Ex Parte POI

Patent Trial and Appeal Board

Jan 30, 2015

11694316 (P.T.A.B. Jan. 30, 2015)

UNITED STATES PATENT AND TRADEMARK OFFICE
____________

BEFORE THE PATENT TRIAL AND APPEAL BOARD
____________

Ex parte PARIND POI
____________

Appeal 2012-008801
Application 11/694,316
Technology Center 2600
____________

Before JOHN A. JEFFERY, DAVID M. KOHUT, and
ANDREW J. DILLON, Administrative Patent Judges.

JEFFERY, Administrative Patent Judge.

DECISION ON APPEAL
Appellant appeals under 35 U.S.C. § 134(a) from the Examiner’s
decision to reject claims 1, 2, 4–6, and 21–33. We have jurisdiction under
35 U.S.C. § 6(b). We affirm.

STATEMENT OF THE CASE
Appellant’s invention involves speech recognition training based on
confirmed speaker utterances. In one aspect, the system uses a grammar
database populated with possible combinations of phrases and a confidence
database derived from the grammar database to recognize user utterances
accurately. See generally Abstract; Spec. 7–8; Figs. 1, 3. Claim 1 is
illustrative:
Appeal 2012-008801
Application 11/694,316

2
1. A method comprising:
receiving a signal representing a voice utterance;
processing, by a processor, the signal to determine a textual phrase
from the utterance;
determining that the textual phrase corresponds to one of a plurality of
phrases stored in at least one of a grammar database and a confidence
database, the confidence database being derived from the grammar database;
requesting confirmation from a user who originated the utterance in
response to the determination;
assigning a score to the utterance depending on whether confirmation
is received from the user; and
modifying an acoustic model for speech recognition processing using
the assigned score.

THE REJECTION
The Examiner rejected claims 1, 2, 4–6, and 21–33 under 35 U.S.C.
§ 102(e)1 as anticipated by Marcus (US 2006/0085184 A1; Apr. 20, 2006).
Ans. 4–9.2

CONTENTIONS
The Examiner finds that Marcus discloses every recited element of
claim 1 including determining that a textual phrase determined from a voice
utterance corresponds to one of plural phrases stored in at least one of (1) a
grammar database, and (2) a confidence database derived from the grammar
database. Ans. 4–6, 10–16. Among other things, the Examiner refers to the

1 Because Marcus was published before the present application’s filing date,
Marcus also qualifies as prior art under § 102(a). Nevertheless, we deem
any error associated with this omission harmless as it does not affect our
assessment of the merits of the anticipation rejection.
2 Throughout this opinion, we refer to (1) the Appeal Brief filed February
22, 2012 (“App. Br.”); (2) the Examiner’s Answer mailed March 15, 2012
(“Ans.”); and (3) the Reply Brief filed May 15, 2012 (“Reply Br.”).
Appeal 2012-008801
Application 11/694,316

3
functionality of Marcus’ speech recognition mechanism whose confidence
scores are said to be derived from an associated grammar database used to
recognize speech. Ans. 5, 12–13, 15–16.
Appellant argues that Marcus does not disclose a grammar database,
let alone a confidence database derived from a grammar database as
claimed, nor does Marcus’ speech recognition necessarily rely on such
databases. App. Br. 6–9; Reply Br. 2–5. Appellant adds that Marcus also
does not determine that the textual phrase corresponds to one of plural
phrases stored in at least one of the grammar and confidence databases as
claimed. App. Br. 9–11; Reply Br. 5–7.

ISSUE
Under § 102, has the Examiner erred in rejecting claim 1 by finding
that Marcus determines that a textual phrase determined from a voice
utterance corresponds to one of plural phrases stored in at least one of (1) a
grammar database, and (2) a confidence database derived from the grammar
database?

ANALYSIS
As noted above, claim 1 recites determining that the textual phrase
corresponds to one of plural phrases stored in at least one of (1) a grammar
database, and (2) a confidence database derived from the grammar database.
We emphasize “at least one of” here, for only one of the two alternatives
need be met by the prior art to anticipate the limitation. That is, to satisfy
alternative (1) above, the prior art need only disclose the recited
determination involving only a grammar database storing plural phrases: a
Appeal 2012-008801
Application 11/694,316

