Paper No. 18
`
`
`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
`____________________
`
`Inter Partes Review No. IPR2017-00626
`__________________________________________________________________
`
`Petitioner’s Reply
`
`
`
`
`
`
`
`
`
`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
`____________________
`
`Inter Partes Review No. IPR2017-00626
`__________________________________________________________________
`
`Petitioner’s Reply
`
`
`
`
`
`
`
`
`IPR2017-00626
`
`
`
`Petitioner’s Reply
`
`Table of Contents
`
`2.
`
`Introduction .................................................................................................... 1
`I.
`II. Claim Construction ....................................................................................... 2
`III. Hirsch Anticipates Claims 1-3, 12-13, 21, 23, and 38 ................................. 2
`IV. Hirsch and Martin Render Claims 4-11, 25, 39-42, and 46 Obvious ........ 3
`A. Martin Discloses the “Future Minimum” of Claims 4 and 39 ........ 3
`1.
`Andrea’s Argument Does Not Apply Where the Martin
`Algorithm Is Configured to Use 1 Sub-Window ........................ 5
`Andrea’s Argument Depends on the Non-Existent Claim
`Requirement that a “Future Minimum” Be a Minimum “Across
`the Entire Window L” ................................................................. 8
`B. Martin Discloses Setting a Current Minimum to a Future
`Minimum Value “Periodically,” as Required by Claim 6 ............. 11
`C. Martin Discloses the “Current Minimum Value” of Claim 10 ..... 13
`D. A Skilled Artisan Would Have Combined Hirsch and Martin ..... 16
`E.
`Andrea’s Criticisms of Dr. Hochwald Are Unfounded ................. 21
`V. A POSA Would Have Considered it Obvious to Modify Hirsch and
`Martin with Conventional Spectral Subtraction Techniques ................. 22
`A. A POSA Would Have Combined Hirsch with Martin and Boll ... 23
`B.
`A POSA Would Have Combined Hirsch with Boll and Arslan .... 24
`C. A POSA Would Have Combined Hirsch with Martin and
`Uesugi.................................................................................................. 25
`VI. Conclusion .................................................................................................... 26
`
`
`
`
`
`ii
`
`
`
`IPR2017-00626
`
`
`
`Petitioner’s Reply
`
`TABLE OF AUTHORITIES
`
`Cases
`KSR Intern. Co. v. Teleflex Inc.,
`127 S.Ct. 1727 (2007) ....................................................................... 19, 21, 24, 25
`
`
`Page(s)
`
`
`
`
`
`iii
`
`
`
`IPR2017-00626
`
`
`
`Petitioner’s Reply
`
`Exhibit List
`
`Reference Name
`U.S. Patent No. 6,363,345
`U.S. Patent No. 6,363,345 File History
`Declaration of Bertrand Hochwald
`[Reserved]
`H. G. Hirsch and C. Ehricher, “Noise estimation techniques for
`robust speech recognition,” Proc. IEEE Int. Conf. Acoustics,
`Speech, Signal Processing, vol. 1, pp. 153 -156, 1995
`(“Hirsch”)
`Rainer Martin, “An Efficient Algorithm to Estimate the
`Instantaneous SNR of Speech Signals,” Proc. Eurospeech, pp.
`1093-96, 1993 (“Martin”)
`Letter from Technische Informationsbibliothek re: Proc.
`Eurospeech 1993 (2 Jan. 2017)
`Proc. Eurospeech 1993 Vol. 2 Table of Contents from
`Technische Informationsbibliothek
`Steven F. Boll, “Suppression of Acoustic Noise in Speech
`Using Spectral Subtraction,” IEEE Transactions on Acoustics,
`Speech, and Signal Processing, Vol. ASSP-27, No. 2, April
`1979 (“Boll”)
`U.S. Patent No. 5,550,924 to Helf (“Helf”)
`U.S. Patent No. 5,706,395 to Arslan (“Arslan”)
`Excerpts from Deller et al., Discrete-Time Processing of Speech
`Signals (1993)
`Excerpt from Merriam-Webster Dictionary (1993)
`Excerpts from Oppenheim and Willsky, Signals and Systems
`(1997)
`U.S. Patent No. 5,459,683 to Uesugi
`Lim and Oppenheim, “Enhancement and Bandwidth
`Compression of Noisy Speech,” Proceedings of the IEEE, vol.
