`
`Filed on behalf of:
`Andrea Electronics Corporation
`By: William D. Belanger (Lead Counsel)
`
`belangerw@pepperlaw.com
`
`Frank D. Liu (Back-up Counsel)
`
`liuf@pepperlaw.com
`
`Andrew P. Zappia (Back-up Counsel)
`
`zappiaa@pepperlaw.com
`
`
`
`Paper No. ____
`Date: April 22, 2020
`
`
`
`UNITED STATES PATENT AND TRADEMARK OFFICE
`_________________
`
`BEFORE THE PATENT TRIAL AND APPEAL BOARD
`_________________
`
`APPLE INC.,
`Petitioner,
`
`v.
`
`ANDREA ELECTRONICS CORPORATION,
`Patent Owner.
`
`_________________
`
`Case IPR2017-00626
`Patent 6,363,345
`_________________
`
`
`PATENT OWNER’S OPENING REMAND BRIEF
`
`
`
`
`
`#57830529
`
`
`
`IPR2017-00626
`
`
`
`PO’s Opening Remand Brief
`
`
`TABLE OF CONTENTS
`
`
`PAGE
`
`
`TABLE OF AUTHORITIES .................................................................................... ii
`
`I.
`
`II.
`
`INTRODUCTION ........................................................................................... 1
`
`THE ORDINARY ARTISAN WOULD NOT HAVE BEEN
`MOTIVATED TO COMBINE HIRSCH AND MARTIN ............................... 1
`
`
`III. CLAIMS 6-9 ARE PATENTABLE OVER HIRSCH AND MARTIN ........... 6
`
`
`A. Martin Does Not Disclose the Claimed “Future Minimum” Recited
`in Claims 6-9 of the ’345 Patent............................................................ 6
`
`
`
`B. Apple Fails to Show How Martin Teaches the “Periodically”
`Limitation Recited in Claims 6-9 as Construed by the Board .............. 9
`
`
`IV. CONCLUSION .............................................................................................. 10
`
`
`
`i
`
`
`
`IPR2017-00626
`
`
`
`PO’s Opening Remand Brief
`
`
`
`
`TABLE OF AUTHORITIES
`
`PAGE
`
`
`CASES
`
`Apple Inc. v. Andrea Elec. Corp.,
`949 F.3d 697, 707-09 (Fed. Cir. 2020) ............................................................ 9
`
`
`Arctic Cat Inc. v. Bombardier Recreational Prods. Inc.,
`876 F.3d 1350 (Fed. Cir. 2017), cert. denied, 139 S. Ct. 143 (2018) ............. 4
`
`
`In re Ethicon, Inc.,
`844 F.3d 1344 (Fed. Cir. 2017) ....................................................................... 4
`
`
`In re Pagliaro,
`657 F.2d 1219 (CCPA 1981) ........................................................................... 3
`
`
`In re Schweickert,
`676 Fed. Appx. 988 (Fed. Cir. 2017) .............................................................. 5
`
`
`In re Van Os,
`844 F.3d 1359 (Fed. Cir. 2017) ....................................................................... 6
`
`
`Polaris Indus., Inc. v. Arctic Cat, Inc.,
`882 F.3d 1056 (Fed. Cir. 2018) ....................................................................... 4
`
`
`
`
`ii
`
`
`
`IPR2017-00626
`
`
`
`I.
`
`INTRODUCTION
`
`PO’s Opening Remand Brief
`
`
`This case was remanded to consider Apple’s Reply argument that claims 6-9
`
`are obvious in view of the Hirsch and Martin combination.1 That argument should
`
`be rejected and for multiple reasons. First, a skilled artisan would not be motivated
`
`to combine these references. Hirsch specifically identifies Martin and expressly
`
`explains the “disadvantage” of Martin’s approach, thus teaching away from it.
`
`Apple’s combination is motivated only by hindsight. And far from a “simple
`
`modification,” the combination would defeat Hirsch’s purpose, by eliminating its
`
`algorithm—i.e., its core teaching. The alleged benefits of this radical
`
`transformation of Hirsch are not substantiated in Hirsch, Martin, or anywhere else.
`
`Second, even if one were to disregard the teaching away, the combination of
`
`Hirsch and Martin fails to teach or make obvious all limitations of claims 6-9.
`
`Apple can neither show that Martin’s multiple sub-window teachings disclose the
`
`claimed “future minimum,” nor show that Martin’s multiple sub-window teachings
`
`disclose the “periodically” limitation recited in claims 6-9 as that term has been
`
`construed by the Board and affirmed by the Federal Circuit.
