de.mpg.escidoc.pubman.appbase.FacesBean
English
 
Help Guide Disclaimer Contact us Login
  Advanced SearchBrowse

Item

ITEM ACTIONSEXPORT

Released

Talk

User friendly signal processing web services for annotators in AVATecH and AUVIS

MPS-Authors
http://pubman.mpdl.mpg.de/cone/persons/resource/persons4

Auer,  Eric
The Language Archive, MPI for Psycholinguistics, Max Planck Society;

Locator
There are no locators available
Supplementary Material (public)
There is no public supplementary material available
Citation

Auer, E. (2013). User friendly signal processing web services for annotators in AVATecH and AUVIS. Talk presented at the 23rd Meeting of Computational Linguistics in the Netherlands (CLIN 2013). Enschede, The Netherlands. 2013-01-18.


Cite as: http://hdl.handle.net/11858/00-001M-0000-000E-7825-6
Abstract
User friendly signal processing web services: The joint Max Planck Fraunhofer project AVATecH aims to support the very time intensive work of annotating audio and video recordings, letting signal processing modules (recognizers) assist annotators. -*- We designed a small, flexible framework where XML metadata describes input, output and settings of recognizers. Building blocks are audio and video files, annotation tiers and numerical data, packaged in simple formats. Text pipes allow flexibility in implementation details. The popular TLA ELAN software even lets the user control recognizers directly in their annotation environment: It generates consistent user interfaces for all installed recognizers based on their metadata. -*- We realized that full recognizers can be inconvenient to install for the user. Hardware, operating system and license requirements can add complexity. AVATecH supported intranet recognizers early, but those are limited by the need for shared network drives between user and server. -*- Recently, we developed a system where recognizers are run on a server using the free open source CLAM software. With suitable configuration, CLAM can run any command line tool, controlled by remote REST requests. On the user side, only a small proxy tool is installed instead of a real recognizer: The tool dynamically mimicks a recognizer based on the same metadata as before, but actually transfers data to a remote server and back where the real recognizer is installed. -*- We present details of our setup and workflow, with an outlook towards future extensions within the successor project, AUVIS.