`
`
`UNITED STATES PATENT AND TRADEMARK OFFICE
`
`____________________
`
`BEFORE THE PATENT TRIAL AND APPEAL BOARD
`
`____________________
`
`
`APPLE INC.
`Petitioner,
`
`v.
`
`ANDREA ELECTRONICS INC.,
`Patent Owner.
`
`Patent No. 6,363,345
`____________________
`
`IPR2017-00626
`__________________________________________________________________
`
`Petitioner’s Responsive Remand Brief
`
`
`
`
`
`
`
`
`IPR2017-00626
`
`
`
`
`
`Table of Contents
`
`Introduction .................................................................................................... 1
`I.
`II. A Skilled Person Had Reason to Combine Hirsch and Martin................. 1
`III. Martin Discloses the Claimed “Future Minimum” Value ......................... 4
`IV. Martin Discloses Periodically Updating the Minimums ............................ 5
`V. Conclusion ...................................................................................................... 5
`
`
`
`
`ii
`
`
`
`IPR2017-00626
`
`
`
`
`
`I.
`
`Introduction
`Andrea’s main argument—that a skilled person would not have considered
`
`Hirsch and Martin together—was already rejected by the Board in its obviousness
`
`finding on claim 25. Andrea is barred from contesting that finding because it did
`
`not appeal it. Andrea is also wrong. The skilled person would have combined the
`
`references. For example, Hirsch teaches it can be combined with known technique
`
`and cites to Martin as one such known approach. Martin also provides unique
`
`benefits not found in Hirsch. Andrea has no persuasive response to these facts.
`
`Andrea also challenges whether Martin teaches the “future minimum” and
`
`“periodically” elements of the claims, but its arguments are based on
`
`mischaracterizations of Martin’s teachings and the ’345 patent’s specification.
`
`Martin teaches those elements, and the Board should find claims 6-9 unpatentable.
`
`II. A Skilled Person Had Reason to Combine Hirsch and Martin
`Apple explained why the skilled person would have had reason to
`
`incorporate Martin’s noise floor algorithm into Hirsch. Pet., 34-38. Hirsch states
`
`it performed well in stationary noise environments but is silent about non-
`
`stationary noise, indicating it did not perform well. Pet., 35-36; Ex. 1003, ¶¶128-
`
`32. Hirsch also states it can be combined with known techniques and cites to
`
`Martin. Ex. 1003, ¶125. Martin teaches its algorithm performed well in non-
`
`stationary environments and is well-suited for identifying noise. Pet., 35-37; Ex.
`
`
`
`1
`
`
`
`IPR2017-00626
`
`
`
`
`
`1006, abstract, 1095-96. These teachings would have motivated a skilled person to
`
`look for ways to incorporate Martin’s algorithm into Hirsch to provide the same
`
`benefit—better performance in non-stationary noise environments. Pet., 35-36;
`
`Ex. 1003, ¶¶126-32. Apple also explained how the skilled person would have
`
`combined them: by using Martin’s noise floor algorithm to set Hirsch’s adaptive
`
`threshold for identifying noise. Pet., 35-36, 41-42. This is a simple substitution:
`
`instead of calculating the adaptive threshold as a multiple of the noise estimate, it
`
`would be calculated as a multiple of the noise floor. Pet., 36-37, 47; Reply, 19-20.
`
`Other parts of Hirsch would remain unchanged. Id.
`
`Andrea argues that Hirsch teaches away from Martin, (Andrea Br., 3), but
`
`the Board rejected that argument when it found that Hirsch and Martin render
`
`claim 25 obvious. Final Written Decision (FWD), 13, 15-16. Andrea did not
`
`appeal that finding and cannot now contest it, as that finding that was necessary to
`
`the finding of obviousness. Pivoting, Andrea asserts that even if it does not teach
`
`away, Hirsch’s description of Martin’s “disadvantages” would have discouraged a
`
`skilled artisan from combining them. Andrea Br., 3-4. The Board specifically
`
`rejected that proposition when it found that Hirsch’s reference to Martin would
`
`have led the skilled person to consider these references together, FWD, 13, and
`
`that person “would have recognized that techniques such as those shown in both
`
`Hirsch and Martin are routinely combined.” FWD, 16 (quoting Pet., 34-35).
`
`
`
`2
`
`
`
`IPR2017-00626
`
`
`
`
`
`Next, Andrea asserts that the skilled person would not have combined the
`
`references because Hirsch intended his algorithm to work fast and including
`
`Martin, which is slower, would add delay. Andrea Br., 3. Andrea’s assertion does
`
`not withstand scrutiny. First, Andrea ignores how the references are combined:
`
`using Martin’s noise floor algorithm to calculate Hirsch’s adaptive threshold. In
`
`that configuration, Hirsch would continue to calculate its noise estimate as the
`
`average of the noise values (i.e., the values less than the adaptive threshold; values
`
`above the threshold are considered to be speech). Pet., 37; Reply, 18-19. Thus, the
`
`only possible “delay” is in updating the adaptive threshold—there is no delay in
`
`calculating noise estimates. Second, as Apple explained, the difference between
`
`Hirsch and Martin is 225 milliseconds, and a skilled person would have been
`
`motivated to try combining them, recognizing the tradeoff between running time
`
`and performance. Pet., 35-36; see Reply, 18-19.
