A few years ago, I had a contract project to parse and organize several
terabytes of speech recognition engine logs captured from interactive voice
response applications.  The goal was to organize the extracted data in a
No-SQL columnar database, which would be used to improve recognition
accuracy.

The logs contained voice recordings, recognition results, and other
metadata about each recording. The logs came from multiple different
systems, so there were several different log formats to parse.

I asked Raul Miller, a J expert and frequent contributor to this forum, to
help me develop the parsing code using the J language. In order to meet
project deadlines, at times we spun up dozens of AWS virtual machines
running J, to do the data extraction and database build. Even then, the
project took most of a year.

The project was quite successful, and the logs helped to significantly
improve the recognition accuracy of a vendor's speech recognition engine.

Skip

Skip Cave
Cave Consulting LLC

On Tue, Oct 10, 2017 at 7:09 AM, Rob B <[email protected]> wrote:

> My many trips around the J home website have given me the impression that
> J is very much geared to maths and puzzles.
>
> I would be genuinely interested in reading about real world appllications
> of J.
>
> Regards, Rob Burns.
>
> > On 10 Oct 2017, at 02:43, Don Kelly <[email protected]> wrote:
> >
> > The possibility of a forum dealing with the extended variations would
> be  useful in that the present forum would then deal with more realistic
> problems  rather  than the more esoteric ones. We don't need to slow down
> operations, useful to all, to satisfy the needs of a few.
> >
> > Any digital expression is limited by machine word limits  and data
> limits. Machine limits exceed data limits in most cases- -useful where
> multiple operations   result in digital fuzz that is insignificant relative
> to real data fuzz.
> >
> > Don Kelly
> >
> >
> >> On 2017-10-09 12:29 PM, Raul Miller wrote:
> >> I think this proposal would require, at minimum, a rewrite of ve.c
> >>
> >> Also, each primitive which takes numeric arguments and/or produces a
> >> numeric result would need a test to make sure it behaved properly for
> >> each different mode of number handling.
> >>
> >> Want to take a crack at it?
> >>
> >> Thanks,
> >
> > ----------------------------------------------------------------------
> > For information about J forums see http://www.jsoftware.com/forums.htm
> ----------------------------------------------------------------------
> For information about J forums see http://www.jsoftware.com/forums.htm
>
----------------------------------------------------------------------
For information about J forums see http://www.jsoftware.com/forums.htm

Reply via email to