CMU Sphinx / News: Recent posts

New release: sphinxbase-0.8, pocketsphinx-0.8 and sphinxtrain-0.8

We are pleased to announce that today a pack of CMUSphinx packages was released:

For the download links see:

The biggest update of this release is a new sphinxtrain. The code sharing between sphinxbase and sphinxtrain significantly increased bringing more consistent codebase and interface, accurate memory management and increased usability.... read more

Posted by Nickolay V. Shmyrev 2012-12-16

CMUSphinx 0.7 Is Released

We are pleased to announce the availability of the updated CMUSphinx toolkit. You can find updated sphinxbase, pocketsphinx, sphinxtrain, cmuclmtk and sphinx4 in downloads section

Major changes include

  • Sphinxtrain actively uses sphinxbase functions
  • Training is more user-friendly
  • Various advanced training techniques are implemented
  • Pocketsphinx is way faster on big FSG grammars
  • Many bug fixes and user-friendly improvements... read more
Posted by Nickolay V. Shmyrev 2011-04-16

Sphinx4-1.0beta5 Released

New Features and Improvements:

* Alignment demo and grammar to align long speech recordings to
  transcription and get word times
* Lattice grammar for multipass decoding
* Explicit-backoff in LexTree linguist
* Significant LVCSR speedup with proper LexTree compression
* Simple filter to drop zero energy frames
* Graphviz for grammar dump vizualization instead of AISee
* Voxforge decoding accuracy test
* Lattice scoring speedup
* JSAPI-free JSGF parser... [read more](/p/cmusphinx/news/2010/08/sphinx4-10beta5-released/)
Posted by Nickolay V. Shmyrev 2010-08-31

Spanish Voxforge Model

As of today, CMUSphinx project is proud to present you the Spanish acoustic model. It was trained on the acoustic data to support both wideband recognition of microphone speech recordings and narrowband recognition of the telephone speech. Now you can build Spanish IVR using CMUSphinx tools.

You can find the model on our download page read more

Posted by Nickolay V. Shmyrev 2010-08-28

PocketSphinx 0.6.1 release

This is a bugfix release, addressing a number of important issues in the 0.6 release. Specifically:

* The GStreamer plugin was broken with old versions of GStreamer (such as the one shipped on Nokia Internet Tablets) which did not accept BSD as a valid license type.
* Runtime performance of the statistical LM based decoder was significantly worse than 0.5 when not using phoneme lookahead.  It is now about 10% faster and also uses less memory.
* The FSG decoder now consumes drastically less memory and CPU power.  (It is still not as good as it should be, but no longer completely embarrassing)
* The Python modules were undocumented.  Docstrings compatible with epydoc have been added.
* Raw audio logging (-rawlogdir) was broken.
* Adding new words works properly, including words with unknown triphones (which never worked before).
* The SphinxBase Python module can now use class-based language models.
* The pitch estimation utility (sphinx_pitch) now builds properly on Win32.
* Some problems with the N-Gram iterator API have been fixed.
* Bestpath search has been fixed to handle recognition failure gracefully.... [read more](/p/cmusphinx/news/2010/07/pocketsphinx-061-release/)
Posted by Nickolay V. Shmyrev 2010-07-30

PocketSphinx 0.6 release

We are pleased to announce the long-awaited PocketSphinx 0.6 release, including SphinxBase 0.6. This release corresponds to SVN revision 9898.

PocketSphinx is a small-footprint continuous speech recognition system, freely licensed under a simplified BSD license, suitable for handheld and desktop applications. It features:

* Cross-platform: Linux, Windows, Mac OS X, iPhoneOS
* Experimental support for Nokia S60v3 and Windows Mobile
* Support for semi-continuous, phonetically-tied, and fully continuous acoustic models
* Model footprint on disk of about 10MB per language
* Memory footprint under 20MB for medium-vocabulary continuous recognition
* Trigram language models and JSGF finite-state grammars
* Acoustic models for English and Mandarin
* Small language models for English and Mandarin (simplified and traditional characters)
* Python language bindings
* GStreamer multimedia framework integration... [read more](/p/cmusphinx/news/2010/03/pocketsphinx-06-release/)
Posted by Nickolay V. Shmyrev 2010-03-21

Sphinx4-1.0 beta 4 released

Congratulations with the new release.

