Regarding google summer of code 2015

[Abhilash Mhaisne]

Sir / Madam,
                   I am an engineering student, and gsoc aspirant. I want
to develop a easy-to-use voice recognition system, which would be capable
of judging even minor changes in accents, for fedora systems.
                   I wish to implement this system in python, making use of
the open source CMU Sphinx system. I would love to do this under a mentor
for gsoc 2015.

Regards!

Abhilash Mhaisne

[Nico Kadel-Garcia]

On Tue, Feb 17, 2015 at 5:49 AM, Abhilash Mhaisne
<abhilashmhaisne(a)gmail.com&gt; wrote:
>  Sir / Madam,
>                     I am an engineering student, and gsoc aspirant. I want to
>  develop a easy-to-use voice recognition system, which would be capable of
>  judging even minor changes in accents, for fedora systems. 
And I'd like a pony, with pretty wings that can fly to Jupiter and
makes cookies.

More seriously, voice recognition is one of the great challenges of
the last 50 years of speech analysis, approached by large and small
companies and research labs around the world. "Judging even minor
changes in accents" is a nightmare in real work: every commercially
available computer microphone system for the last few decades has made
the same basic mistake for collecting speech. They completely screw up
"plosives", sounds like "b" and "p" that have a lot of high
frequency
information that gets completely munged by the digitization and
undersampling. And do not get me *started* on the ongoing fallacy that
"if we just collect more speech samples, we'll somehow be able to
analyze them". I'm aware of companies with software patents spending
many millions of investment money in the work, and they're still
getting their ass handed to them in the marketplace by smaller, local
apps that handle basic speech locally and *don't try to get fancy*.

And in case it's not clear: I did artificial hearing research for a
dozen years, designing electronics for cochlear implants. Every modern
microphone=>speech analysis system screws up the plosives, for a stack
of reasons we could discuss elsewhere.


>                     I wish to implement this system in python, making
> use of
>  the open source CMU Sphinx system. I would love to do this under a mentor
>  for gsoc 2015.
>
>  Regards!
>
>  Abhilash Mhaisne 

See above. "voice recognition" and accent immunity is not a "python"
sort of problem, it's a massive ongoing computer and speech research
issue. Pick something *smaller*.

[Abhilash Mhaisne]

Oh my! Thank you Mr. Nico! :) I will pick up a smaller project for google
summer of code. However, I am too much obsessed with voice recognition
systems. But now I understand that it is not a small issue. Someday, I will
surely work on speech synthesis. As you mentioned above, it is certainly
not doable in small span of SoC. Thanks once again! :)

On Tue, Feb 17, 2015 at 6:01 PM, Nico Kadel-Garcia <nkadel(a)gmail.com&gt; wrote:

>  On Tue, Feb 17, 2015 at 5:49 AM, Abhilash Mhaisne
>  <abhilashmhaisne(a)gmail.com&gt; wrote:
>  > Sir / Madam,
>  >                    I am an engineering student, and gsoc aspirant. I
>  want to
>  > develop a easy-to-use voice recognition system, which would be capable of
>  > judging even minor changes in accents, for fedora systems.
>
>  And I'd like a pony, with pretty wings that can fly to Jupiter and
>  makes cookies.
>
>  More seriously, voice recognition is one of the great challenges of
>  the last 50 years of speech analysis, approached by large and small
>  companies and research labs around the world. "Judging even minor
>  changes in accents" is a nightmare in real work: every commercially
>  available computer microphone system for the last few decades has made
>  the same basic mistake for collecting speech. They completely screw up
>  "plosives", sounds like "b" and "p" that have a lot of high
> frequency
>  information that gets completely munged by the digitization and
>  undersampling. And do not get me *started* on the ongoing fallacy that
>  "if we just collect more speech samples, we'll somehow be able to
>  analyze them". I'm aware of companies with software patents spending
>  many millions of investment money in the work, and they're still
>  getting their ass handed to them in the marketplace by smaller, local
>  apps that handle basic speech locally and *don't try to get fancy*.
>
>  And in case it's not clear: I did artificial hearing research for a
>  dozen years, designing electronics for cochlear implants. Every modern
>  microphone=>speech analysis system screws up the plosives, for a stack
>  of reasons we could discuss elsewhere.
>
>
>  >                    I wish to implement this system in python, making use
>  of
>  > the open source CMU Sphinx system. I would love to do this under a mentor
>  > for gsoc 2015.
>  >
>  > Regards!
>  >
>  > Abhilash Mhaisne
>
>
>  See above. "voice recognition" and accent immunity is not a "python"
>  sort of problem, it's a massive ongoing computer and speech research
>  issue. Pick something *smaller*.
>  --
>  devel mailing list
>  devel(a)lists.fedoraproject.org
>  https://admin.fedoraproject.org/mailman/listinfo/devel
>  Fedora Code of Conduct: http://fedoraproject.org/code-of-conduct

[Kevin Kofler]

Nico Kadel-Garcia wrote:
>  See above. "voice recognition" and accent immunity is not a
> "python"
>  sort of problem, it's a massive ongoing computer and speech research
>  issue. Pick something *smaller*. 
I would also add to that that Fedora would probably be the wrong mentoring 
organization for such a project to begin with, even if it were doable. These 
things need to be worked on with upstream projects. But they are very likely 
to reject this proposal for the same reasons. Improving the quality of the 
speech recognition, if it's possible at all, has to be done inside the C 
(Pocketsphinx) or Java (Sphinx4) code of CMU Sphinx. Trying to do it in a 
higher layer in Python doesn't make any sense whatsoever.

        Kevin Kofler

[Abhilash Mhaisne]

Oh I see. Did not know about these. I put it on fedora just cause it's my
favorite distro. Thank you!

On Tue, Feb 17, 2015 at 6:42 PM, Kevin Kofler <kevin.kofler(a)chello.at&gt;
wrote:

>  Nico Kadel-Garcia wrote:
>  > See above. "voice recognition" and accent immunity is not a
> "python"
>  > sort of problem, it's a massive ongoing computer and speech research
>  > issue. Pick something *smaller*.
>
>  I would also add to that that Fedora would probably be the wrong mentoring
>  organization for such a project to begin with, even if it were doable.
>  These
>  things need to be worked on with upstream projects. But they are very
>  likely
>  to reject this proposal for the same reasons. Improving the quality of the
>  speech recognition, if it's possible at all, has to be done inside the C
>  (Pocketsphinx) or Java (Sphinx4) code of CMU Sphinx. Trying to do it in a
>  higher layer in Python doesn't make any sense whatsoever.
>
>          Kevin Kofler
>
>  --
>  devel mailing list
>  devel(a)lists.fedoraproject.org
>  https://admin.fedoraproject.org/mailman/listinfo/devel
>  Fedora Code of Conduct: http://fedoraproject.org/code-of-conduct
>