Ex Parte Gruenstein et al

Patent Trial and Appeal Board

Dec 12, 2018

13838379 (P.T.A.B. Dec. 12, 2018)

UNITED STA TES p A TENT AND TRADEMARK OFFICE
APPLICATION NO. FILING DATE FIRST NAMED INVENTOR
13/838,379 03/15/2013 Alexander H. Gruenstein
26192 7590 12/14/2018
FISH & RICHARDSON P.C.
PO BOX 1022
MINNEAPOLIS, MN 55440-1022
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
ATTORNEY DOCKET NO. CONFIRMATION NO.
16113-4437001 7937
EXAMINER
OPSASNICK, MICHAEL N
ART UNIT PAPER NUMBER
2658
NOTIFICATION DATE DELIVERY MODE
12/14/2018 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):
P ATDOCTC@fr.com
PTOL-90A (Rev. 04/07)
UNITED STATES PATENT AND TRADEMARK OFFICE
BEFORE THE PATENT TRIAL AND APPEAL BOARD
Ex parte ALEXANDER H. GRUENSTEIN and
PETARALEKSIC
Appeal2017-002237
Application 13/838,379
Technology Center 2600
Before JEREMY J. CURCURI, JUSTIN BUSCH, and
PHILLIP A. BENNETT, Administrative Patent Judges.
BUSCH, Administrative Patent Judge.
DECISION ON APPEAL
Pursuant to 35 U.S.C. § 134(a), Appellants appeal from the
Examiner's decision to reject claims 1, 3-8, 10-15, and 17-21. Claim 22 is
objected to as being dependent on a rejected base claim. Final Act. 1, 8.
Oral arguments were heard on November 14, 2018. A transcript of the
hearing was placed in the record. We have jurisdiction over the pending
claims under 35 U.S.C. § 6(b). Claims 2, 9, and 16 were cancelled
previously.
We reverse.
CLAIMED SUBJECT MATTER
Appellants' invention generally relates to speech recognition and,
more specifically, to mixed model speech recognition. Spec. ,r,r 2, 18, Title.
Appeal2017-002237
Application 13/838,379
Even more specifically, aspects of Appellants' invention relate to using both
a client-side limited language model speech recognizer developed using
user-specific data and a server-side speech recognizer model with a larger
vocabulary developed independent of user-specific data. Spec. ,r 18. Claims
1, 8, and 15 are the independent claims. Claim 1 is illustrative and
reproduced below, with the disputed limitations italicized:
1. A computer-implemented method comprising:
accessing audio data generated by a computing device
based on audio input from a user, the audio data encoding one or
more user utterances;
generating a first transcription of the utterances by
performing speech recognition on the audio data using a first
speech recognizer, wherein the first speech recognizer employs
a language model that is based on user-specific data;
generating a second transcription of the utterances by
performing speech recognition on the audio data using a second
speech recognizer, wherein the second speech recognizer
employs a language model independent of user-specific data;
determining that the second transcription of the utterances
includes a term from a predefined set of one or more terms
associated with actions that are performable by the computing
device; and
based on determining that the second transcription of the
utterance includes the term from the predefined set of one or
more terms, providing an output of the first transcription of the
utterance.
REJECTION
Claims 1, 7, 8, 14, 15, and 20 stand provisionally rejected on the
ground of obviousness-type double-patenting over claims 1, 3-9, 11-16, and
18-20 of U.S. Application No. 13/892,590. Final Act. 3.
2
Appeal2017-002237
Application 13/838,379
Claims 1, 3-8, 10-15, and 17-20 stand rejected under 35 U.S.C. § 102
as anticipated by Phillips (US 2011/0066634 Al; Mar. 17, 2011 ). Final
Act. 4---6.
Claim 21 stands rejected under 35 U.S.C. § 103 as obvious in view of
Phillips and Ma (US 2011/0144996 Al; June 16, 2011). Final Act. 7.
