GFT Group is developing a new speech recognition engine with superior
accuracy compared to current speech recognition engines such as Dragon
Naturally Speaking, ViaVoice, Microsoft Speech, and the open-source
SPHINX speech recognition engine. The engine should enable or enhance
rapid jumping to targets such as files, web sites, PowerPoint slides,
submenu selections and so forth on computers by simple voice command,
dictation of documents, and hands-free operation of computers,
cell-phones, automobile accessories and many other devices. Our target
is to achieve 100% accurate speaker-independent phoneme recognition in
the presence of typical background sounds such as car noises that do
not impair human speech recognition. We do not expect to solve the
homonym resolution problem for general unstructured human speech. The
effective speech recognition accuracy, the word error rate, of the
engine will be determined by the frequency of homonyms and
near-homonyms, words and phrases that sometimes sound the same, in the
recognized speech.
The Mathematician will help translate advanced concepts in human speech
to specific mathematical formulas that can be tested on human speech
data and, if successful, converted quickly to software for a real-time
commercial speech recognition engine written in a portable compiled
language such as ANSI C. The engine will include a Microsoft Speech
compatible wrapper.
The task is similar to the inference of mathematical formulas such as
differential equations from experimental data and from concepts
expressed in words, pictures, and rough mathematical formulas. It may
resemble, for example, the translation of Michael Faraday's ideas
about electricity and magnetism from the words and pictures that
Faraday used to a set of new differential equations by James Clerk
Maxwell. Experience with this process is most valuable for this
position but is not a requirement.
A strong knowledge of the physical processes and/or visual
representations corresponding to individual terms and factors within
terms in differential equations and other mathematical formulas should
be helpful. A strong knowledge of English verbal descriptions, words,
and phrases used for these physical processes and/or visual
representations should also be helpful. This should make it easier to
identify the mathematics -- for example construct a new differential
equation -- corresponding to features of data and advanced concepts
expressed in words and pictures.
Experience in the following areas may be helpful, but is in no way a
specific requirement of the position:
1. Non-linear differential equations
2. Classical invariant theory
3. Differential geometry and tensor analysis
4. Hidden Markov Model speech recognition methods
5. Adaptive signal processing
6. Music theory and practice
7. Acoustics of Speech Communication
8. Implementation of advanced mathematical algorithms in C or similar
languages
The Mathematician must be able to think creatively, "outside the
box", and have a solid foundation in advanced mathematics including
the ability to learn new areas rapidly as needed. An advanced degree
in mathematics, applied mathematics, or theoretical physics may be
helpful, but is not required. This is primarily a pencil and paper
task; computer skills (for example, knowledge of a computer algebra
system such as Mathematica) may be helpful but are not required.
The Mathematician will be an independent contractor and will need to
sign a non-disclosure agreement to perform the task. The Mathematician
will need to work closely with a scientist from GFT Group in a
collegial style. This is not implementing someone else's ideas but
requires shared creativity by skilled experts in complementary fields.
The opportunity to share license fees from the technology and inventor
status (on patent applications for example) may exist.
GFT Group is a private research and development and contracting
organization with several products in the speech recognition field.
For more information, see http://www.Petrana.net
Please submit a cover letter and a r
