ICSI corpus download

Use this page to download audio and annotations from the ICSI corpus. The annotations, which include the orthographic transcription, come all together in two zip files: one for core annotations: transcription and dialogue acts, and one containing third party annotations. The audio files are too large to package in this way, so you need to use the chooser to indicate which ones you wish to download.

Annotations, including transcription

Annotations are in NXT format. To use with signals downloaded below, unzip one or both of these files into the 'amicorpus' directory. Requires NXT version 1.4.4.

• ICSI core annotations v1.0 22-July-2016 (19MB): transcripts plus dialogue act coding
• ICSI core plus contributed annotations v1.0 (53MB): all the above plus third-party annotations for topic, hotspot, summarization etc.
• ICSI original MRT format transcripts with documentation (4MB)

To use the signals below with NXT, download the Headset mix files, and unzip them into the directory where your ICSI-metadata.xml file is. Some programs assume the wav files are directly in the Signals directory, so you may need to use soft links to use those programs with audio.

Signals

1) Select one or more ICSI meetings

Bdb001 Bed002 Bed003 Bed004 Bed005 Bed006 Bed008 Bed009 Bed010 Bed011 Bed012 Bed013 Bed014 Bed015 Bed016 Bed017 Bmr001 Bmr002 Bmr003 Bmr005 Bmr006 Bmr007 Bmr008 Bmr009 Bmr010 Bmr011 Bmr012 Bmr013 Bmr014 Bmr015 Bmr016 Bmr018 Bmr019 Bmr020 Bmr021 Bmr022 Bmr023 Bmr024 Bmr025 Bmr026 Bmr027 Bmr028 Bmr029 Bmr030 Bmr031 Bns001 Bns002 Bns003 Bro003 Bro004 Bro005 Bro007 Bro008 Bro010 Bro011 Bro012 Bro013 Bro014 Bro015 Bro016 Bro017 Bro018 Bro019 Bro021 Bro022 Bro023 Bro024 Bro025 Bro026 Bro027 Bro028 Bsr001 Btr001 Btr002 Buw001

2) Select Audio streams

For detailed information about formats and exact descriptions of each signal type, please visit the ICSI corpus documentation.

Media types	average size per meeting	comments
Headset mix	120M	single wav file
Individual channel headsets	350M	individual SPH files: original ICSI release

3) Press the button once your selection is done

All of the signals and transcription, and some of the annotations, have been released publicly under the Creative Commons Attribution 4.0 International Licence (CC BY 4.0).