ANALYSIS
THE PROVISIONAL OBVIOUSNESS-TYPE DOUBLE-PATENTING REJECTION
Although Appellants do not contest the provisional non-statutory
(obviousness-type) double-patenting rejection of claims 1, 7, 8, 14, 15, and
20, see generally Appeal Br. 4--8; Reply Br. 1-3, we decline to reach this
provisional rejection. We note U.S. Application No. 13/892,590 issued as
U.S. Patent No. 9,058,805 on May 27, 2015, prior to the mailing date of the
Office Action from which this Appeal is taken. We also note the Examiner
included claims 2, 9, and 16 in this rejection, but these claims were canceled
previously,
THE 35 U.S.C. § 102 REJECTION
Appellants argue the rejection of claims 1, 3-8, 10-15, and 17-20
under 35 U.S.C. § 102 as a group. See Appeal Br. 4---6; Reply Br. 1-3.
Independent claims 8 and 15 recite commensurate limitations to those
limitations argued with respect to claim 1 and claims 3-7, 10-14, and 17-20
depend from, and incorporate the limitations of, claims 1, 8, and 15,
respectively. Thus, we select independent claim 1 as representative of this
group, and the remaining claims 2-8, 10-15, and 17-20 stand or fall with
claim 1. 37 C.F.R. § 4I.37(c)(l)(iv).
The Examiner finds Phillips discloses every limitation recited in
independent claim 1. Final Act. 4--5. Specifically, the Examiner finds:
3
Appeal2017-002237
Application 13/838,379
( 1) Phillips' language models output based on usage history in one pass of
speech recognition discloses the recited output of the first transcription,
Final Act. 4 ( citing Phillips ,r 60, Fig. 2); Ans. 7 ( citing Phillips ,r 66);
(2) Phillips' acoustic model output and/or Phillips' "recognized/accepted
text" disclose the recited second transcription, compare Final Act. 4--5
(finding "acoustic models to detect the sound" disclose the recited
"generating a second transcription" (citing Phillips ,r,r 10, 69)) and Ans. 3
(same) with Ans. 7 (finding "the ASR [(Automated Speech Recognition)]
server analyzes recognized/ accepted text ( second transcript)" ( citing Phillips
,r 69)); (3) Phillips' "user action" discloses determining the second
transcription includes a predefined term associated with an action, Final
Act. 5 (citing Phillips ,r 47, Fig. 7a); and (4) Phillips' use of "the
combination of both models to generate a recognized result" discloses
providing an output for the first transcription based on determining the
second transcription includes a predefined term. Final Act. 5 ( citing Phillips
i166).
The Examiner finds the ASR server determines whether speech
recognition results meet certain criteria, which "includes, among other items,
client state information, recognized text, accepted text, timing information,
user actions, and the like." Ans. 7 (citing Phillips ,r,r 10, 174, 189--190). If
the results meet the criteria, the ASR server outputs the first transcription
without changing the language models. Ans. 7. If the results do not meet
the criteria, the ASR server determines the language models need to be
updated and the output is different than the first transcription. Ans. 7.
Appellants acknowledge that Phillips discloses using client state
information to select speech recognition models and alter words in speech
4
Appeal2017-002237
Application 13/838,379
recognition results depending on the application for which speech
recognition is performed----e.g., "applying address-specific changes ... in a
location-based application." Appeal Br. 5 ( citing Phillips ,r 66). Appellants
also acknowledge Phillips discloses using the speech recognition output to
determine and perform appropriate actions on a phone. Appeal Br. 5 ( citing
Phillips ,r 103). Appellants argue, however, that nothing the Examiner cites
in Phillips relates to providing an output of one transcription based on a
determination that a different transcription includes a predefined term.
Appeal Br. 5---6.
We agree with Appellants that the Examiner has not demonstrated
Phillips discloses the disputed limitations. In particular, we agree that
Phillips does not disclose providing an output of a first transcription of
utterances, which is generated employing a language model based on user-
specific data, based on a determination that a different second transcription
of the utterances, which is generated employing a language model
independent of user-specific data, includes a predefined term associated with
an action that is performable by a device. As discussed above, the Examiner
finds Phillips' output from using language models based on usage history for
one pass discloses the recited "generating a first transcription ... by
performing speech recognition ... using a first speech recognizer [that]
employs a language model that is based on user-specific data." Final Act. 4
( citing Phillips ,r 60, Fig. 2).
The Examiner's Mapping of the Second
Transcription to Phillips' Acoustic Model Output
Even accepting the Examiner's findings that Phillips' use of language
models for one pass discloses the recited generating a first transcription and
5
Appeal2017-002237
Application 13/838,379
Phillips' use of an acoustic model discloses the recited generating a second
transcription, Appellants argue Phillips does not disclose providing an
output of the transcription using the language model based on a
determination that transcription using the acoustic model includes a
predefined term, as required by claim 1. Reply Br. 1-3.
