Jump to navigation
Who We Are
Policies & Guidelines
Learn at LTI
Explore Our Work
Back to the catalogue
If you have a problem in filling in this form, contact lti.catalogue AT gmail.com.
Fields marked with (
) are required.
The email is only for internal purposes.
the submission will not be considered until you confirm that you own this email.
You will receive a confirmation email upon submitting. The confirmation email might get into your junk email, so check your SPAM folder. If you do not receive an email, please contact us.
A proof for not being spam
This information will appear in the catalogue.
Natural Language Processing/Computational Linguistics
Information Retrieval, Text Mining and Analytics
Spoken Interfaces and Dialogue Processing
Keywords (comma separated, internal and not shown to public)
Direct Download Link
(If you provide a direct download link, please also provide the IP agreement, and the Required Acknowledgement)
<p><strong>Eesen</strong> is a toolkit to build speech recognition (ASR) systems in a <strong>completely end-to-end fashion</strong>. The goal of Eesen is to <strong>simplify</strong> the existing complicated, expertise-intensive ASR pipeline into a straightforward learning problem. Acoustic modeling in Eesen involves training <strong>a single recurrent neural network</strong> (RNN) which models the sequence-to-sequence mapping from speech to transcripts. Eesen <strong>discards the following elements</strong> required by the existing ASR pipeline:</p> <ul> <li>Hidden Markov models (HMMs)</li> <li>Gaussian mixture models (GMMs)</li> <li>Decision trees and phonetic questions</li> <li>Dictionary, if characters are used as the modeling units</li> <li><strong>...</strong></li> </ul> <p>Eesen is developed on the basis of the popular <a href="http://kaldi.sourceforge.net/">Kaldi</a> toolkit. However, Eesen is fully self-contained, requiring no dependencies from Kaldi to funciton. </p> <p>Eesen is released as <strong>an open-source project</strong> under the highly non-restrictive <strong>Apache License Version 2.0</strong>. We welcome community participation and contribution.</p> <p>For <strong>more information</strong>, please refer to our manuscript: <a href="http://arxiv.org/abs/1507.08240">EESEN: End-to-End Speech Recognition using Deep RNN Models and WFST-based Decoding</a>.</p>
Availability (e.g. source code, binary only, XML file, etc.)
Support Status (e.g. as-is, maintained, etc.)
Prerequisites (e.g. Windows XP, Java 1.6, etc.)
<p>Kaldi toolkit: <a href="http://kaldi.sourceforge.net/">Kaldi</a></p>
Required Acknowledgement (e.g. paper to cite)
<p><a href="http://arxiv.org/abs/1507.08240">EESEN: End-to-End Speech Recognition using Deep RNN Models and WFST-based Decoding</a></p>
might be helpful)
Contact (e.g. e-mail)
Additional Comments (internal and not shown to public)