`
`Andrea also mischaracterizes Hirsch, again arguing Hirsch stops working
`
`after speech starts. Andrea Br., 2. But the Board rejected that argument when it
`
`found Hirsch anticipates claim 3, agreeing that Hirsch operates during continuous
`
`speech. FWD, 8; Pet., 29. Andrea did not appeal that decision, and it is final.
`
`Andrea also asserts that Martin does not perform well in non-stationary
`
`environments because Martin says its algorithm is “capable of” estimating noise in
`
`those environments as opposed to performing well. Andrea Br., 5. But as Apple
`
`
`
`3
`
`
`
`IPR2017-00626
`
`
`
`
`
`explained in its Reply, Martin published his algorithm to report positive results not
`
`to suggest that it did not work well. Reply, 19-20. In sum, the Board should find
`
`that a skilled person would have combined Hirsh and Martin.
`
`III. Martin Discloses the Claimed “Future Minimum” Value
`Andrea argues Martin’s PMmin cannot be the “future minimum” of claims 6-9
`
`based on its mistaken belief that PMmin is not a minimum of a frequency bin. Br., 6.
`
`For the monotonically increasing scenario, Andrea incorrectly asserts that
`
`PMmin cannot be the “future minimum” because the “future minimum” is a “noise
`
`floor tracker” that “continuously adjust[s]… downward” and PMmin can increase
`
`across sub-windows. Br., 7. But PMmin is a noise floor tracker—Martin’s algorithm
`
`“estimate[s] the noise floor… [as] the minimum of a smoothed power estimate
`
`within a window of finite length.” Ex. 1006, 1093. Martin’s PMmin also is used to
`
`set the estimate Pn(i)—i.e., the
`
`noise floor—and thus, it is a noise
`
`floor tracker. Andrea is also
`
`wrong that the “future minimum”
`
`is only adjusted down. Its expert,
`
`Dr. Douglas, illustrated how the
`
`’345 algorithm works (reproduced) and admitted the “future minimum” (red) can
`
`increase across periods. Ex. 1026, 52:21-53:13, 59:20-60:5 (discussing Ex. 1027).
`
`
`
`4
`
`
`
`IPR2017-00626
`
`
`
`
`
`In the non-monotonically increasing case, Andrea asserts that PMmin cannot
`
`be the “future minimum” because the noise estimate Pn is set using min_vec and
`
`not PMmin. But as Apple has explained, min_vec is used to store PMmin values, and
`
`even Dr. Douglas admitted that use of min_vec results in setting Pn(i) to the
`
`smallest PMmin value in a window. Apple Remand Br., 6-7; Reply, 10-11.
`
`IV. Martin Discloses Periodically Updating the Minimums
`Andrea asserts that Martin does not show “periodically” setting (i) a “current
`
`minimum” to the “future minimum” as required in claims 6-8, or (ii) a “future
`
`minimum” to the “current magnitude” as required in claim 9. For claims 6-8,
`
`Andrea’s argument rests on its flawed position that Martin’s use of min_vec to find
`
`the smallest PMmin value means that Pn(i) is not set to a PMmin value—a position
`
`contradicted by Dr. Douglas, as explained above. Martin uses PMmin to update Pn(i)
`
`at a fixed interval—at the end of every sub-window of M samples. Ex. 1026,
`
`77:16-80:19; Apple Remand Br., 8-9; Ex. 1023, ¶20. Andrea’s argument on claim
`
`9 is new and improper, Paper 36 (limiting briefing to arguments previously raised),
`
`and in any event, is wrong. As explained in the Petition, Martin sets PMmin equal to
`
`the current value of the signal at the start of every sub-window of M samples. Pet.,
`
`43-44, 46. That is at regular intervals of time and satisfies claim 9.
`
`V. Conclusion
`Apple respectfully submits that the Board should cancel claims 6-9.
`
`
`
`5
`
`
`
`IPR2017-00626
`
`Dated: May 6, 2020
`
`Respectfully Submitted,
`
`/Jeffrey P. Kushan/
`Jeffrey P. Kushan
`Registration No. 43,401
`Sidley Austin LLP
`1501 K Street NW
`Washington, DC 20005
`jkushan@sidley.com
`(202) 736-8914
`
`Lead Counsel for Petitioner
`
`6
`
`
`
`IPR2017-00626
`
`
`
`
`
`CERTIFICATE OF SERVICE
`
`I hereby certify that on this 6th day of May, 2020, copies of this Petitioner’s
`
`Responsive Remand Brief, and Exhibits have been served in its entirety by email
`
`on the following counsel of record for Patent Owner:
`
`
`William D. Belanger, belangerw@pepperlaw.com
`Frank D. Liu, liuf@pepperlaw.com
`Andrew P. Zappia, zappiaa@pepperlaw.com
`BN_IPR-Andrea@pepperlaw.com
`
`Respectfully submitted,
`
`/Jeffrey P. Kushan/
`Jeffrey P. Kushan
`Reg. No. 43,401
`Attorney for Petitioner
`
`
`
`
`Dated: May 6, 2020
`
`
`
`
`
`
`
`