Google Inc.

Patent Trials and Appeals Board

Jul 1, 2020

14982567 - (D) (P.T.A.B. Jul. 1, 2020)

UNITED STATES PATENT AND TRADEMARK OFFICE
UNITED STATES DEPARTMENT OF COMMERCE
United States Patent and Trademark Office
Address: COMMISSIONER FOR PATENTS
P.O. Box 1450
Alexandria, Virginia 22313-1450
www.uspto.gov
APPLICATION NO. FILING DATE FIRST NAMED INVENTOR ATTORNEY DOCKET NO. CONFIRMATION NO.
14/982,567 12/29/2015 Petar Aleksic 231441-461050 4284
101874 7590 07/01/2020
HONIGMAN LLP-GOOGLE
DOCKETING DEPARTMENT
650 Trade Centre Way
Suite 200
KALAMAZOO, MI 49002-0402
EXAMINER
WOZNIAK, JAMES S
ART UNIT PAPER NUMBER
2657
NOTIFICATION DATE DELIVERY MODE
07/01/2020 ELECTRONIC
Please find below and/or attached an Office communication concerning this application or proceeding.
The time period for reply, if any, is set in the attached communication.
Notice of the Office communication was sent electronically on above-indicated "Notification Date" to the
following e-mail address(es):
google@honigman.com
PTOL-90A (Rev. 04/07)
UNITED STATES PATENT AND TRADEMARK OFFICE
BEFORE THE PATENT TRIAL AND APPEAL BOARD
Ex parte PETAR ALEKSIC and PEDRO MORENO MENGIBAR
Appeal 2019-001884
Application 14/982,567
Technology Center 2600
Before ALLEN R. MacDONALD, JEREMY J. CURCURI, and
PHILLIP A. BENNETT, Administrative Patent Judges.
CURCURI, Administrative Patent Judge.
DECISION ON APPEAL
STATEMENT OF THE CASE
Pursuant to 35 U.S.C. § 134(a), Appellant1 appeals from the
Examiner’s decision to reject claims 1, 3, 4, 8, 9, 11, 13, 15, 16, 21–23, and
25–32. We have jurisdiction under 35 U.S.C. § 6(b).
We REVERSE.

1 We use the word Appellant to refer to “applicant” as defined in 37 C.F.R.
§ 1.42(a). Appellant identifies the real party in interest as Google LLC.
Appeal Br. 1.
Appeal 2019-001884
Application 14/982,567

2
CLAIMED SUBJECT MATTER
The claims are directed to “speech recognition and the use of class-
based language models to transcribe speech to text.” Spec. ¶ 1. Claim 1,
reproduced below, is illustrative of the claimed subject matter:
1. A computer-implemented method, comprising:
receiving, at a computing system, speech data that
characterizes an utterance of a user;
generating a candidate transcription of the utterance using
a static class-based language model that includes a plurality of
static classes, each static class populated with a pre-defined set
of class-based terms that were selected for inclusion in the static
class independently of a context of the utterance or the user;
determining whether the candidate transcription includes
a class-based term from at least one static class that is capable of
being customized based on the context of the utterance or the
user;
in response to determining that the candidate transcription
includes a class-based term from at least one static class that is
capable of being customized based on the context of the utterance
or the user:
(i) generating a dynamic class-based language
model that includes at least one dynamic class that is
customized based on the context of the utterance or the
user, including generating a first dynamic class by adding
dynamic class-based terms that were selected based on the
context of the utterance or the user to the pre-defined set
of class-based terms for a first static class of the plurality
of static classes, and
(ii) providing, using the dynamic class-based
language model, a speech recognition result for the
utterance.
Appeal 2019-001884
Application 14/982,567

3
REFERENCES
The prior art relied upon by the Examiner is:
Name Reference Date
US 2005/0055210 A1 Venkataraman Mar. 10, 2005
US 2011/0055256 A1 Phillips Mar. 3, 2011
US 2015/0058018 A1 Georges Feb. 26, 2015