`67, no. 12, pp. 1586-1604, December 1979
`Affidavit of Service in Andrea Elecs. v. Apple Inc., EDNY
`
`Exhibit #
`1001
`1002
`1003
`1004
`1005
`
`1006
`
`1007
`
`1008
`
`1009
`
`1010
`1011
`1012
`
`1013
`1014
`
`1015
`1016
`
`1017
`
`
`
`iv
`
`
`
`1019
`
`1020
`
`1021
`
`1022
`
`IPR2017-00626
`
`
`
`Petitioner’s Reply
`
`Exhibit #
`1018
`
`Reference Name
`In the Matter of Certain Audio Processing Hardware and
`Software and Products Containing the Same, Inv. No. 337-TA-
`949, Claim Construction Order (U.S.I.T.C. Jan. 27, 2016) (“949
`CC Order”)
`In the Matter of Certain Audio Processing Hardware and
`Software and Products Containing Same, Inv. No. 337-TA-949,
`Complainant Andrea Electronics Corp.’s Initial Claim
`Construction Brief (U.S.I.T.C. Oct. 19, 2015) (“Andrea CC
`Br.”)
`In the Matter of Certain Audio Processing Hardware and
`Software and Products Containing Same, Inv. No. 337-TA-949,
`Commission Investigative Staff’s Initial Markman Brief
`(U.S.I.T.C. Oct. 19, 2015) (“OUII CC Br.”)
`Letter from the parties in 337-TA-949 informing ALJ they
`agreed to certain constructions (Nov. 10, 2015) (prior litigation)
`In the Matter of Certain Audio Processing Hardware, Software,
`and Products Containing The Same, Inv. No. 337-TA-1026,
`Verified Complaint Against Apple Inc. and Samsung Inc.
`Under Section 337 of the Tariff Act of 1930, as Amended
`(U.S.I.T.C. Sept. 19, 2016)
`[NEW] 1023 Hochwald Reply Decl.
`[NEW] 1024 Reserved
`[NEW] 1025 Exhibit 2 from Hochwald Deposition
`[NEW] 1026 Transcript from Deposition of Scott Douglas dated Jan. 17,
`2018
`[NEW] 1027 Exhibit 1 from Douglas Dep., Figure 27 depicting Current and
`Future Minima
`[NEW] 1028 Exhibit 2 from Douglas Dep., Dr. Douglas’s mark up of Exhibit
`1
`[NEW] 1029 Exhibit 8 from Douglas Dep., Declaration of Scott Douglas in
`Support of Complainant Andrea’s Claim Construction Brief in
`Inv. No. 337-TA-949 (Oct. 19, 2015)
`[NEW] 1030 Transcript from Deposition of Scott Douglas dated June 16,
`2017, taken in Inv. No. 337-TA-1026
`
`
`
`v
`
`
`
`IPR2017-00626
`
`
`
`Petitioner’s Reply
`
`Reference Name
`Exhibit #
`[NEW] 1031 Rainer Martin, Spectral Subtraction Based on Minimum
`Statistics, Proc. EUSIPCO-94, pp. 1182-85 (1994) (“Martin
`94”)
`[NEW] 1032 H. G. Hirsch, “Estimation of Noise Spectrum and its
`Application to SNR Estimation and Speech Enhancement,”
`Technical Report TR-93-012, International Computer Science
`Institute (1993) (reference [7] in Martin 93)
`[NEW] 1033 D. Van Campernolle, “Noise Adaptation in a Hidden Markov
`Model Speech Recognition System”, Computer Speech and
`Language, Vol. 3, pp. 151-167 (1989) (reference [3] in Martin
`93)
`
`vi
`
`
`
`
`
`
`
`
`
`IPR2017-00626
`
`
`
`Petitioner’s Reply
`
`I.
