Welcome to this forum :) This post marks the transition of CMU Sphinx to Open Source. The current code will be placed, pre-release, under CVS here on SourceForge soon. It's not perfect by any means, but it is a completely working system -- at least on the tested platforms, and linux in particular.
There are some issues about getting sound set up, and about using the libraries, that I'm sure will come up right away. If you can record and play back audio on your machine, that's a good start.
kevin
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Anonymous
-
2000-01-31
Hi there --
What sort of acoustic and/or language model training is included with this release, if any?
I noticed that there are tools to do your own language modelling. Are there tools for acoustic training as well?
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
In this release, we have some not-great 4k-senone models trained from TIMIT. These aren't good enough. We're in the process of training better models from more data, and for both broadband and telephone speech.
We will be releasing the acoustic trainer also, but we're working up to that. Another point is that there isn't speaker adaptation in the current code yet -- the code is as we use it ourselves for our speaker-independent realtime systems like the Communicator and the MovieLine, though those use telephone models.
Regarding data, we actually have been using a lot from the
LDC (Linguistic Data Consortium), and collecting some of our own. The LDC allows licensors to distribute derived models, so it's OK for us to put out acoustic models trained from that. We're going to put out some of our own data too, but i think what we really need is a way for people to contribute data and make sure it remains free. Something like a SETI-at-home where you say some things at your desktop and these get analyzed and stored in a public database; these public data could then be used for anyone to train upon.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Anonymous
-
2000-02-01
Hi Lenzo,
>but i think what we really need is a way for
>people to contribute data and make sure it >remains free. Something like a SETI-at-home
>where you say some things at your desktop and
>these get analyzed and stored in a public
>database; these public data could then be used
>for anyone to train upon.
Cool idea!
--------------------------------------------------
Alan Cobb - alancobbREMOVETHIS@jps.net (remove REMOVETHIS)
MFC, GUI, Speech, C++ Programming Consultant, California, USA
--------------------------------------------------
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
We have a couple of things here: one is the CMU-Cambridge Statistical Language Modeling Toolkit, which is already available. The second is the web-based LMTool page, where you can upload some example sentences and it will build a language model.
We will put our a few more language models too, but it's important that we have a set of tools available for this. The CMU-Cambridge SLM is good for heavy lifting, and the LMTool for small domains right now, but in order to build language models on-the-fly we'll need to have some more code.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Licensing for CMU/CU LMTK. Can we get some licensing pinned down for this? I think Roni Rosenfeld (roni@cs.cmu.edu) and Philip Clarkson (prc14@eng.cam.ac.uk) will need to be consulted. As far as I can tell, there is no specific licensing assigned to either version 1 or version 2.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Well, all the source files contain the notice attached below. Basically you are free to use it for research purposes, which might be a clever way to avoid patent problems...
Isn't that specific enough?
/*=====================================================================
======= COPYRIGHT NOTICE =======
Copyright (C) 1996, Carnegie Mellon University, Cambridge University,
Ronald Rosenfeld and Philip Clarkson.
All rights reserved.
This software is made available for research purposes only. It may be
redistributed freely for this purpose, in full or in part, provided
that this entire copyright notice is included on any copies of this
software and applications and derivations thereof.
This software is provided on an "as is" basis, without warranty of any
kind, either expressed or implied, as to any matter including, but not
limited to warranty of fitness of purpose, or merchantability, or
results obtained from use of this software.
======================================================================*/
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
There are no patent issues! If one wanted to develop a commercial product from the codebase, they'd have to do their own research on the patents in the area to make themselves happy about such things.
I think awb spoke to Philip about this recently. We need Philip, Roni, and Cambridge's approval and we can make it a truly Open Source license. Outlook good.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Welcome to Open Discussion
Hi --
Welcome to this forum :) This post marks the transition of CMU Sphinx to Open Source. The current code will be placed, pre-release, under CVS here on SourceForge soon. It's not perfect by any means, but it is a completely working system -- at least on the tested platforms, and linux in particular.
There are some issues about getting sound set up, and about using the libraries, that I'm sure will come up right away. If you can record and play back audio on your machine, that's a good start.
kevin
Hi there --
What sort of acoustic and/or language model training is included with this release, if any?
I noticed that there are tools to do your own language modelling. Are there tools for acoustic training as well?
In this release, we have some not-great 4k-senone models trained from TIMIT. These aren't good enough. We're in the process of training better models from more data, and for both broadband and telephone speech.
We will be releasing the acoustic trainer also, but we're working up to that. Another point is that there isn't speaker adaptation in the current code yet -- the code is as we use it ourselves for our speaker-independent realtime systems like the Communicator and the MovieLine, though those use telephone models.
Regarding data, we actually have been using a lot from the
LDC (Linguistic Data Consortium), and collecting some of our own. The LDC allows licensors to distribute derived models, so it's OK for us to put out acoustic models trained from that. We're going to put out some of our own data too, but i think what we really need is a way for people to contribute data and make sure it remains free. Something like a SETI-at-home where you say some things at your desktop and these get analyzed and stored in a public database; these public data could then be used for anyone to train upon.
Hi Lenzo,
>but i think what we really need is a way for
>people to contribute data and make sure it >remains free. Something like a SETI-at-home
>where you say some things at your desktop and
>these get analyzed and stored in a public
>database; these public data could then be used
>for anyone to train upon.
Cool idea!
--------------------------------------------------
Alan Cobb - alancobbREMOVETHIS@jps.net (remove REMOVETHIS)
MFC, GUI, Speech, C++ Programming Consultant, California, USA
--------------------------------------------------
We have a couple of things here: one is the CMU-Cambridge Statistical Language Modeling Toolkit, which is already available. The second is the web-based LMTool page, where you can upload some example sentences and it will build a language model.
We will put our a few more language models too, but it's important that we have a set of tools available for this. The CMU-Cambridge SLM is good for heavy lifting, and the LMTool for small domains right now, but in order to build language models on-the-fly we'll need to have some more code.
Licensing for CMU/CU LMTK. Can we get some licensing pinned down for this? I think Roni Rosenfeld (roni@cs.cmu.edu) and Philip Clarkson (prc14@eng.cam.ac.uk) will need to be consulted. As far as I can tell, there is no specific licensing assigned to either version 1 or version 2.
Well, all the source files contain the notice attached below. Basically you are free to use it for research purposes, which might be a clever way to avoid patent problems...
Isn't that specific enough?
/*=====================================================================
======= COPYRIGHT NOTICE =======
Copyright (C) 1996, Carnegie Mellon University, Cambridge University,
Ronald Rosenfeld and Philip Clarkson.
All rights reserved.
This software is made available for research purposes only. It may be
redistributed freely for this purpose, in full or in part, provided
that this entire copyright notice is included on any copies of this
software and applications and derivations thereof.
This software is provided on an "as is" basis, without warranty of any
kind, either expressed or implied, as to any matter including, but not
limited to warranty of fitness of purpose, or merchantability, or
results obtained from use of this software.
======================================================================*/
There are no patent issues! If one wanted to develop a commercial product from the codebase, they'd have to do their own research on the patents in the area to make themselves happy about such things.
I think awb spoke to Philip about this recently. We need Philip, Roni, and Cambridge's approval and we can make it a truly Open Source license. Outlook good.