Ex Parte Lloyd et al

Patent Trial and Appeal Board

Sep 27, 2017

13862219 (P.T.A.B. Sep. 27, 2017)

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.
13/862,219 04/12/2013 Matthew I. Lloyd 16113-2135003 4656
26192 7590 09/29/2017
FTSH fr RTfTTARnSON P C EXAMINER
PO BOX 1022 SAINT CYR, LEONARD
MINNEAPOLIS, MN 55440-1022
ART UNIT PAPER NUMBER
2658
NOTIFICATION DATE DELIVERY MODE
09/29/2017 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):
PATDOCTC@fr.com
PTOL-90A (Rev. 04/07)
UNITED STATES PATENT AND TRADEMARK OFFICE
BEFORE THE PATENT TRIAL AND APPEAL BOARD
Ex parte MATTHEW I. LLOYD and TRAUSTI T. KRISTJANSSON
Appeal 2016-004628
Application 13/862,219
Technology Center 2600
Before ROBERT E. NAPPI, JUSTIN BUSCH, and JASON M. REPKO,
Administrative Patent Judges.
REPKO, Administrative Patent Judge.
DECISION ON APPEAL
STATEMENT OF THE CASE
Appellants1 appeal under 35 U.S.C. § 134(a) from the Examiner’s
rejection of claims 1, 4, 6, 7, 9—17, 19, and 20. App. Br. 3.2 Claims 2 and 3
have been canceled. Claims 5, 8, and 18 have not been rejected. We have
jurisdiction under 35 U.S.C. § 6(b). We affirm.
1 Appellants identify the real party in interest as Google Inc. App. Br. 1.
2 Throughout this opinion, we refer to (1) the Final Rejection (“Final Act.”)
mailed March 26, 2015, (2) the Appeal Brief (“App. Br.”) filed September
14, 2015, (3) the Examiner’s Answer (“Ans.”) mailed February 2, 2016, and
(4) the Reply Brief (“Reply Br.”) filed March 30, 2016.
Appeal 2016-004628
Application 13/862,219
THE INVENTION
Appellants’ invention improves speech-recognition accuracy.
Abstract. A user’s regional accent can affect the accuracy of speech
recognition. Spec. 13. To improve accuracy, one embodiment applies
acoustic models to audio signals that are geotagged with location
information. Id. 14. The acoustic model is specific to a single geographic
area or to multiple geographic areas. Id. The acoustic model may be
automatically trained, selected, generated, or otherwise adapted using the
geotagged audio signal. Id.
Claim 1 is reproduced below:
1. A system comprising:
one or more computers; and
one or more computer-readable media coupled to the one
or more computers having instructions stored thereon which,
when executed by the one or more computers, cause the one or
more computers to perform operations comprising:
receiving an audio signal that corresponds to an
utterance spoken by a user,
determining a geographic location associated with
the user,
obtaining audio signals that correspond to
utterances from a group of users that were also associated
with the geographic location,
after receiving the audio signals that corresponds to
the utterance spoken by the user, adapting one or more
acoustic models for the geographic location using the
audio signals that correspond to utterances from the group
of users that were also associated with the geographic
location, and
performing speech recognition on the audio signal
using the one or more acoustic models that are adapted for
the geographic location using the audio signals that
correspond to utterances of the group of users that were
also associated with the geographic location.
2
Appeal 2016-004628
Application 13/862,219
THE REJECTIONS
The Examiner relies on the following as evidence:
Deisher et al.
Veprek et al.
Kadambe et al
Mamou et al.
US 2003/0236099 Al
US 2004/0158457 Al
US 2004/0230420 Al
US 2009/0030894 Al
Dec. 25, 2003
Aug. 12, 2004
Nov. 18, 2004
Jan. 29, 2009
Claims 1, 4, 6, 7, 9—12, 14, 15, 17, 19, and 20 stand rejected under
35 U.S.C. § 103(a) as unpatentable over Deisher and Veprek. Ans. 2—7.
Claim 13 stands rejected under 35 U.S.C. § 103(a) as unpatentable
over Deisher, Veprek, and Mamou. Ans. 7—8.