`
`Introduction
`Patent Owner Andrea raises no challenge to the Board’s initial finding that
`
`Hirsch anticipates claims 1-3, 12-13, 21, 23, and 38, effectively conceding that
`
`these claims are unpatentable. See Paper 11 (“Resp.”), 12. Andrea instead devotes
`
`the bulk of its Response to arguing that dependent claims 4-11 and 39-41—which
`
`cover a process for tracking the noise floor of an audio signal—are patentable over
`
`the combination of Hirsch and Martin. See Resp., 25 (admitting that “claim 4
`
`recites the ‘future minimum’ as a noise floor tracker” (emphasis added)).
`
`But Martin teaches the same noise floor tracking algorithm used in the
`
`claims and disclosed in the patent. As Martin states, “[t]o estimate the noise floor
`
`our algorithm takes the minimum of a [signal] within a window of finite length.”
`
`Ex. 1006(Martin), 1093 (emphasis added). Hirsch and Martin thus together teach
`
`every element of these claims. Unable to seriously dispute that Martin teaches use
`
`of a noise floor tracking algorithm, Andrea next argues that Martin calculates the
`
`noise floor over the wrong data window. Resp., 25. But the claims, by their
`
`explicit terms, do not impose the restraints that Andrea contends differentiates
`
`them from Martin: the ’345 claims do not specify the use of a data “window” nor
`
`do they limit the period over which the noise floor is tracked.
`
`Andrea also asserts a person of ordinary skill in the art (“POSA”) would not
`
`have even considered Hirsch and Martin together. But Andrea’s position lacks
`
`
`
`1
`
`
`
`IPR2017-00626
`
`
`
`Petitioner’s Reply
`
`merit as it rests on non-existent differences between the references. More than a
`
`preponderance of the evidence shows that Hirsch and Martin render these claims
`
`obvious.
`
`Andrea does not dispute that the other challenged claims (which all depend
`
`from either claim 1 or 38) simply recite conventional features of the spectral
`
`subtraction process or that Hirsch and the secondary references disclose these
`
`features. Instead, Andrea contends only that a POSA would not have been
`
`motivated to modify Hirsch to incorporate these known features. But, as explained
`
`the Petition, a POSA would have had ample reason to modify Hirsch to use these
`
`conventional features to solve standard problems present in any spectral
`
`subtraction system. Accordingly, the Board should cancel claims 1-25 and 38-47.
`
`II. Claim Construction
`The Board need not construe any claims because under any reasonable
`
`construction, Hirsch alone or in combination Martin and/or other references
`
`renders the claims obvious.
`
`III. Hirsch Anticipates Claims 1-3, 12-13, 21, 23, and 38
`Andrea does not dispute that claims 1-3, 12-13, 21, 23, and 381 are
`
`anticipated by Hirsch. Resp., 12 (“Patent Owner does not address the anticipation
`
`
`
`1 All challenged claims depend from independent claims 1 and 38.
`
`
`
`2
`
`
`
`IPR2017-00626
`
`
`
`Petitioner’s Reply
`
`grounds with respect to the Hirsch reference.”). The Board should therefore cancel
`
`these claims for the reasons stated in the Petition and Institution Decision.
`
`IV. Hirsch and Martin Render Claims 4-11, 25, 39-42, and 46 Obvious
`Andrea’s primary challenge to these claims is that Martin does not teach the
`
`“future minimum” element of claims 4 and 39, from which claims 5-11 and 40-42
`
`depend, respectively. Andrea also argues that Martin does not teach elements of
`
`claims 6 and 10, and that a POSA would not have combined Hirsch with Martin.2
`
`Andrea’s arguments lack merit.
`
`A. Martin Discloses the “Future Minimum” of Claims 4 and 393
`The Board correctly found that the combination of Hirsch and Martin
`
`renders claims 4 and 39 obvious. Claim 4 depends from claim 1 and specifies
`
`“set[ting] the threshold… in accordance with a current minimum value… of the
`
`corresponding frequency bin; said current minimum value being derived in
`
`accordance with a future minimum value… of the corresponding frequency bin.”
`
`Claim 39 depends from claim 38 and specifies the same limitation.
`
`
`
`2 Because Andrea does not separately address claims 5, 7-9, 11, 25, 40-42, or 46,
`
`Apple does not either.
`
`3 In a heading, Andrea states that Martin does not disclose the “current minimum,”
`
`but it does not substantively analyze that term.