`
`II. THE ORDINARY ARTISAN WOULD NOT HAVE BEEN
`MOTIVATED TO COMBINE HIRSCH AND MARTIN
`
`The Board already rejected Apple’s attempt to combine Hirsch with Martin.
`
`
`1 The defined terms herein have the meanings ascribed to them in the POR.
`
`1
`
`
`
`IPR2017-00626
`
`
`
`PO’s Opening Remand Brief
`
`
`See FD at 13. In its Petition, Apple modified Martin’s multiple sub-window
`
`approach to create a single sub-window scenario that it could try to map to claims
`
`6-9. Apple resorted to this modification because it could not plausibly argue that
`
`Martin’s multiple sub-window teachings rendered the claims obvious. See id. at
`
`12. The Board correctly found Apple’s modifications to be “directly contrary” to
`
`Martin’s teachings. Id. at 12-13.
`
`In Reply, Apple then attempts to combine Hirsch with Martin’s multiple
`
`sub-window approach. The Board should also reject this new argument, for Hirsch
`
`clearly discourages Martin’s multiple sub-window approach. Hirsch expressly
`
`identifies Martin, and it teaches away from use of Martin’s complex estimate.
`
`Hirsch teaches a noise estimation technique that uses an adaptive threshold
`
`to detect the onset of speech thereby allowing noise to be estimated based on the
`
`signal directly before the onset of speech. EX1005 at 153. Hirsch’s noise
`
`estimation algorithm improves upon the prior noise estimation algorithms by
`
`removing the need for explicit speech pauses and relatively long past segments of
`
`noisy speech. Id. Hirsch avoids the need for these long past segments through a
`
`simple recursive accumulation of the signal power until the signal power exceeds
`
`the adaptive threshold. Id. at 153-154. When the threshold is exceeded, Hirsch
`
`stops the recursive accumulation of signal power, thus providing an estimate of the
`
`noise level. Id. In other words, Hirsch’s noise estimate is derived by measuring
`
`2
`
`
`
`IPR2017-00626
`
`
`
`PO’s Opening Remand Brief
`
`
`the signal power during a period when no speech is detected. Id.
`
`Hirsch describes Martin’s (and others’) approach as having a clear
`
`“disadvantage”: the “need [for] relatively long past segments of noisy speech.”
`
`EX1005 at 153 (identifying Martin as reference [6]); see EX2002, ¶¶81-92; see
`
`also FD at 13. Martin’s complex algorithm requires relatively long past segments
`
`of noisy speech in order to generate a sufficient number of sub-windows to
`
`distinguish between rapid and slowly varying noise power. See EX2002, ¶¶60-61.
`
`A consequence of Martin’s algorithm is that it introduces a “bias” (i.e. inaccuracy)
`
`during non-speech periods. See EX2002, ¶81; EX1006 at 1095, 1096. Hirsch
`
`avoids the “disadvantage” of Martin’s required “long past segments of noisy
`
`speech” through its simpler recursive accumulation technique that estimates the
`
`noise using the signal preceding the onset of speech (i.e. non-speech periods).
`
`EX2002, ¶45. Combining Martin and Hirsch would introduce Martin’s “bias” to
`
`Hirsch’s non-speech periods thereby negatively impacting the functionality of
`
`Hirsch and would introduce complex noise estimate computations during speech
`
`periods incompatible with Hirsch’s simple noise reduction scheme. Id., ¶87.
`
`Apple’s attempts to force a combination of Hirsch and Martin despite this clear
`
`discouragement of each other’s methods would be error. See, e.g., In re Pagliaro,
`
`657 F.2d 1219, 1225 (CCPA 1981).
`
`“Prior art teaches away when ‘a person of ordinary skill, upon reading the
`
`3
`
`
`
`IPR2017-00626
`
`
`
`PO’s Opening Remand Brief
`
`
`reference, would be discouraged from following the path set out in the reference,
`
`or would be led in a direction divergent from the path that was taken by the
`
`[patent].’” Arctic Cat Inc. v. Bombardier Recreational Prods. Inc., 876 F.3d 1350,
`
`1360 (Fed. Cir. 2017), cert. denied, 139 S. Ct. 143 (2018). Hirsch’s “clear
`
`discouragement from use” of Martin’s algorithm teaches away. In re Ethicon, Inc.,
`
`844 F.3d 1344, 1351 (Fed. Cir. 2017). Even if it did not teach away, Hirsch’s
`
`“statements regarding preferences are relevant to a finding regarding whether a
`
`skilled artisan would be motivated to combine” it with Martin. Polaris Indus., Inc.
