← Back to team overview

speechcontrolteam team mailing list archive

Fwd: SpeechControl: The Anthology


---------- Forwarded message ----------
From: Jacky Alcine <jackyalcine@xxxxxxxxx>
Date: Thu, Jan 13, 2011 at 11:19 PM
Subject: SpeechControl: The Anthology
To: speechcontrol@xxxxxxxxxxxxxxxxxxx,
speechcontrol-devel@xxxxxxxxxxxxxxxxxxx, Danté <danteashton@xxxxxxxxx>

 SpeechControl is the brainchild of Manuela Popping, from the Netherlands.
In such a short period of time,she was able to accumulate, direct and have
joined a tram of individual s that are committed to the task of giving back
to the Ubuntu society. Her efforts led to start a wave of production that'll
change how Ubuntu users use their distributions, and possibly for all of
Linux. Such a feat is one that would be seized, exploited, and *
closed-sourced* by corporations, industries and other selfish individuals.
You're probably curious as how to *libopenmary*, *python-openmary*, and even
*speechcontrol-daemon* could ever attempt a feat as large as this.

 Truth is, they won't. But *wintermute* will. Wintermute (
https://www.launchpad.net/wintermute) is a project that the developers of
SpeechControl have unknowingly signed up for. This is something I must
deeply apologize. Such a project is one that should I have asked you before
even considering doing such a thing. The primary reason why I felt that
SpeechControl Developers were qualified was this: the amount of work,
interaction and comfortability we had with each other so far; I felt that it
was safe enough for me not to speak on your behalf, but to enlist you into a
project that you'd find beyond interesting, and might freak you out. It took
me off guard as how to this application started as well. Due to a bit of
discussion though, the code isn't fully mine to have it uploaded to
Launchpad, this has to be spoken and mulled over once more by the admins.
But for now, I will link you to a set of websites for you to being wrapping
your mind about the journey as SpeechControl Developers will embark.

 First off, you need to know what a neural network is, skim about the
Wikipedia page for more information. Secondly, you need to understand the
different between *AI* (Artificial Intelligence) and *AGI* (Artificial
General Intelligence). The contrasting attributes of these two will heavily
outline our goals.

 For now, we need to need to achieve the following *after* we finish working
on openMary or after it hits version 0.3:


   *Wiki Knowledge Base*
   Initially the idea was to use a predefined ontology (look this up as
   well), but we'd rather it be up to date and refreshable. So the idea behind
   this snazzy devil is simple. Instead of having every single piece of data
   stored into one chunky file, a directory with individual file as if they
   were defined as nodes within the Wikipedia database are stored and are
   readily more available than having it squished into one store. And instead
   of having a file that defines each word and its corresponding file; each
   word's MD5 hash can be used instead. For words with different meanings, it
   can be enumerated within their corresponding data file. This is the
   preliminary blueprint for project *winter-kb*. The knowledge base will
   have a plethora of sections, of which three stores are implemented.


      The *short-term*, or more immediate data is what's loaded in memory
      for more rapid access and can be flashed for testing purposes.
This data is
      what we'd compare to as the things we'd remember *all* the time; it's
      considered “always on our mind”.

      The *long-term*, or the repository of “infinite knowledge”* *is what
      makes the magical click of the knowledge base, this is where all of the
      information from Wikipedia is stored. This data should be
updated as often
      as possible with a server that we can set up, hosting a cache of
      information so that the need of working back and forth with huge
masses of
      data isn't done on the end-user client.

      The *dynamic*, or the user defined memory, is the more interesting
      aspect of the knowledge base. Kept separate from the long-term
store; this
      is the information indexed off of the user's computer. This information
      should remain on the user's computer unless they choose to push their
      information to our system. The data here should be checked heavily for
      corruption, spell-checking and proper usage so that the user doesn't have
      any irregular interactions.

   The back-end of this should allow each and every object, word, and topic
   to be accessible via a specialized, numeric machine-readable syntax, similar
   to memory locations. As you can see, such data interactions must be speedy,
   snappy and efficient. We could implement a wizard that, as it guides the
   user through installation of Wintermute, index some immediate information
   for the dynamic store and set up topics that are likely to fall within the
   short-term, and perhaps the kind of extra information that they'd like to
   download to the long-term. It's all about the knowledge!


   *Natural Language Composition*
   This is a bit more difficult task than the previously one described. This
   project consists of enabling Wintermute to speak to a user in his or her
   native language. A background of linguistics and NLP (Natural Language
   Processing) will be required for this section. Even after the NLC system has
   the rules of grammar down for a specific languages (grammar-pack), it must
   learn other aspects of the language specified (speech-pack), and those two
   can be put together can shipped off via a langpack. There should be a
   front-end or a simple means of having translations and sample sentences be
   pushed up to Wintermute Translations for scrutiny and analysis. The *
   grammar-pack*, *speech-pack*, and *langpack* are all archives in the 7z
   format, that of which will be included if not found on the user's system.


   A *grammar-pack* consists of the parts of speech as defined for a
   specified language. This way, the sentence “The blue table” and “La mesa
   azul” are both converted properly back to machine-readable syntax, and visa
   versa. The file should be compressed and

   A *speech-pack* consists of the sample setups that conversation would (or
   would be considered) to normally follow in a certain language. For example,
   this is a typical English greeting:

   Wintermute: “Hello, how are you?”
   User : “I'm fine, thanks for asking.”

   As for in Haitian Creole (my country of origin):

   Wintermute : “Sak pase?” (What's up? / Literal: What's cooking?)
   User: “Nap boule, Wintermute.” (Nothing much / Literal: Nothing's

   The phrases are different, but they both do the same thing; they *greet*.
   Simple interactions of the sort can be created, compiled and packed by a
   user and shipped off to Wintermute Translations to scrutinize.

   A *langpack* is a fusion of the *grammar-pack* and *speech-pack*, just to
   make porting a bit more simpler for users. The exception is that it's like a
   dummy .deb; it links speech to grammar and permits updates and edits.

Now that you have a initial understanding of what it is I ask from you; I
have to add this. From now on, at the start of this project, all e-mails
will be encrypted and signed by the SpeechControl key (that of which I'll
personally e-mail to everyone). This key is *not to be shared* or saved
on-line, at any cost. If you lose this key, you *must* contact the admins or
enter the developers' channel and request for me to send you a new copy. The
reason, once stated before, for these security measures are to ensure that
the code, in its early stage, remains open-sourced and not hijacked by a
money-grubber. We'll work on making a derivative of the GPLv3 license that
complies to our needs and copyrighting, but as for now, this project is
something that only SpeechControl should refer to. You can mention it, but
do not, under the maximum circumstances, release any code to anyone without
consulting the admins.

 *Thank you.*

NOTE: We have a general team meeting the 21st of January, 20:00 Ubuntu Time
(UTC). Be there!

*Jacky Alcine* Launchpad <https://launchpad.net/%7Ejackyalcine> |
Blog <http://jdevelopthis.blogspot.com>
fn:Jacky Alcine

Attachment: signature.asc
Description: PGP signature