Appellants acknowledge Phillips discloses the possibility of rerunning
speech recognition on new models when the current models are not
appropriate. Reply Br. 2-3. Appellants argue, however, Phillips does not
disclose outputting the results of the first transcription if it reruns the speech
recognition with a new model. Reply Br. 2-3. Rather, Phillips either
outputs the results of the acoustic model (i.e., the output the Examiner maps
to the second transcription) or Phillips combines the recognition outputs
(transcriptions) of multiple language models. Reply Br. 1-2 ( citing Phillips
,r 70, Fig. 5a). Thus, Appellants argue, even assuming Phillips' speech
recognition using a language model discloses the recited step of generating a
first transcription and Phillips' rerunning speech recognition using an
acoustic model discloses the recited step of generating a second
transcription, Phillips does not disclose the disputed limitations because
Phillips fails to disclose outputting the results of the language model after
rerunning the speech recognition using an acoustic model. Reply Br. 2-3
("Phillips has no disclosure of choosing the recognition output of a first run
... over the recognition output of a second run ... after the recognition is
rerun based on new models.").
We agree with Appellants. We see nothing in Phillips' cited portions
that discloses providing an output of the first speech recognition pass in
response to determining the second transcription from the acoustic model
6
Appeal2017-002237
Application 13/838,379
includes a predefined term. On the contrary, Phillips discloses that, "if the
current recognition models are not appropriate given the characteristics of
the audio data and the client state information," the ASR "may load new or
additional recognition models" and "rerun the recognition based on these
new models." Phillips ,r 69. Phillips, thus, does not disclose providing the
output of the first pass when the ASR server determines the original
recognition models were inappropriate. In fact, Phillips' disclosure suggests
providing the results from rerunning the recognition using the new models
and, presumably, discarding the old results as "not appropriate."
The Examiner's Mapping of the Second
Transcription to Phillips' Recognize/Accepted Text
Again accepting the Examiner's findings that Phillips' use of
language models for one pass discloses the recited generating a first
transcription, Appellants argue the Examiner errs in finding Phillips'
"recognized/accepted text" discloses the recited second transcription. Reply
Br. 1, 3. Specifically, Appellants argue Phillips' "recognized" text involves
optionally applying a process to alter the output words based on client state
information and Phillips' "accepted" text is user-generated. Reply Br. 3
( citing Phillips ,r,r 64, 66, 1 117). Thus, Appellants argue Phillips' recognized
and accepted text fail to disclose the recited step of generating a second
transcription using a second speech recognizer because Phillips'
recognized/accepted text is not the result of a speech recognizer employing a
language model independent of user-specific data, as recited in claim 1.
Reply Br. 3.
1 Appellants erroneously cite paragraph 67, but the language quoted actually
appears in paragraph 66.
7
Appeal2017-002237
Application 13/838,379
We disagree with Appellants characterization of the Examiner's
findings with respect to the Phillips' "recognized/accepted text." Although
the Examiner provides little more than citation to paragraphs in Phillips, the
particular paragraphs cited indicate Appellants' arguments do not address
the particular text the Examiner identifies. See Ans. 7 ("Examiner notes that
recognized/accepted text includes recognized commands (see Phillips, para
0010 for the command format input; para 0174, para 0189-0190, showing
command elements and content elements)").
Nevertheless, regardless of the particular text the Examiner cites as
disclosing the recited second transcript, the Examiner's findings are based
on rerunning speech recognition using new models if Phillips' ASR
determines the original models are inappropriate. Thus, for the same reasons
discussed above, we are persuaded the Examiner erred because Phillips'
cited portions do not disclose providing an output of a first transcription
based on a determination that a second transcription includes a predefined
term.
Moreover, even considering the additionally cited portions Phillips in
the Answer, Phillips does not disclose the subject matter recited in claim 1.
See Ans. 7 (citing Phillips ,r,r 10, 174, 189--190). Paragraphs 10, 174, 189,
and 190 of Phillips do, in fact, disclose generating a first transcription of
audio data ( of captured utterances) using a resident speech recognition
facility, determining that the audio data includes a command, sending at
least a portion of that audio data to a remote speech recognition facility, and
presenting the output of the second speech recognition facility to the user.
