Trefnydd: A Speech Recognition Toolkit
Author: | Ivan A. Uemlianin |
---|---|
Contact: | ivan@llaisdy.com |
What is trefnydd?
Trefnydd is a toolkit for building speech recognition engines. Tools support the compilation of corpora and pronunciation dictionaries, creation of acoustic and language models, and generation of speech recognition engines compatible with the Media Resource Control Protocol (MRCP). For more on MRCP, see my notes on MRCP.
Trefnydd can be used either via its GUI, or by incorporating the tools directly into your own software.
For examples of GUI-based uses, plase see the Quick Start and User's Guides. For information on programmatic uses, please refer to the Developer's Manual.
Trefnydd is released under the GNU GPL license.
Requirements
Trefnydd is developed and tested on linux and Microsoft Windows. It has not been tested on other systems.
Requirements:
Essential:
For the Model Builder:
- HTK (I hope also to support SphinxTrain, and eventually to offer an interface mechanism to enable arbitrary model builders).
For the Engine Builder:
- HTK (I hope also to support Sphinx, and eventually to offer an interface mechanism to enable arbitrary recognition engine builders).
- htk2s3conv: this converts HTK acoustic models into Sphinx3 format.
- Cairo: This is not actually a requirement for Trefnydd to run, but the Engine Builder produces Sphinx4 .jar files, which can be used with Cairo as an MRCP speechrecog resource.
Download
Trefnydd can be downloaded from here: [trefnydd-0.0.3.tar.gz | trefnydd-0.0.3.zip]. The current release is 0.0.3. Older releases are available here.
Please read the Release Notes before using. Releases earlier than 0.1 should be regarded as 'request for comments' releases.
Installation
Trefnydd doesn't currently use distutils. Just unpack the tarball into a convenient home directory (the site-packages directory of your Python installation is best, if possible).
Usage
The trefnydd GUI can be started by running the trefnydd_gui.py script in the trefnydd/ directory. If trefnydd is installed in site-packages/, you should be able to take (copy/move/link/shortcut) this script and run it from anywhere. For example, on Windows you could have a shortcut to it on your desktop. Clicking the shortcut icon will fire up the Trefnydd GUI.
See the documents listed in the sidebar for guides on usage.
Release schedule
Version | Release Date | Added Features |
---|---|---|
0.0.1 | April 2007 | Skeleton SpeechCorpus Builder and Pronunciation Dictionary Builder |
0.0.2 | July 2007 | Skeleton Model Builder |
0.0.3 | December 2007 | Skeleton Engine Builder |