`
`v. Arctic Cat, Inc., 882 F.3d 1056, 1069 (Fed. Cir. 2018). Apple’s combination is
`
`driven by hindsight in contravention of both Hirsch’s and Martin’s teachings.
`
`Apple contends: (1) that the ordinary artisan would overlook the difference
`
`in the length of past segments of noisy speech in Hirsch and Martin as “trivial”
`
`(Reply at 17); (2) that Hirsch and Martin can be easily combined (id. at 20-21); (3)
`
`that the ordinary artisan would been motivated to “modify[ ] Hirsch’s adaptive
`
`threshold calculation ‘to use Martin’s noise floor algorithm instead of Hirsch’s
`
`running average algorithm for deciding the threshold” (id. at 20); and (4) that this
`
`modification would have been “one simple substitution for one parameter in
`
`Hirsch’s algorithm” (id. at 21). But Apple’s modification would defeat the
`
`purpose of Hirsch’s noise estimation technique by removing the heart of Hirsch’s
`
`noise estimation technique – the recursive accumulation noise estimate – and
`
`4
`
`
`
`IPR2017-00626
`
`
`
`PO’s Opening Remand Brief
`
`
`replace it with the biased and inaccurate process that Hirsch was designed to
`
`improve. EX2002, ¶¶87-91; see also EX1005 at 153. Far from a “simple
`
`modification,” this is a fundamental change of Hirsch’s methodology. See In re
`
`Schweickert, 676 Fed. Appx. 988, 994 (Fed. Cir. 2017) (proposed substitution was
`
`“illogical[,] as it would leave [the reference] void of the mechanism it relies on”;
`
`“removing [that element] defeats the purpose of [the reference]’s system.”).
`
`Apple’s Reply further suggests there is a motivation for this modification to
`
`improve performance in non-stationary environments. Reply at 18-19. The record
`
`offers no support for this theory. Nothing in Hirsch discusses how its algorithm
`
`might perform in non-stationary noise environments. EX2002, ¶88. There is no
`
`evidence that Martin’s algorithm would have so improved Hirsch’s estimation
`
`performance as to motivate the ordinary artisan to modify Hirsch to incorporate it.
`
`See EX2002, ¶87. As for Martin, all it says is that its algorithm is capable of
`
`estimating non-stationary noise variations; it boasts of no special benefit of its
`
`algorithm in tracking such variations. Id. On the contrary, it admits that its
`
`algorithm does not perform as well in challenging environments, such as rapid
`
`noise variations, and when speech is not present. Id. Such admitted weaknesses,
`
`without evidence of countervailing advantages, undermine the purported
`
`motivation to improve non-stationary performance by incorporating Martin’s
`
`algorithm into Hirsch. Apple’s new theory is classic “ex post reasoning” that “fails
`
`5
`
`
`
`IPR2017-00626
`
`
`
`PO’s Opening Remand Brief
`
`
`to identify any actual reason why a skilled artisan would have combined the
`
`elements in the manner claimed,” In re Van Os, 844 F.3d 1359, 1361 (Fed. Cir.
`
`2017) (emphasis in original), and it is at least as deficient as Apple’s prior rejected
`
`effort to combine Hirsch and Martin.
`
`III. CLAIMS 6-9 ARE PATENTABLE OVER HIRSCH AND MARTIN
`
`Even if, arguendo, Martin were properly combinable with Hirsch (it is not),
`
`that combination still could not render claims 6-9 obvious, for multiple reasons.
`
`Even under Apple’s new reliance on Martin’s multiple sub-window teaching, the
`
`references do not disclose or render obvious the “future minimum” of claims 6-9,
`
`and even if they did, Martin fails to disclose or render obvious “periodically”
`
`setting the “future minimum” to the “current minimum” of claims 6-8, or (ii)
`
`“periodically” setting the “future minimum” to the “current magnitude” of claim 9.
`
`A. Martin Does Not Disclose the Claimed “Future Minimum”
`Recited in Claims 6-9 of the ’345 Patent
`
`Apple alleges that Martin’s PMmin parameter teaches the “future minimum”
`
`limitation in claims 6-9. Reply at 8-11; Pet. at 38-41. Martin’s PMmin parameter
`
`represents the most recent sub-window minimum and is used to distinguish
`
`between slowly and rapidly varying noise power. EX2002, ¶66; EX1006 at 1094.