`
`
`
`3
`
`
`
`IPR2017-00626
`
`
`
`Petitioner’s Reply
`
`In the Petition, Apple explained how Martin’s algorithm mapped to the
`
`elements of claim 4. Martin teaches a noise estimation algorithm that tracks the
`
`noise floor of an audio signal over windows of L digital samples, and each window
`
`can correspond to a period of, e.g., 0.625 seconds. Pet., 33, 40; Ex. 1003, ¶¶135-
`
`39. Martin shows that the window L can be divided into an arbitrary number of
`
`sub-windows W, each of length M samples (where L=MxW). Ex. 1003, ¶¶135,
`
`139. Martin uses an example that has 4 sub-windows (W=4), but teaches that any
`
`number of sub-windows can be used. Id.
`
`During each sub-window, Martin uses the variable PMmin (“future minimum”)
`
`to track the signal’s minimum value. Pet., 38-39, 42. At the end of a sub-window,
`
`PMmin is used to update the noise floor Pn(i) (“current minimum”) by either (1)
`
`setting Pn(i) equal to the PMmin observed in that sub-window or (2) setting Pn(i)
`
`equal to the smallest PMmin observed over the past L samples (which corresponds to
`
`W sub-windows). Ex. 1003, ¶¶135-36. Where the PMmin value over the past W sub-
`
`windows is monotonically increasing (meaning that each is higher than the
`
`previous one), update option (1) is used, otherwise option (2) is used. Ex.
`
`1006(Martin), 1094. Martin also teaches a process for immediately updating the
`
`noise floor before a sub-window is over. Pet., 45. If the current value of the signal
`
`((cid:1842)(cid:3364)x) ever drops below the noise floor Pn(i), the noise floor is immediately updated
`
`to be equal to the signal’s current value. Ex. 1003, ¶142.
`
`
`
`4
`
`
`
`IPR2017-00626
`
`
`
`Petitioner’s Reply
`
`In Response, Andrea asserts that Martin does not teach a “future minimum
`
`value,” alleging that PMmin “is not the minimum magnitude of a frequency bin”
`
`because it is the minimum of a sub-window of length M in Martin and not of a
`
`window of length L. Resp., 25-27. Andrea’s argument should be rejected because
`
`it both mischaracterizes Martin and reads non-existent limitations into the claims.
`
`1.
`
`Andrea’s Argument Does Not Apply Where the Martin
`Algorithm Is Configured to Use 1 Sub-Window
`Where Martin is configured to use 1 sub-window (W=1), the sub-window
`
`(M) and window (L) are the same length, and thus the minimum of the sub-window
`
`is also the minimum of the window. Ex. 1003, ¶139; Ex. 1023, ¶¶5, 11. In this
`
`configuration, any purported distinction between windows and sub-windows in
`
`Martin vanishes and Andrea’s argument is irrelevant.
`
`In an attempt to distinguish Martin, Andrea asserts that Martin does not
`
`disclose using a single sub-window. See Resp., 21-22. But as Dr. Hochwald
`
`previously explained, Martin discloses that the number of sub-windows in his
`
`algorithm is a variable W and teaches that the number of sub-windows can be
`
`changed and adapted to a desired configuration. Ex. 1003, ¶139; Ex. 2005, 77:3-
`
`22; see Ex. 1006(Martin), 1094. Dr. Hochwald explains that Martin describes the
`
`operation of his algorithm using equations that use the variables M, L, and W, and
`
`that the equations all work for any positive value of W. Ex. 1023, ¶¶5-7, 9-10; see
`
`Ex. 1003, ¶139. A POSA reading Martin would understand that Martin discloses
`
`
`
`5
`
`
`
`IPR2017-00626
`
`
`
`Petitioner’s Reply
`
`setting the number of sub-windows W to 1. Ex. 1003, ¶¶135, 139; Ex. 2005, 77:3-
`
`22, 88:10-22; Ex. 1023, ¶¶4-5.
