data/fisher-callhome-corpus/index.html

---
layout: documentation
title: Fisher / CALLHOME Parallel Corpus
---

    <div class="container">

      <div class="row">
        <div class="span8">
          <h1>Datasets</h1>
          <h2>Fisher / CALLHOME Spanish&ndash;English Parallel Corpus</h2>
          <span id="download">
            <a href="https://github.com/joshua-decoder/fisher-callhome-corpus/zipball/master">Download</a>
          </span>
        </div>
      </div>
      
      <hr />

      <div class="row">
        <div class="span8">

          <p>
            This paper describes the release of a set of English translations (obtained
            on <a href="http://mturk.com">Amazon's Mechcanical Turk</a>) and ASR lattice output
            (produced with <a href="http://kaldi.sf.net">Kaldi</a>). Together, this data supplements
            existing LDC datasets (in the form of audio and Spanish transcriptions), yielding a
            four-way parallel corpus for research in Spanish&ndash;English spoken language
            translation. 
          </p>

          <p>
            The LDC datasets that this dataset extends are as follows:
          </p>

          <p style="text-align: center"><center>
            <table style="border: 1px solid lightgray">
              <tr>
                <th></th>
                <th>Audio</th>
                <th>Transcripts</th>
              </tr>
              <tr>
                <td>Fisher Spanish</td>
                <td><a href="http://catalog.ldc.upenn.edu/LDC2010S01">LDC2010S01</a></td>
                <td><a href="http://catalog.ldc.upenn.edu/LDC2010T04">LDC2010T04</a></td>
              </tr>
              <tr>
                <td>CALLHOME Spanish</td>
                <td><a href="http://catalog.ldc.upenn.edu/LDC96S35">LDC96S35</a></td>
                <td><a href="http://catalog.ldc.upenn.edu/LDC96T17">LDC96T17</a></td>
              </tr>
            </table>
          </center></p>

          <p>
            If you use this dataset, please cite the following paper, which also contains a number
            of experiments to compare against:
          </p>

          <blockquote>
            <i>Improved Speech-to-Text Translation with the Fisher and Callhome Spanish&ndash;English
            Speech Translation Corpus</i> <br/>
            Matt Post, Gaurav Kumar, Adam Lopez, Damianos Karakos, Chris Callison-Burch and Sanjeev
            Khudanpur <br/>
            <a href="http://www.iwslt2013.org">IWSLT 2013</a> <br/>
            <a class="pdf" href="http://cs.jhu.edu/~post/papers/post2013improved.bib">PDF</a>
            <a class="bibtex" href="http://cs.jhu.edu/~post/papers/post2013improved.bib">BIB</a>
          </blockquote>

          <h2>Download & License</h2>

          The Fisher / CALLHOME corpus
          is <a href="https://github.com/joshua-decoder/fisher-callhome-corpus">hosted on
          Github</a>.  You can clone that, or download a release tarball by clicking the big green
          button above. The corpus is licensed under
          the <a href="http://creativecommons.org/">Creative
          Commons</a> <a href="http://creativecommons.org/licenses/by-sa/3.0/">Attribution-Sharealike
          3.0 Unported License</a> (CC BY-SA 3.0).

          <h2>Scores</h2>

          <p>
            Below are the best translation scores (case-insensitive BLEU-4) that have been reported
            on the provided test sets.  The Google results were recorded in the fall of 2011 (and
            are described in Post et al. (2012)).  Google does not have a Malayalam system.
          </p>

        </div>

        <div class="span4">
          <div style="border: 1px solid lightgray">
            <p style="text-align: center">
              <img width="250px" src="images/lattice.png"/><br/>
            </p>
            <p style="text-align: center">
              An example lattice from the dataset
            </p>
          </div>
        </div>
      </div>
    </div>