[Accessibility] Re: Can you help write a free version of HTK?

[Bill Cox]

Sorry about the delay in response.  I had a couple of days off.  Also,
sorry if I offended anyone.  In particular, I know practically nothing
about speech recognition, Simon's implementation, and such, so I do
make mistakes out of ignorance.

On Fri, Jul 9, 2010 at 8:56 AM,  <address@hidden> wrote:
> Hi!
>
> First of all: If you really want the HTK maintainers to change their
> license, I don't think writing angry mail is the best strategy.

The copyright holders are Microsoft.  I assumed Microsoft would
neither listen to nor care about anything you or I have to say.
However, if you think there is a chance, we should try.  Do you know
anyone who we could talk to?

> When we were evaluating the existing software for simon it's not that we
> didn't choose free alternative X because we were "tricked" by the HTK it's
> just that if we had chosen something else, simon would most likely have
> failed completely. If that means using the HTK to get started, so be it.

I'm all for using whatever tools are available to get a project going.
 I just want to help take Simon to the next step, and currently the
HTK non-redistribution clause is a major hangup.  I think we can live
with the Julius license.  I can put Simon in the Vinux repositories,
minus HTK, but in my experience, having a barrier like that means a
lot of potential Simon users wont get there.  I was not actually able
to install HTK without assistance from the HTK site admin, and I'm a
bit more persistent than most.  Also, I suspect Ubuntu, our upstream
repository, will not be very likely to include Simon with the HTK
dependency, and distros like Debian and Fedora are even more picky.
If it is possible to fix this issue, I'd make it a high priority at
this point, because I think Simon is close to being ready for lots of
users and contributors.

> Also, in simon 0.3 you can set up a complete voice recognition without
> ever touching the HTK.

Can we train without HTK?  The static models worked for the existing
vocabulary, but will it work well with custom commands?

>> Fortunately, we can freely read the HTK source code, and can learn how
>> it works.  We can then go rewrite it, and hopefully do a better job.
> Thats actually illegal. Nobody who did read the HTK source code is
> technically allowed to work on GHMM because of copyright issues.

That's not true.  You may not copy any of their code, and in a lawsuit
happy environments, we sometimes adopt a "clean room" approach to
rewriting software tools, where one group reads the old code, and
advises development done by a second group.  That minimises the chance
of actual copying.  However, if you're careful not to copy code, you
should be ok without a clean room approach.  Personally, I rarely find
code I want to copy anyway.  I'd generally rather just learn the
algorithms, and come up with my own implementation.  You may patent an
algorithm, but you can't copyright one.

> The law basically is that if you get your information from a copyright
> protected source (that doesn't explicitly allow you to use it to create a
> derived work) and create something similar then it boils down to the
> question: Could you have created your derived work without ever looking at
> the "original"? If the answer is no, than you violated the original works
> copyright.

Copyright law says I may not make a "derivative work" of copyrighted
material without permission.  That work does not in any way include
the ideas and algorithms implemented in the program.  Some licenses
restrict users from reading the code in order to learn how it works,
but the HTK does not.

> Well I think it's much more realistic to just use SPHINX instead of
> Julius/HTK which works _right now_ and shouldn't be that hard to integrate
> into the simon framework.
>
> The only reasons its not there yet are:
> 1. Limited time on my part
> 2. Very little documentation available online

Would Sphinx do as good a job as Julius?  I haven't dived into any of
the Sphinx code, but I have tried several Sphinx based programs in the
past, and all of them had trouble recognising speech at a rate that
would prove useful for real work.  Simon is basically the only mostly
open source program that has impressed me with it's recognition rate,
though I don't understand the details of why this is the case.

If switching to Sphinx would not hamper the recognition rate, I'd
support such a move.

By the way, I'm no open-source fanatic, though I am a fan.  My
computer dual-boots Windows and Vinux.  I have Skype, the Firefox
Flash plugin, and a proprietary closed-source "voxin" speech synthesis
program, and I write proprietary software for a living.  These
programs make me more productive, which is what counts right now.  I'm
very pragmatic about what we need to do to move forward, as I see you
were with the decision to use HTK.  I just think that from a pragmatic
point of view, we need to solve this license issue pretty soon, to
help Simon reach it's audience.

Bill