Claim 16 stands rejected under 35 U.S.C. § 103(a) as unpatentable
over Deisher, Veprek, and Kadambe. Ans. 8—9.
THE REJECTION OVER DEISHER AND VEPREK
Claims 1, 19, and 20
The Examiner finds that Deisher teaches every limitation recited in
independent claims 1,19, and 20 except for (1) obtaining the recited audio
signals, (2) adapting the acoustic models using the signals, and (3)
performing speech recognition using the adapted models. Final Act. 2—3. In
concluding that claims 1,19, and 20 would have been obvious, the Examiner
cites Veprek as teaching these features. Id. at 3^4.
Appellants argue that Veprek does not teach adapting the acoustic
model after receiving the audio signal corresponding to the user’s utterance.
App. Br. 3^4; Reply Br. 1—2. According to Appellants, Veprek trains
models for “new users.” App. Br. 4 (citing Veprek 135). And Appellants
contend that Veprek does not train the model after receiving the user’s audio
3
Appeal 2016-004628
Application 13/862,219
signal because Veprek’s new users have not yet provided any speech data.
App. Br. 4 (citing Veprek 135). We are unpersuaded.
The limitation at issue in claim 1 recites, in part, “adapting one or
more acoustic models for the geographic location using the audio signals
that correspond to utterances from the group of users that were also
associated with the geographic location” (emphasis added). Claims 19 and
20 recite similar limitations. For example, adapting the acoustic models
“may include selecting the one or more acoustic models generated for the
geographic location associated with the mobile device.” Spec. 172
(emphasis added). Or adapting “may include incorporating data that
references the geographic location into state information included in the
acoustic model.” Id. (emphasis added). But the claim broadly recites
“using” the recited audio signals, which does not limit the claim to either
embodiment.
Given this breadth, the recited adapting step encompasses, among
other techniques, Veprek’s teaching of adapting “using” an acoustic model
that was initially trained on utterances from a group of users (Tflf 20, 35), as
cited by the Examiner. See Ans. 4, 11.
In particular, Veprek’s pre-loaded acoustic models do not reflect the
user’s voice characteristics. Veprek 120. Veprek, however, transforms
these acoustic models “[o]nce the user has been speaking to the system for
some time.” Id. For instance, to improve accuracy, Veprek stores the user’s
speech and uses the stored speech to transform the model. Id. This
transformed model better reflects the user’s voice characteristics. Id. That
is, like the recited adapting, Veprek adapts the model after receiving audio
signals corresponding to the user’s utterances. See id.
4
Appeal 2016-004628
Application 13/862,219
Appellants argue that Veprek’s adapting in this example only exploits
information in the user’s stored speech, not the group’s. Reply Br. 1—2.
Veprek, however, teaches using a group’s speech to initially train the new-
user model—i.e., the model that is later adapted. See Veprek || 20, 35, cited
in Ans. 4, 11. In one embodiment, Veprek teaches training the new-user
acoustic models using data from many different users. Veprek 135. For
example, the acoustic model for a new male customer from Georgia is
trained on speech from males in Georgia. Id.
Because the acoustic model used in Veprek’s adaptation (| 20) was
initially trained on utterances from the group of users (135), we agree that
Veprek teaches adapting the model “using” the recited audio signals
corresponding to the utterances from the group of users. See Ans. 4 (citing
Veprek || 20, 35). On this record, Appellants’ arguments (App. Br. 4;
Reply Br. 1—2) are unpersuasive because the arguments do not take into
account the collective teachings of paragraphs 20 and 35, which are relied
upon by the Examiner in the obviousness rejection (Ans. 4, 11).
Accordingly, we sustain the Examiner’s rejection of independent
claims 1,19, and 20, as well as dependent claims 7, 9—12, 14, 15, and 17,
which are not argued separately with particularity (see App. Br. 3—4).
Claim 4
Claim 4 recites, in part, “receiving geotagged audio signals that
correspond to audio recorded by multiple mobile devices in multiple
geographic locations.”