REJECTIONS
Claims 1, 3, 4, 8, 9, 11, 13, 15, 16, 21–23, 25–30, and 32 are rejected
under 35 U.S.C. § 103 as obvious over Phillips and Venkataraman. Final
Act. 3–11.
Claim 31 is rejected under 35 U.S.C. § 103 as obvious over Phillips,
Venkataraman, and Georges. Final Act. 11–12.
OPINION
The Obviousness Rejection of Claims 1, 3, 4, 8, 9, 11, 13, 15, 16, 21–23, 25–
30, and 32 over Phillips and Venkataraman
The Examiner finds Phillips and Venkataraman teach all limitations of
claim 1. Final Act. 3–5; see also Ans. 11–22. In particular, the Examiner
finds Philips teaches most limitations of claim 1. Final Act. 3–4. In
particular, the Examiner further finds
Although Phillips selects a dynamic language model
populated with terms based upon the context of utterance
hypothesis terms and also considers a user/client context, Phillips
does not specifically recite that at least one class in the initial
language model has the capability of being customized based on
the context of the utterance of the user and then actually adding
dynamic class-based terms that are selected based on the context
of the utterance or the user to the predefined set of class-based
terms for a first static class of the plurality of static classes
Appeal 2019-001884
Application 14/982,567

4
because Phillips merely replaces the first, initial language model
with dynamic, secondary language model containing terms
indicated by words/terms in the utterance transcription generated
in an initial speech recognition pass. Thus, Phillips explicitly
lacks the customizable/expandable initial language models
indicated in claim 1. Venkataraman, however, discloses a
refining/updating of an original, utterance-independent,
customizable domain-based static language model by adding
terms/words based upon domain-based terms that are identified
in a transcription generated by an initial speech recognition pass
followed by the utilization of the updated language model in a
second speech recognition pass.
Final Act. 4–5 (citing Venkataraman ¶¶ 25–26, 29–34, 36, 39–40).
The Examiner reasons
it would have been obvious to one of ordinary skill in the art to
modify the teachings of Phillips with the dynamic language
model generation technique that adds dynamic domain-based
terms to an existing static domain-based language model taught
by Venkataraman in order to provide a predictable result in the
form of increasing the accuracy of speech recognition results.
Final Act. 5 (citing Venkataraman ¶ 46).
Among other arguments, Appellant presents the following principal
arguments:
i. Phillips and Venkataraman do not teach “generating a dynamic
class-based language model that includes at least one dynamic class that is
customized based on the context of the utterance or the user,” as recited in
claim 1. See Appeal Br. 7–9. “Phillips never describes a ‘class-based’
language model—let alone a ‘dynamic’ class-based language model having
a ‘dynamic class’ as recited in claim 1.” Appeal Br. 8 (citing Phillips ¶¶ 71,
88); see also Reply Br. 1–2. Venkataraman mentions class-based language
models, but does not teach the language recited in claim 1. See Appeal Br.
Appeal 2019-001884
Application 14/982,567

5
8–9 (citing Venkataraman ¶ 26); see also Reply Br. 2–3 (“Venkataraman
generates the second language model through analysis of the content of the
utterance itself (as reflected by the ‘high-confidence words’ from the initial
recognition hypotheses) rather than based on a context of the utterance”).
ii. Phillips and Venkataraman do not teach “generating a first
dynamic class by adding dynamic class-based terms that were selected based
on the context of the utterance or the user to the pre-defined set of class-
based terms for a first static class of the plurality of static classes,” as recited
in claim 1. See Appeal Br. 9–11. “Venkataraman teaches essentially the
opposite of what is claimed—instead of adding terms to a class as recited in
claim 1, Venkataraman ‘narrows’ a search space so that fewer terms are
included.” Appeal Br. 9 (citing Venkataraman ¶ 32); see also Appeal Br.
10–11, Reply Br. 4–6.
Appellant’s arguments persuade us that the Examiner erred.
Phillips discloses determining an initial set of language models,
running recognition, determining a new set of language models, and running
recognition again. See Phillips ¶ 88; see also Phillips Fig. 5a. Phillips
discloses “n-gram statistical language models,” but does not explicitly state
that the models are class-based language models. Phillips ¶ 71.
Venkataraman discloses “a class-based language model may be used.”
Venkataraman ¶ 26. However, even if we were to conclude a skilled artisan
would have modified Phillips to use class-based language models, resulting
in “generating a dynamic class-based language model” (Phillips’s new set of
language models), as recited in claim 1, we find Phillips and Venkataraman
do not teach the specific class-based model recited in the claim. That is, the
cited references do not teach the further recitation of
Appeal 2019-001884
Application 14/982,567

