Data processing: speech signal processing – linguistics – language – Linguistics – Translation machine
Reexamination Certificate
2000-12-23
2004-05-18
{haeck over (S)}mits, T{overscore (a)}livaldis Ivars (Department: 2654)
Data processing: speech signal processing, linguistics, language
Linguistics
Translation machine
C704S227000, C704S277000
Reexamination Certificate
active
06738738
ABSTRACT:
BACKGROUND OF THE INVENTION
1. Field of the Invention
This invention relates to the field of phonetics. In particular, the invention relates to technologies for transforming pronunciations appropriate for American English into pronunciations appropriate for British English.
2. Description of the Related Art
A. Notation
Before turning to definitions, some notational concerns will be addressed. A standard notational alphabet, the International Phonetic Alphabet (IPA) can be used to represent the pronunciation of words using phonemes. However, the IPA uses symbols that are difficult to represent easily in ASCII systems and further many of the symbols lack appropriate representational glyphs in standard computer fonts. (Newer systems that handle Unicode can represent IPA symbols directly and frequently include newer fonts with appropriate glyphs for IPA symbols.) Accordingly, it is more convenient and has become industry standard practice to use the Computer Phonetic Alphabet (CPA) in computer speech recognition and pronunciation generation tools such as “autopron”, from Nuance Communications, Menlo Park, Calif. and “namepro”, from E-Speech Corporation, Princeton, N.J.
The CPA has the advantage that it can be represented using standard ASCII characters using the glyphs in commonly available fonts. The following tables show the correspondence between CPA and IPA symbols for American English and British English.
TABLE 1
American English: Computer Phonetic Alphabet (CPA) to
International Phonetic Alphabet (IPA) Correspondence
CPA
Example
IPA
CPA
IPA
CPA
IPA
Vowels
Stops
Fricatives
i
fl
ee
t
p
f
I
d
i
mple
t
T
e
d
a
te
k
s
E
b
e
t
b
S
a
c
a
t
d
v
aj
s
i
de
g
D
Oj
t
oy
Flaps
z
{circumflex over ( )}
c
u
t
!
Z
u
bl
ue
Nasals
h
U
b
oo
k
m
Approximants
o
sh
ow
n
j
O
c
augh
t
g~
r
A
f
a
ther, cot
Affricates
w
aw
c
ou
ch
tS
l
*r
b
i
rd
dZ
*
a
live
TABLE 2
British English: Computer Phonetic Alphabet (CPA) to
International Phonetic Alphabet (IPA) Correspondence
CPA
Example
IPA
CPA
IPA
CPA
IPA
Vowels
Stops
Fricatives
i
b
ea
n
p
f
I
b
i
n
t
T
e
b
a
ne
k
s
E
b
e
t
b
S
a
b
a
t
d
v
A
f
a
ther
g
D
@
c
o
t
Flaps
z
O
c
au
ght
!
Z
o
g
o
Nasals
h
U
b
oo
k
m
Approximants
u
t
oo
t
n
j
{circumflex over ( )}
c
u
p
g~
r
3
b
ir
d
Affricates
w
*
a
live, ride
r
tS
l
aj
f
i
ve
dZ
Oj
b
oy
aw
c
ow
i*
b
eer
e*
b
ear
u*
p
oor
Throughout the remainder of this document, the CPA symbols will be used to represent phonemes in transcriptions. When relevant, transcriptions written in CPA symbols will be identified as corresponding to British English (UK) or American English (US) if it is not clear from the context and it is relevant to understanding the material. Additionally, to minimize confusion, US English conventions for spelling and style will be used throughout the body of this specification, except in examples and rules. Additionally, the UK CPA forms are used for Australian and New Zealand pronunciations.
The range of possible sounds that a human being can produce by moving the lips, tongue, and other speech organs, are called phones. These sounds are generally grouped into logically related groups, each a phoneme. In a given language only certain sounds are distinguished (or distinguishable) by speakers of the language, i.e. they conceptualize them as different sounds. These distinguishable sounds are phonemes. In fact, a phoneme may be defined as a group of related phones that are regarded as the same sound by speakers. The different sounds that are part of the same phoneme are called allophones (or allophonic variants).