`
`To contest Martin’s explicit disclosure, Andrea alleges that a POSA would
`
`not be able to make Martin’s algorithm work where W=1 because that person
`
`would not know which of Martin’s two noise update equations (the monotonically
`
`increasing option or non-monotonically increasing option) to use to update the
`
`noise floor. Resp., 18. But as Dr. Hochwald explains, where W=1 both update
`
`equations are identical, and thus, the choice is irrelevant. Ex. 1023, ¶¶11-12; see
`
`Ex. 2005, 82:1-85:1. As shown in the annotated excerpt below, the noise floor
`
`update equations (blue box) are defined using the variables W and M.
`
`Ex. 1023, ¶20; Ex. 1003, ¶140. As shown on the right, when power is
`
`monotonically increasing Pn(i)=PMmin. When power is not monotonically
`
`
`
`
`
`6
`
`
`
`IPR2017-00626
`
`
`
`Petitioner’s Reply
`
`increasing, Pn(i) is set equal to the minimum of the past W PMmin values that are
`
`stored in min_vec. As Dr. Hochwald explains, after plugging in W=1 to the
`
`equation, the non-monotonically increasing update simplifies to Pn(i)=PMmin. Ex.
`
`1023, ¶¶11-12. Therefore, when W=1, Pn(i) is set to the same PMmin value
`
`irrespective of which update equation is used. See Ex. 2005, 82:1-85:1.
`
`Implicitly recognizing that Martin’s algorithm works with 1 sub-window,
`
`Andrea incorrectly argues that the use of sub-windows is central to Martin’s
`
`algorithm. Resp., 21-22, 44-45. But Martin explains that the central concept of its
`
`algorithm is using minimum values to track a noise floor of an audio signal. Ex.
`
`1006(Martin), 1093 (“The algorithm is based on the observation that a noise power
`
`estimate can be obtained using minimum values of a smoothed power estimate”),
`
`1093 (“[t]o estimate the noise floor our algorithm takes the minimum of a
`
`smoothed power estimate within a window of finite length.”). In contrast to the
`
`central concept of using minimum values to track a noise floor, Martin describes
`
`sub-windows as an optional feature that can reduce “computational complexity
`
`and delay.” Ex. 1006(Martin), 1094; Ex. 1023, ¶7. While Martin states that the
`
`window length L “must be large enough to bridge any peak of speech activity, but
`
`short enough to follow nonstationary noise variations,” he says nothing about how
`
`long a sub-window must be. Ex. 1006(Martin), 1094. Nothing in Martin suggests
`
`that the use of sub-windows is required. Ex. 1023, ¶¶6-8.
`
`
`
`7
`
`
`
`IPR2017-00626
`
`
`
`Petitioner’s Reply
`
`As further evidence of this fact, Prof. Martin wrote a follow-up article
`
`(“Martin 94”) that uses the same noise floor algorithm but does not distinguish
`
`between the monotonically and non-monotonically increasing power cases. Ex.
`
`1023, ¶13 (explaining that Martin 94 does not include a “monotonic decision
`
`block”); see Ex. 1031(Martin94), 1183. As Dr. Hochwald explains, if the
`
`“monotonic decision block” were of central importance to Martin’s algorithm, the
`
`Martin 94 paper describing the same algorithm would presumably at least mention
`
`it. Ex. 1023, ¶13. Instead, Martin 94 does not mention the “monotonic decision
`
`block,” which confirms that a POSA would have considered the block to be an
`
`optional feature of the algorithm. Id.
`
`2.
`
`Andrea’s Argument Depends on the Non-Existent Claim
`Requirement that a “Future Minimum” Be a Minimum
`“Across the Entire Window L”
`Even where Martin’s algorithm is configured to use multiple sub-windows,
`
`it still teaches the claimed “future minimum.” To argue otherwise, Andrea reads
`
`non-existent limitations into the claims to require the “future minimum” to be
`
`calculated over a particular period of time. Resp., 25-27.
`
`Andrea argues that when a signal is monotonically increasing, Martin fails to
`
`disclose a “future minimum” because Martin’s variable PMmin tracks the minimum
`
`of the sub-window of length M and “is not the minimum across the window L.”