`
`Critically, Martin’s PMmin parameter does not represent a minimum power value of
`
`the frequency bin, as required by the claims. EX2002, ¶¶66-71.
`
`Claim 5 of the ’345 Patent (from which claims 6-9 depend) requires the
`
`6
`
`
`
`IPR2017-00626
`
`
`
`PO’s Opening Remand Brief
`
`
`“future minimum” to be a “minimum value of the magnitude of the corresponding
`
`frequency bin.” See EX1001, 9:61-10:12. In Martin’s case of rapidly varying
`
`(monotonically increasing) noise, PMmin will always represent the most recent sub-
`
`window minimum, which is always the largest minimum of all sub-windows so far
`
`across window L – not a minimum value of the magnitude as required by claims 6
`
`to 9. See EX1006 at 1094; EX2002, ¶¶66-68; POR at 24-26. Dispositively, in its
`
`Reply, Apple concedes exactly that. Reply at 4.
`
`Apple asserts that nevertheless Martin’s PMmin meets the “future minimum”
`
`limitation because PMmin is the minimum of at least one sub-window. Id. at 8-9.
`
`Apple misinterprets both the “future minimum” limitation and the teachings of
`
`Martin. The ’345 Patent explains that its “future minimum” is a noise floor
`
`tracker, which finds the smallest magnitude value of the signal within a frequency
`
`bin by continuously adjusting the future minimum downward based on smaller
`
`noise magnitudes to prevent too high of a noise estimation. See EX1001, 6:23-32,
`
`6:42-45, 9:54-10:12; EX2002, ¶67; POR at 24-26. The ordinary artisan would
`
`understand that Martin’s PMmin does the opposite of what the “future minimum”
`
`does: the “future minimum” continuously adjusts downward, while Martin’s PMmin,
`
`continually adjusts upward as each sub-window across window L features a
`
`successively higher minimum. Martin does this to achieve its purpose of detecting
`
`rapidly varying noise. Without this operation of PMmin, Martin is not able to
`
`7
`
`
`
`IPR2017-00626
`
`
`
`PO’s Opening Remand Brief
`
`
`identify the signal condition. See EX1006 at 1094; EX2002, ¶¶67-68. As such,
`
`Martin’s PMmin fundamentally differs from the ’345 Patent’s “future minimum.”
`
`Apple’s Reply is forced to contend that Martin’s PMmin parameter discloses
`
`the “future minimum” in the slowly varying (non-monotonically increasing) noise
`
`case because “Martin sets the noise floor Pn(i) equal to the smallest PMmin value
`
`from the past W sub-windows.” See Reply at 9-10. However, Martin’s PMmin
`
`represents the minimum of the most current sub-window, and does not represent
`
`the minimum power value of the frequency bin – which is the “future minimum”
`
`of the ’345 Patent. EX2002, ¶¶69-70.
`
`Apple next contends that PMmin is a minimum value of a frequency bin
`
`because PMmin values are used to populate an array min_vec, and the minimum of
`
`those min_vec values is used to set the value of Pn(i) in the slowly varying (non-
`
`monotonically increasing) noise case. Reply at 9-10. Apple’s argument flatly
`
`mischaracterizes the operation of Martin’s PMmin. PMmin merely determines the
`
`minimum power of the current sub-window and does not purport to represent the
`
`minimum magnitude of the frequency bin. EX2002, ¶¶69-71. Instead, as Martin
`
`describes in its slowly varying noise case, the minimum power is determined by
`
`finding the minimum value of the min_vec array, which is stored as Pn(i) – which
`
`Apple insists is the “current minimum,” not the “future minimum.” See FD at 10
`
`(noting Petitioner identifies Pn(i) as the “current minimum”).
`
`8
`
`
`
`IPR2017-00626
`
`
`
`PO’s Opening Remand Brief
`
`
`B. Apple Fails to Show How Martin Teaches the “Periodically”
`Limitation Recited in Claims 6-9 as Construed by the Board
`
`Even if Martin taught the claimed “future minimum,” Martin’s algorithm
`
`does not “periodically” set a “current minimum” to the “future minimum” as
`
`required in claims 6-8, or “periodically” set a “future minimum” to the “current
`
`magnitude” as required in claim 9. See EX2002, ¶¶72-75; POR at 28-30. As the
`
`Board found and the Federal Circuit affirmed, “periodically” means “at regular
`
`intervals of time,” not merely from time to time. Apple Inc. v. Andrea Elec. Corp.,
`
`949 F.3d 697, 707-09 (Fed. Cir. 2020).
