS Keyword Spotting|
|Date of Submission||Oct. 23, 2017, 3:59 p.m.|
|Resource Type||Primary Text|
|Media Type||Text, Audio|
|Language||North Levantine Arabic, Persian, South Levantine Arabic|
|Format/MIME Type||audio/x-flac, text/plain|
|Access Medium||Hard Drive|
RATS Keyword Spotting was developed by the Linguistic Data Consortium (LDC) and is comprised of approximately 3,100 hours of Levantine Arabic and Farsi conversational telephone speech with automatic and manual annotation of speech segments, transcripts and keywords generated from transcript content. The corpus was created to provide training, development and initial test sets for the keyword spotting (KWS) task in the DARPA RATS (Robust Automatic Transcription of Speech) program.
The goal of the RATS program was to develop human language technology systems capable of performing speech detection, language identification, speaker identification and keyword spotting on the severely degraded audio signals that are typical of various radio communication channels, especially those employing various types of handheld portable transceiver systems. To support that goal, LDC assembled a system for the transmission, reception and digital capture of audio data that allowed a single source audio signal to be distributed and recorded over eight distinct transceiver configurations simultaneously. Those configurations included three frequencies -- high, very high and ultra high -- variously combined with amplitude modulation, frequency hopping spread spectrum, narrow-band frequency modulation, single-side-band or wide-band frequency modulation. Annotations on the clear source audio signal, e.g., time boundaries for the duration of speech activity, were projected onto the corresponding eight channels recorded from the radio receivers.
The source audio consists of conversational telephone speech recordings collected by LDC: (1) data collected for the RATS program from Levantine Arabic and Farsi speakers; and (2) material from Levantine Arabic QT Training Data Set 5, Speech (LDC2006S29) and CALLFRIEND Farsi Second Edition Speech (LDC2014S01).
Annotation was performed in two steps. Transcripts of calls were either produced or already available from the source corpora. For the CALLFRIEND Farsi calls, transcripts were updated by native Farsi speakers. Potential target keywords were selected from the transcripts on the basis of overall word frequencies to fall within a given range of target-word likelihood per hour of speech. The selected words were then reviewed by native speakers to confirm that each selection was a regular word or multi-word expression of more than three syllables.
All audio files are presented as single-channel, 16-bit PCM, 16000 samples per second; lossless FLAC compression is used on all files; when uncompressed, the files have typical "MS-WAV" (RIFF) file headers.
The data is divided for use as training, initial development set, and initial evaluation set (note that the initial evaluation only used Levantine Arabic data).
This material is based upon work supported by the Defense Advanced Research Projects Agency (DARPA) under Contract No. D10PC20016. The content does not necessarily reflect the position or the policy of the Government, and no official endorsement should be inferred.
|Creator||Xiaoyi Ma , David Graff , Stephanie Strassel , Kevin Walker , Karen Jones|
|Distributor||Linguistic Data Consortium|
|Rights Holder||Portions © 1995-1996, 2003-2006, 2014, 2017 Trustees of the University of Pennsylvania|