5
Appeal 2016-004628
Application 13/862,219
Regarding this limitation, the Examiner finds that Veprek collects
speech from users from different demographic origins, such as Texas and
Georgia using their devices. Ans. 12—13 (citing Veprek || 10, 20, 34, 35).
Appellants argue that Deisher and Veprek do not receive geotagged
audio signals that correspond to audio recorded by multiple mobile devices
in multiple geographic locations. App. Br. 5. Appellants contend that,
although Deisher uses the identity of nearby people, Deisher does not
receive audio signals from them. Id. Appellants further contend that the
cited portion of Veprek does not teach different end users or different
devices are in multiple geographic locations. Id. (citing Veprek 137).
We note that the Examiner does not rely on Deisher for the limitation
at issue. See Ans. 12—13. Furthermore, we disagree that Veprek does not
teach or suggest audio recorded by multiple mobile devices in multiple
geographic locations.
In particular, Veprek pools data from different males in Georgia.
Veprek 135, cited in Ans. 5. Veprek explains that specialized acoustic
models can be trained “for sub-groups of new users.” Veprek 135
(emphasis added). That is, Veprek suggests that Georgian males are a sub­
group of new users, which can be from other geographic locations. See id.
For example, Veprek discloses that some users are from Texas. Id. 134,
cited in Ans. 13. Furthermore, Veprek collects data from the user’s cellular
phones and other devices. Veprek 110, cited in Ans. 13. Therefore, we are
unpersuaded that the Examiner erred in this regard. See App. Br. 5.
Claim 4 further recites, in part, “adapting one or more acoustic models
for the geographic location further comprises adapting one or more acoustic
6
Appeal 2016-004628
Application 13/862,219
models for the geographic location using a subset of the geotagged audio
signals.”
Appellants argue that Veprek does not adapt the acoustic model using
a subset of the speech data. App. Br. 5—6. We disagree. As discussed in
connection with the independent claims, the Examiner has shown that
Veprek uses the speech data from males from Georgia to adapt the model.
See Veprek 135. Veprek further teaches that males from Georgia are a
“sub-group” (id.) and suggests other users are from Texas (id. 134).
Because the speech data are at least from the Georgia sub-group, the
Examiner reasonably concludes that the data corresponds to the subset
recited in claim 4. Ans. 13 (discussing collecting data from a specific group
and citing Veprek 135).
Accordingly, we sustain the Examiner’s rejection of claim 4.
Claim 6
Claim 6 depends from claim 4. Appellants argue that Deisher and
Veprek do not teach the selecting limitation recited in claim 6 because the
cited portions of Deisher and Veprek do not receive geotagged audio signals
from multiple mobile devices in multiple geographic locations, which is
recited in claim 4. App. Br. 6. For the reasons discussed in connection with
claim 4, we find this argument (id.) unpersuasive. Accordingly, we also
sustain the Examiner’s rejection of claim 6.
THE REJECTION OVER DEISHER, VEPREK, AND MAMOU
Claim 13 recites, in part, “executing a search query using the one or
more candidate transcriptions.”
7
Appeal 2016-004628
Application 13/862,219
Appellants argue that the Examiner erred by finding that Deisher and
Veprek disclose the features of claim 13. App. Br. 7. The Examiner,
however, finds neither Deisher nor Veprek teaches the features of claim 13,
and turns to Mamou in concluding that the subject matter of claim 13 would
have been obvious. Final Act. 8—9; see also Ans. 15. Because Appellants’
argument (App. Br. 13) does not squarely address, let alone persuasively
rebut, the Examiner’s reliance on Mamou as teaching the limitation at issue,
we sustain the rejection of claim 13.
THE REJECTION OVER DEISHER, VEPREK, AND KADAMBE
Claim 16 depends from claim 1. Appellants do not separately argue
claim 16. See App. Br. 3—8; Reply Br. 1—2. Therefore, for the reasons
discussed in connection with claim 1, we also sustain the rejections of claim
16.
DECISION
We affirm the Examiner’s rejection of claims 1, 4, 6, 7, 9—17, 19, and
20.
No time period for taking any subsequent action in connection with
this appeal may be extended under 37 C.F.R. § 1.136(a)(l)(iv).
AFFIRMED
8