1 / 12

NXT meets the ICSI Corpus

NXT meets the ICSI Corpus. Jean Carletta and Jonathan Kilgour University of Edinburgh HCRC Language Technology Group. ICSI Meeting Corpus. 75 natural meetings from research groups close-talking and far-field microphones orthographic transcription "speech quality" tags (e.g., emphasis)

laszlo
Download Presentation

NXT meets the ICSI Corpus

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript

  1. NXT meets the ICSI Corpus Jean Carletta and Jonathan Kilgour University of Edinburgh HCRC Language Technology Group

  2. ICSI Meeting Corpus • 75 natural meetings from research groups • close-talking and far-field microphones • orthographic transcription • "speech quality" tags (e.g., emphasis) • dialogue acts using MRDA • hot spots

  3. The NITE XML Toolkit • library support for data handling and search using a data model that can express both timing and complex structure • multiple file stand-off XML data storage • some standard GUIs, data utilities • library support for writing tailored GUIs

  4. Stand-off XML extract from Bdb001.A.speech-quality.xml <speechquality nite:id="Bdb001.emphasis.16" type="emphasis"> <nite:child href="Bdb001.A.words.xml#id(Bdb001.w.1,342)..id(Bdb001.w.1,344)" /> </speechquality> extract from Bdb001.A.words.xml <w nite:id="Bdb001.w.1,342" starttime="356.39" endtime="" c="W">time</w> <w nite:id="Bdb001.w.1,343" starttime="" endtime="" c="HYPH">-</w> <w nite:id="Bdb001.w.1,344" starttime="" endtime="356.59" c="W">line</w>

  5. Tasks • pre-NXT: up-translation and tokenization • hand annotation (topic segmentation, dialogue acts, extractive summaries, ...) • automatic annotation/indexing by query match

  6. Queries in NXT ($a w):(TEXT($a) ~ /th.*/):: ($s speechquality):($s ^ $a) && ($s@type="emphasis") • Find instances of words starting with “th” • For each find instances of speech quality tags of type "emphasis" that dominate the word • Discard words that are not dominated by at least one such tag Use queries to understand data, verify quality, index.

  7. NXT as Meeting Browser • Browser = display + signal indexing + search • NXT data displays: • synchronize with signal • highlight search results

  8. Issues • Already can't load all the ICSI data at once on some machines • NXT supports display of one meeting at a time but browsing may be over several meetings • Really complicated queries are often too slow for browser response times Key: Pre-indexing of query results, tailored data builds

  9. Conclusions • NXT available, free, open source, useful in a surprising number of ways http://www.ltg.ed.ac.uk/NITE

More Related