Get it here:

New Features and Improvements:

* Large arbitrary-order language models
* Simplified and reworked model loading code
* Raw configuration and and demos
* HTK model loader
* Many code optimizations
* JSAPI-independent JSGF parser
* Noise filtering components
* Lattice rescoring
* Server-based language model... [read more](/p/cmusphinx/news/2010/03/sphinx4-10-beta-4-released/)
Posted by Nickolay V. Shmyrev 2010-03-01

Phonetically Tied Mixtures (with models)

Support for phonetically-tied mixture acoustic models has been added to the Subversion repository for SphinxTrain, Sphinx3, and PocketSphinx. Briefly, phonetically-tied mixture models are somewhere between semi-continuous and fully-continuous models, offering most of the speed of the former combined with the ability of the latter to effectively use large amounts of training data.

Parameter settings for training PTM models are present in the template sphinx_train.cfg file created by SphinxTrain, and can be enabled by setting $CFG_HMM_TYPE to “.ptm.”. The development version of PocketSphinx will automatically recognize PTM models, while Sphinx3 requires you to add “-senmgau .ptm.” to the command line.... read more

Posted by Nickolay V. Shmyrev 2010-01-20

Sphinx4-1.0beta3 released

We are happy to announce the new sphinx4 release. This is still a development version, so bug reports and testing are very appreciated.


New Features and Improvements:

* BatchAGC frontend component

* Complete transition to defaults in annotations

* ConcatFeatureExtrator to cooperate with cepwin models

* End of stream signals are passed to the decoder for end of stream handling... [read more](/p/cmusphinx/news/2009/08/sphinx4-10beta3-released/)
Posted by Nickolay V. Shmyrev 2009-08-17

SphinxTrain 1.0 Released

SphinxTrain is the acoustic model training system for the Sphinx family of continuous speech recognition systems.

After years of not having an actual release of SphinxTrain, it was time to make one, in anticipation of potentially restructuring the training code. This trainer can produce acoustic models for all versions of Sphinx, and supports VTLN, speaker adaptation and dimensionality reduction.

Future releases will support discriminative and speaker-adaptive training, and will be more closely integrated with the Sphinx decoders.

Posted by David Huggins-Daines 2009-02-12

Sphinx-4 1.0 beta2 released

It is with great pleasure that we announce the 1.0 beta2 release of Sphinx-4, which is a state-of-the-art, speaker-independent, continuous speech recognition system written entirely in the Java programming language:

In this release, we have provided the following new features and improvements:

  • new much cleaner and more robust configuration system
  • refactored to make use of java5 language features
  • new API to setup speech recognizers without any xml
  • improved feature extraction (better voice activity detection, many bug-fixes)
  • Cleaned up some of the core APIs
  • 'include' and 'extends' for configuration files
  • better JavaSound support
  • fully qualified grammar names in JSGF (Roger Toenz)
  • support for dictionary addenda in the FastDictionary (Gregg Liming)
  • added batch tools for measuring performance on NIST corpus with CTL files
  • many performance and stability improvements

You're welcome to post your questions, comments, and feedback to the CMU Sphinx4 Forum:

We can also be reached at


The Sphinx-4 Team

Posted by Holger Brandl 2009-02-07

Sphinx3 0.8 released

This is largely a maintenance release, to bring Sphinx3 up to date with the current version of SphinxBase.

Posted by David Huggins-Daines 2009-01-01

PocketSphinx 0.5.1 and SphinxBase 0.4.1

These co-ordinated bugfix releases address a number of problems with the 0.5/0.4 release, including some compilation issues. As well, a number of features which were missing have been reintroduced.

For a full list, see the changelogs on the download page. We apologize for the delay in making them available.

Posted by David Huggins-Daines 2008-11-29

PocketSphinx: 0.5 release

Sphinx is a speaker-independent large vocabulary continuous speech recognizer released under a BSD style license. It is also a collection of open source tools and resources that allows researchers and developers to build speech recognition systems.

The newest stable version of PocketSphinx is now available. This release is about 50% smaller and up to 18% faster than the previous one, and introduces a new, re-entrant and "modern" API.

