Software in Appraisal of

matchmoaningΤεχνίτη Νοημοσύνη και Ρομποτική

17 Νοε 2013 (πριν από 3 χρόνια και 8 μήνες)

96 εμφανίσεις

The Potential for Use of
Voice Recognition
Software in Appraisal of
Oral History Tapes


Sonia Yaco

University of Wisconsin
-
Madison

School of Library and Information Studies

May 4, 2007

Introduction


Voice recognition software could
vastly increase intellectual access to
oral histories



70 years of development history



Transcription or keyword search
capability is tantalizingly close

The problem

Oral histories tapes


content unknown

Image Copyright @ Canadian Broadcasting Corporation (CBC) / Society Radio
-
Canada (SRC) 2005

The solution?

Transcription by voice
recognition
software

Image source: howstuffworks.com

Voice recognition
technology milestones

1876


Alexander Graham Bell
patents telephone


1939


The World's Fair features
Bell Labs' Voder (Voice
Operating Demonstrator).


Image source: http://www.keyboardmuseum.org/pre60/1930/voder.html

Voice recognition
technology milestones

1950’s


Recognition of sound waves of single digits

1959


Development of statistical prediction of word choice

1970’s


AT & T and IBM begin development


Image source: Photographs of formant 1 vs. formant 2 presentation of the digits.


(H. Davis, R. Biddulph, and S. Balashek, Automatic Recognition of Spoken Digits,

Journal Acoust. Soc. Am., Vol 24, No. 6, pp. 627
-
642, 1952.)

Methodology


My pilot test evaluated two types of
software:


Speaker dependent


Dragon
NaturallySpeaking


Speaker independent



AudioMining


Test Data


Seventeen interviews recorded on
high quality audio tape, converted
from analog cassette tapes to digital
.wav files


Interviewees were college educated
Caucasians, 60 to 80 years old


Native English speakers raised in the
Midwestern or eastern part of the
United States



Results of Dragon Naturally
Speaking Transcription


Test data sound clip


“As or a clean July 3 a in a any
as you know you and I on and
I am rocking in and where we
were a time that he is very
fast lane eight at him are
behind format with
evolutionary flat on said who
worked in the last time she
had all been with all of these
resources the heated rings in
their sieges if any action is
tiny and flashes of who defy a
knee in the last angle or for a
minute for everything he owns
or any there was actually
became one of our in our is
unknown after you”


Results of AudioMining
Transcription

“To pitch their 2004 between giveaways.
And at a list of the need to check
many sure and I give you a call, and
I may move off the table and where
we were at a time with which he just
very successfully is not a bench and
bar behind the divorce Reform Act,
which was quite evolutionary and it's
why on and so that was her work
there with pretty much done. She
had all of its goodwill and all of these
resources that she could bring to
bear. And she just finished a
tremendous job on a major piece of
legislation itself can be defined in
another age in the Atlanta life said
any move for protecting women at
work to protect the people during the
ongoing marriage, which actually
became one of our big arguments as
I know you all
--

started out because
like”


Findings


Dragon NaturallySpeaking low
accuracy rate when used with multiple
speakers



AudioMining more accurate but
expensive


Discussion


Neither software accurate enough to
be useful currently



Other new software shows great
potential for keyword searching of
audio and video material

Conclusion


Voice recognition software could vastly
increase intellectual access to oral histories



The technology has a 70 year development
history



Transcription or keyword search capability
is tantalizingly close to possible