See Phillips ,r,r 10, 174, 189--190. Thus, Phillips' resident speech
recognition facility is similar to the recited second speech recognizer and
8
Appeal2017-002237
Application 13/838,379
Phillips' remote speech recognition facility is similar to the recited first
speech recognizer because Phillips provides output of the remote speech
recognition facility based on a determination that the output of the resident
speech recognition facility includes a command (i.e., a "term from the
predefined set of one or more terms, providing an output of the first
transcription of the utterance"). However, to the extent Phillips' resident
speech recognition facility and remote recognition facility disclose the
recited second speech recognizer and first speech recognizer, respectively,
Phillips does not disclose the particular characteristics of the recited speech
recogmzers.
Specifically, Phillips, at most, suggests the remote speech recognition
facility (similar to the recited first speech recognizer) uses a model based on
user-specific data. Phillips discloses "generating speech-to-text results
utilizing the remote speech recognition facility based at least in part on the
speech and on the information related to the mobile communication facility."
Phillips ,r 190 ( emphasis added). Phillips further discloses transmitting the
information related to the mobile communication facility, which "includes
information about a command recognized by the resident speech recognition
facility and information about contacts stored on the mobile communication
facility," from the mobile communication facility to the remote speech
recognition facility. Phillips ,r 190. Thus, although Phillips may suggest to
a person of ordinary skill in the art that the information related to the mobile
communication facility is "user-specific data" and that such information is
used to transcribe the audio data, Phillips does not explicitly disclose that the
remote speech recognition facility uses a language model based on user-
specific data.
9
Appeal2017-002237
Application 13/838,379
Furthermore, claim 1 recites that "the second speech recognizer
employs a language model independent of user-specific data." Even
assuming Phillips discloses or suggests its remote speech recognition facility
uses a language model based on user-specific data, this would lead to a
conclusion that Phillips' resident speech recognition facility (similar to the
recited second speech recognizer) also uses a language model based on user-
specific data. That is, to the extent Phillips' cited portions are understood to
disclose or suggest that Phillips' remote speech recognition facility language
model is based on the information about contacts stored on the mobile
communication facility, Phillips also discloses that the resident speech
recognition facility uses a language model based on user-specific data. See
Phillips ,r,r 173-17 6, 189-190. In particular, Phillips discloses that the
resident speech recognition facility processes the speech "to recognize
command elements and content elements, wherein the content elements
include the contact name for at least one of a text message and an email
message." Phillips ,r 190. Phillips also discloses limiting processing on a
mobile device and improving accuracy by processing on the device only the
part of the speech that is most predictable, and processing the remainder of
the speech remotely. Phillips ,r 173; see Phillips ,r,r 174--175. Notably,
Phillips discloses that, because "name detection is done on the device, it is
easier to manage the creation of user-chosen names without network
interaction." Phillips ,r 175 ( emphasis added).
Thus, although Phillips discloses providing output of one transcription
based on a determination that a different transcription includes a predefined
term, Phillips does not disclose the particular arrangement claimed.
10
Appeal2017-002237
Application 13/838,379
Summary of Analysis of 3 5 US. C. § 102 Rejection
For the reasons discussed above, we are persuaded the Examiner
erred. Accordingly, we do not sustain the Examiner's rejection of
independent claim 1. For similar reasons, we also do not sustain the
Examiner's rejection of independent claims 8 and 15, which recite similar
limitations, or claims 3-7, 10-14, and 17-20, which depend from claims 1,
8, and 15, respectively.
THE 35 U.S.C. § 103 REJECTION
Appellants separately argue the patentability of claim 21 under
35 U.S.C. § 103 as obvious in view of Phillips and Ma. Appeal Br. 6-7.
Claim 21 depends directly from, and incorporates the limitations of,
independent claim 1. The Examiner neither relies on Ma nor provides a
rationale for modifying Phillips to cure the deficiencies identified above
with respect to the disputed limitations of claim 1. Thus, we reverse the
rejection of claim 21 for the same reasons discussed above with respect to
claim 1.
DECISION
The Examiner's decision to reject claims 1, 3-8, 10-15, and 17-21 is
reversed.
REVERSED
11