6
that includes at least one dynamic class that is customized based
on the context of the utterance or the user, including generating
a first dynamic class by adding dynamic class-based terms that
were selected based on the context of the utterance or the user to
the pre-defined set of class-based terms for a first static class of
the plurality of static classes,
as recited in claim 1.
According to Venkataraman, an initial language model is applied to a
speech signal to determine high-confidence words, and a data set query is
made with the high-confidence words. See Venkataraman ¶¶ 27–29.
Venkataraman discloses “the second language model is constructed by
updating the initial language model based on the query results.”
Venkataraman ¶ 30.
In describing the updating of the initial language model,
Venkataraman discloses “analyz[ing] the query results to find novel words”
that are not contained in the initial language model, and then “add[ing] the
novel words . . . to the method 400’s pronunciation dictionary, thereby
enabling the novel words to be identified the next time they are spoken by
the user.” Venkataraman ¶ 31.
In our view, adding words to the “pronunciation dictionary” does not
teach “adding dynamic class-based terms . . . to the pre-defined set of class-
based terms for a first static class of the plurality of static classes,” as recited
in claim 1.
In reaching our decision, we recognize that Venkataraman discloses
“a class-based language model may be used.” Venkataraman ¶ 26. However,
it is unclear to us how or why the further disclosures in Venkataraman of
adding words to a pronunciation dictionary (Venkataraman ¶ 31) would
have been applied to a class-based language model, and in particular, how or
Appeal 2019-001884
Application 14/982,567

7
why adding words to a pronunciation dictionary suggests adding terms (or
words) to a static class of a plurality of static classes to generate a dynamic
class, as recited in claim 1.
Thus, on the record before us, we determine the Examiner erred in
finding Phillips and Venkataraman teach
(i) generating a dynamic class-based language model that
includes at least one dynamic class that is customized based on
the context of the utterance or the user, including generating a
first dynamic class by adding dynamic class-based terms that
were selected based on the context of the utterance or the user to
the pre-defined set of class-based terms for a first static class of
the plurality of static classes,
as recited in claim 1.
We, therefore, do not sustain the Examiner’s rejection of claim 1.
We also do not sustain the Examiner’s rejection of claims 3, 4, 8, 9,
11, 22, 28–30, and 32, which depend from claim 1.
Independent claim 13 recites similar language as claim 1. We,
therefore, also do not sustain the Examiner’s rejection of claim 13.
We also do not sustain the Examiner’s rejection of claims 15, 16, and
23, which depend from claim 13.
Independent claim 21 recites similar language as claim 1. We,
therefore, also do not sustain the Examiner’s rejection of claim 21.
We also do not sustain the Examiner’s rejection of claims 25–27,
which depend from claim 13.

Appeal 2019-001884
Application 14/982,567

8
The Obviousness Rejection of Claim 31 over Phillips, Venkataraman, and
Georges
Claim 31 depends from claim 1. The Examiner does not find Georges
cures the deficiency of Phillips and Venkataraman. See Final Act. 11–12.
We, therefore, do not sustain the Examiner’s rejection of claim 31.
CONCLUSION
The Examiner’s decision to reject claims 1, 3, 4, 8, 9, 11, 13, 15, 16,
21–23, and 25–32 is reversed.
DECISION SUMMARY
In summary:
Claims
Rejected
35 U.S.C.
§
References Affirmed Reversed
1, 3, 4, 8, 9,
11, 13, 15,
16, 21–23,
25–30, 32
103 Phillips,
Venkataraman
1, 3, 4, 8, 9,
11, 13, 15,
16, 21–23,
25–30, 32
31 103 Phillips,
Venkataraman,
Georges
31
Overall
Outcome
1, 3, 4, 8, 9,
11, 13, 15,
16, 21–23,
25–32
REVERSED