Returning to notation issues, the phonemic transcription of a word will be shown between slashes (“/ /”). For clarity, the glyph “·” will be placed between each phoneme in the transcription, e.g./k·O·r·n·*r/ for “corner” (US), to represent the space character visibly. In many computer programs a space character is used to represent the boundary between phonemes; however, in a printed publication using the standard glyph for the space character, “ ”, might lead to ambiguities, e.g. between /*r/ and /*·r/ (US), etc.
If used, phonetic transcriptions will be shown in brackets (“[ ]”). Phonetic transcriptions distinguish between the different phones that are allophones of the phoneme.
B. Role of Phonemic Transcriptions in Speech Software
Speech recognizers (both speaker independent and speaker dependent varieties) rely on pronunciations to perform recognition. For example, in order for the Nuance™ speech recognition software from Nuance Communications, to recognize a word in a recognition grammar, a pronunciation (e.g. phonemic transcription) must be available. To support recognition, Nuance provides a large phonemic dictionary that includes pronunciations for many American English words. The content of the dictionary typically excludes proper nouns and made up words, e.g. “Kodak”; however, there may be extensions for particular purposes, e.g. for US equity issues (stocks).
Additionally, Nuance provides an automated tool, “autopron”, that attempts to generate (simply from the spelling of the word) a usable pronunciation. Other companies, e.g. E-Speech, specialize in providing software that they claim can do a better job at generating such pronunciations.
Symmetrically, a good pronunciation is also important to producing good synthesized speech (or in the case where a human is reading a script, providing the human with extra guidance about the correct pronunciation). Thus, a useful phonemic transcription is important to many aspects of computer speech technology.
C. British English and American English
Although American English and British English share a common origin, there are significant differences in grammar (word choice, vocabulary, spelling, etc.), pronunciation, and text normalization (e.g. time formats, data formats, etc.). One can typically purchase an electronic dictionary of British English, e.g. for use in spell checking, or even a phonetic one for use with products such as the Nuance speech recognition system. However, such a pronunciation dictionary assumes that materials have already been prepared in British English form.
For example, given a particular word like “attorney” in a production script for a voice application (e.g. yellow pages), that was prepared for American English speakers there are several problems. First, if presented a list of options, “attorney” will sound awkward to a British native since they expect the term “solicitor” (or perhaps if trying to get out of gaol a “barrister”). Similarly, the native British speaker is unlikely to provide the verbal command “attorney” to the speech recognition system. Lastly, even if the British speaker did provide the word “attorney”, the pronunciation will be different from the one used by Americans. This also has an impact on the recording of the program script where prompts for categories such as “attorneys” would need to be re-recorded.
These problems may be further exacerbated in the realm of proper nouns, e.g. names and places, as well as made up words, e.g. company names, movie/book titles, etc., where even if a British English dictionary were provided the term would not likely be present.
D. Noting Stress in CPA
Presently, (as seen above in Tables 1 and 2) the CPA does not support the representation of stress within a word. This limits its usefulness (as compared to IPA representations) in designating differences in pronunciation. For example “advertisement” is pronounced in US English with the stress on the penultimate syllable of the word, whereas UK English places the stress on the second syllable of the word. Shifting the stress changes the pronunciation.
Although present generation speech recognition systems (e.g. Nuance) do not make use of stress (see absence of the same from CPA, above) the stress information is essential for a voice talent performing a script and may potentially be useful in enhanced speech recognition.
E. Conclusion
Prior techniques for converting US English to UK English have required humans to perform textual normalization and pronunciation transformations
Bever Hoffman & Harms LLP
Harms Jeanette S.
Spooner Lamont M
Tellme Networks, Inc.
{haeck over (S)}mits T{overscore (a)}livaldis Ivars
LandOfFree
Automated transformation from American English to British... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Automated transformation from American English to British..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Automated transformation from American English to British... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3268476