`
`In its analysis of claims 6-8, Apple maps Martin’s Pn(i) and PMmin
`
`parameters to the claimed “current minimum” and “future minimum,” respectively.
`
`Pet. at 38. While Martin does set Pn(i) to PMmin in the rapidly varying
`
`monotonically increasing noise case, it does not do so periodically. Instead, it sets
`
`Pn(i) to PMmin randomly. EX1006 at 1094; EX2002, ¶¶72-75; POR at 28-30. In the
`
`non-monotonically increasing case, on the other hand, Martin sets Pn(i) to one of
`
`the stored values in the min_vec array—not to PMmin. EX1006 at 1094; EX2002,
`
`¶¶72-75; POR at 28-30. And even if the values in the min_vec array were
`
`considered PMmin values, the non-monotonically increasing case also occurs
`
`randomly. Thus, Martin does not teach setting Pn(i) to PMmin at “regular intervals
`
`of time,” but, rather based upon that happenstance of signal characteristics. See
`
`EX1006 at 1094; EX2002, ¶¶72-75; POR at 28-30.
`
`9
`
`
`
`IPR2017-00626
`
`
`
`PO’s Opening Remand Brief
`
`
`Still attempting to argue that Martin somehow teaches setting Pn(i) to PMmin
`
`“periodically,” Apple tries to conflate PMmin and the min_vec array and argue that
`
`Martin sets Pn(i) to PMmin regardless of the signal characteristics. Reply at 11-13.
`
`As discussed, the min_vec array does not represent a “future minimum” – it is an
`
`array of minimum sub-window values, which Martin needs to subsequently
`
`evaluate to determine a minimum power of the window. That minimum power
`
`value is stored in Pn(i), a value Apple does not allege to be the “future minimum.”
`
`See supra III.A. Even if, arguendo, Martin’s algorithm did set Pn(i) to PMmin in
`
`both the monotonically and non-monotonically increasing cases, PMmin still does
`
`not satisfy the “future minimum,” for all the reasons already discussed. Id.
`
`As for claim 9, Apple never even attempted to show Martin “periodically”
`
`sets the “future minimum” to a “current magnitude” under the Board’s construction
`
`of “periodically.” See Pet. at 46. Apple asserts that Martin sets the alleged “future
`
`minimum” (PMmin) to the alleged “current magnitude” (𝑃̅x(i)): “whenever 𝑃̅x(i)<
`
`PMmin.” Id. That contention is insufficient to meet the Board’s construction of
`
`“periodically,” which requires the setting of the values to occur “at regular
`
`intervals of time” – not “whenever.” Apple incorrectly relies on a condition in
`
`Martin that occurs at random, unpredictable times. See, e.g., EX2002, ¶¶72, 75.
`
`IV. CONCLUSION
`
`Petitioner has not met its burden of proving invalidity of claims 6-9.
`
`10
`
`
`
`IPR2017-00626
`
`
`
`
`
`PO’s Opening Remand Brief
`
`
`Respectfully submitted,
`
`
`
`
`/Frank D. Liu/
`Frank D. Liu (Back-up Counsel)
`Registration No. 64,682
`
`Date: April 22, 2020
`
`Pepper Hamilton LLP
`125 High Street, 19th Floor
`Boston, MA 02110
`Tel: (617) 204-5117
`Fax: (617) 204-5150
`E-mail: liuf@pepperlaw.com
`
`
`11
`
`
`
`IPR2017-00626
`
`
`
`PO’s Opening Remand Brief
`
`
`
`CERTIFICATE OF SERVICE
`
`Pursuant to 37 C.F.R. § 42.6(e), this is to certify that on this 22nd day of
`
`April, 2020, I caused to be served a true and correct copy of the foregoing
`
`PATENT OWNER’S OPENING REMAND BRIEF by e-mail on the following
`
`counsel of record (as agreed in the Service Information section of the Petition):
`
`Jeffrey P. Kushan (Lead Counsel)
`E-mail: jkushan@sidley.com
`Service e-mail: iprnotices@sidley.com
`
`Steven S. Baik (Back-up Counsel)
`E-mail: sbaik@sidley.com
`Service e-mail: iprnotices@sidley.com
`
`Thomas A. Broughan III (Back-up Counsel)
`E-mail: tbroughan@sidley.com
`Service e-mail: iprnotices@sidley.com
`
`Counsel for Petitioner
`
`
`
`
`
`
`
`
`
`
`/Frank D. Liu/
`Frank D. Liu (Back-up Counsel)
`Registration No. 64,682
`
`
`
`