Posted by David Huggins-Daines 2008-07-08

cmudict.0.7a release

This is the first public release of cmudict since version 0.6d (11/1998). Version 0.7a adds about 4k new words, but more importantly includes a large number of corrections to existing entries.
Also part of this release is cmudict.0.7a_SPHINX_40, a Sphinx-ready version of the dictionary. You can download it from

Posted by Alex Rudnicky 2008-02-19

New IRC channel and documentation wiki

An IRC channel for CMU Sphinx discussion and development has been created on freenode. To participate, go to and join #cmusphinx.

We have also created an official documentation wiki on the project website.

Posted by David Huggins-Daines 2007-12-20

PocketSphinx 0.4.1 bugfix release

Due to some unforseen issues with the Windows port and the Python module in PocketSphinx 0.4, we have made a bugfix release. This is particularly recommended for people on the Windows platform. It incorporates the patch pocketsphinx-0.4-pythonfix.diff released earlier.

This corresponds to Subversion revision 7200.

Posted by David Huggins-Daines 2007-08-23

Sphinx3 0.7 released

Sphinx is a speaker-independent large vocabulary continuous speech recognizer released under a BSD style license. It is also a collection of open source tools and resources that allows researchers and developers to build speech recognition systems.

After a year since the last release candidate, Sphinx3 0.7 is finally available. This version adds integrated N-best list generation and rescoring for all modes of operation, HTK word lattice output, much improved phoneme recognition, faster and more accurate flat-lexicon decoding, and various other improvements.... read more

Posted by David Huggins-Daines 2007-08-21

PocketSphinx 0.4, SphinxBase 0.3 released

Sphinx is a speaker-independent large vocabulary continuous speech recognizer released under a BSD style license. It is also a collection of open source tools and resources that allows researchers and developers to build speech recognition systems.

The latest release of PocketSphinx is now available. It requires SphinxBase 0.3, which has been released simultaneously. This release includes better support for Windows CE (thanks to Marco Contenti), better support for Mac OS X and Solaris, better acoustic models, and all sorts of other better stuff.

Posted by David Huggins-Daines 2007-08-17

PocketSphinx 0.3, SphinxBase 0.2 released

New versions of PocketSphinx and SphinxBase are now available.

This version of SphinxBase will be used as the base for the next version of Sphinx3. It includes many updates to the utility library, the feature extraction code, and the dynamic feature computation code.

This version of PocketSphinx is quite different from the last one. It is considerably smaller due to the elimination of large amounts of redundant and legacy code. It is also up to 20% faster in some cases. The command-line arguments have been synchronized with Sphinx3, and support for Sphinx2-format acoustic models has been removed.

Posted by David Huggins-Daines 2007-01-30

PocketSphinx, SphinxBase in SVN now compile with eVC 4.0

I've added project and workspace files to sphinxbase and pocketsphinx that (should) enable them to compile and link for Windows CE using Embedded Visual C++ 4.0. This is, however, totally untested. We are looking for someone who understands Sphinx and WinCE moderately well to help get this running.

Posted by David Huggins-Daines 2006-11-06

French language acoustic and language models released

The speech group at LIUM (Universite du Maine, Le Mans, France) has prepared a set of acoustic and language models created from broadcast news in French. The models are being released as part of the cmusphinx project. The models, both acoustic and language, are fully compatible with Sphinx-3. Please check the README file distributed with the package for more detailed information about the settings for best performance.... read more

Posted by Evandro B. Gouvea 2006-10-18

PocketSphinx 0.2.1 Released

This is a bug-fix release to make fixed-point computation work, which was inadvertently broken in the 0.2 release.

Posted by David Huggins-Daines 2006-10-14

Sphinx3 in Subversion now requires SphinxBase

As of revision 5979, Sphinx3 now requires SphinxBase in order to build and run. A pre-split version is available in

If you are a developer using Subversion, just make sure that you have checked out the 'sphinxbase' module in the same parent directory as 'sphinx3'. If you simply pull the entire cmusphinx tree, it should work fine.... read more

Posted by David Huggins-Daines 2006-08-06

PocketSphinx 0.2 Released

This is the first real release of PocketSphinx. PocketSphinx is a derivative of Sphinx-II designed for embedded and handheld systems.

See the release notes for more information.

This release corresponds to revision 5967 in Subversion.

Posted by David Huggins-Daines 2006-08-05