4
confidence database is not required, let alone that it is derived from the
grammar database. But to satisfy alternative (2), both databases are
required.
Given this breadth, we see no error in the Examiner’s reliance on
Marcus’ speech recognition mechanism 210 for anticipating the recited
determination step. As shown in Marcus’ Figure 2(a), speech recognition
mechanism 210 (1) receives an input waveform 207 generated from the
user’s speech, and (2) produces a recognition result as a word or sequence of
words and associated confidence score—a sequence that can be used to
recognize a phrase. Marcus ¶¶ 25–26 (noting that the speech recognition
mechanism may be capable of recognizing a phrase from a sequence of
words.) Marcus further notes that in some applications, after the user says a
certain phrase (e.g., “local phone service” or “long distance service”), the
speech recognition mechanism outputs recognized phrases and an associated
confidence score. Marcus ¶ 27.
Leaving aside Marcus’ confidence scores for the moment, the very
recognition of the uttered phrase by the speech recognition mechanism fully
meets the recited first determination alternative, namely determining that the
textual phrase3 corresponds to one of plural phrases stored in a grammar
database. According to Appellants’ Specification, a grammar database, in
general, stores all possible combinations of user utterances and associated
values that are validly accepted by a particular speech application. Spec.
10:14–15. Appellants also describe various forms of grammar databases on

3 The Examiner’s finding that Marcus teaches determining a textual phrase
from a voice utterance in paragraphs 25 and 26 and Figures 2a and 2b (Ans.
5) is undisputed.
Appeal 2012-008801
Application 11/694,316

5
pages 7 and 8. But apart from these descriptions, Appellants’ Specification
does not define the term “grammar database” to so limit its interpretation.4
Accordingly, we construe the term with its plain meaning. The term
“database” is defined quite broadly as “[a]ny electronically-stored collection
of data.” Alan Freedman, The Computer Glossary 86 (9th ed. 2001).
Accord In re Comiskey, 554 F.3d 967, 981 (Fed. Cir. 2009) (citing this
definition from an earlier edition). Therefore, a “grammar database” is any
electronically-stored collection of data pertaining to grammar.
Given this construction, there must be some sort of electronically-
stored collection of data or “grammar database” in Marcus for the speech
recognition mechanism to correlate an utterance-based textual phrase5 with a
stored phrase to recognize the phrase and provide an associated recognition
result. See Marcus ¶¶ 25–27. That this recognition can involve multiple
phrases (see id.) only bolsters the Examiner’s position that plural phrases are
stored in this database, and that this word-based and phrase-based data at
least pertains to grammar. Accord Ans. 13 (finding that Marcus’ speech
recognition incorporates grammar.)
This functionality, then, anticipates at least the first recited alternative
of the determination step in claim 1, as well as the commensurate limitation
in independent claim 21. To the extent that Appellants’ arguments are based
on these claims requiring both databases (see App. Br. 6–11; Reply Br. 5–7),
such arguments are not commensurate with the scope of the claims.

4 The Specification, however, defines at least one other term. See, e.g.,
Spec. 15 (defining “computer-readable medium”).
5 As noted above, the Examiner’s finding that Marcus teaches determining a
textual phrase from a voice utterance in paragraphs 25 and 26 and Figures 2a
and 2b (Ans. 5) is undisputed.
Appeal 2012-008801
Application 11/694,316

6
Nevertheless, we see no error in the Examiner’s rejection even if these
claims required both databases (which they do not)—a dual-database feature
required by independent claim 26 that positively recites both databases.6
Notably, Marcus’ speech recognition mechanism can assign
confidence scores not only on a word-by-word basis, but also on a phrase-
by-phrase basis in connection with the recognition result. See Marcus
¶¶ 25–27. To achieve this end, there must be some sort of electronically-
stored collection of data pertaining to confidence or “confidence database.”
And because this confidence data is tied to the corresponding words and
phrases in the “grammar database” as the Examiner indicates (Ans. 13), this
stored confidence data collection or “confidence database” is derived from
the “grammar database” at least in that sense. Appellants’ arguments to the
contrary (App. Br. 6–9; Reply Br. 2–5) are unavailing and not
commensurate with the scope of the claim.
Therefore, we are not persuaded that the Examiner erred in rejecting
claim 1, and claims 2, 4–6, and 21–33 not argued separately with
particularity.

CONCLUSION
The Examiner did not err in rejecting claims 1, 2, 4–6, and 21–33
under § 102.

6 Although claim 26 recites both grammar and confidence databases, the
claim lacks the textual phrase correspondence determination in independent
claims 1 and 21.
Appeal 2012-008801
Application 11/694,316

7
DECISION
The Examiner’s decision rejecting claims 1, 2, 4–6, and 21–33 is
affirmed.
No time period for taking any subsequent action in connection with
this appeal may be extended under 37 C.F.R. § 1.136(a)(1)(iv).

AFFIRMED

cdc