`
`Resp., 26 (emphasis added); see id., 25 (“Thus, PMmin is not the minimum of the
`
`
`
`8
`
`
`
`IPR2017-00626
`
`
`
`Petitioner’s Reply
`
`samples in the window L… Martin intentionally does not use the minimum power
`
`value across the entire window L”). Thus, Andrea asserts that when Martin sets
`
`the noise floor Pn(i) equal to the most recent PMmin value, the noise floor is set to
`
`the minimum of the sub-window M but not the minimum of the window L. Id., 25-
`
`26.
`
`Andrea’s argument rests on a limitation found nowhere in the claims.
`
`Nothing in claims 4 or 39 specifies the period over which the “future minimum”
`
`must be calculated. Claims 5 and 40 are the only claims that specify anything
`
`about the period over which the “future minimum” must be calculated, and they
`
`specify simply that the “future minimum” is calculated “over a predetermined
`
`period of time”; they are silent as to how long that period must be. Andrea admits
`
`that Martin shows that PMmin tracks the minimum value during the period M,
`
`(Resp., 24 (“In the monotonically increasing case, PMmin will always represent the
`
`most recent subwindow minimum…”)), and that admission is fatal to Andrea’s
`
`argument. Nothing in claims 4 and 39 (or any other claim) requires use of a
`
`predetermined period of length L, or prohibits tracking the minimum of a sub-
`
`window. Thus, in the scenario where a signal is not monotonically increasing,
`
`Martin satisfies the claims. Ex. 1023, ¶18.
`
`In the scenario where a signal is not monotonically increasing, Andrea
`
`argues that PMmin is not the “future minimum” because Martin sets the noise floor
`
`
`
`9
`
`
`
`IPR2017-00626
`
`
`
`Petitioner’s Reply
`
`Pn(i) equal to the smallest PMmin value from the past W sub-windows, which is not
`
`necessarily the minimum value of the most recent sub-window. Resp., 26-27.
`
`Again, Andrea is reading limitations into the claims. Nothing requires calculating
`
`the “future minimum” over a particular data window. Nor do the claims prohibit
`
`segmenting a data window into sub-windows and using the minimum of a previous
`
`sub-window as the “future minimum”.
`
`Andrea also tries to manufacture a difference between Martin and the claims
`
`by arguing that Martin’s use of the variable “min_vec” to temporarily store the
`
`PMmin values means that the noise floor Pn(i) is never set equal to PMmin because it is
`
`set to min_vec instead. Resp., 27. But Andrea ignores that min_vec stores the
`
`PMmin values and that Martin uses min_vec to determine which PMmin value in the
`
`data window L is the smallest. Ex. 1006(Martin), 1094 (showing
`
`“min_vec(r*M)=PMmin”); Ex. 1023, ¶¶14-16, 19. The result is that Martin sets the
`
`noise floor Pn(i) equal to the smallest PMmin value from min_vec.4 That the smallest
`
`PMmin is temporarily stored in min_vec does not change the fact that the noise floor
`
`Pn(i) is set to it. Ex. 1023, ¶¶14-16, 19.
`
`
`
`4 For example, if W=4, Martin will set Pn(i) equal to the smallest of the past 4 PMmin
`
`values stored in min_vec.
`
`
`
`10
`
`
`
`IPR2017-00626
`
`
`
`Petitioner’s Reply
`
`The distinction Andrea attempts to draw between the “collection of
`
`min_vec” values and PMmin is illusory. As Dr. Douglas has admitted, the min_vec
`
`variable stores the PMmin values, and Martin sets the noise floor Pn(i) equal to the
`
`smallest PMmin value in the data window.
`
`Q And the min vec values are the PMmin values, correct?
`A They are the PMmin values.
`Q So this assignment in Figure 2 of Martin assigns Pn(i) to the
`smallest PMmin in the data window, correct?
`A Yes.
`Q If the data window was segmented into four subwindows, that
`would be the smallest of four PMmin values, correct?
`A That is correct.
`
`Ex. 1030, 177:5-16; Ex. 1026, 87:15-20, 103:7-104:12 (providing similar
`
`testimony).
`
`B. Martin Discloses Setting a Current Minimum to a Future
`Minimum Value “Periodically,” as Required by Claim 6
`With respect to claim 6, Andrea argues that Martin does not teach updating
`
`the current minimum “periodically” by incorrectly stating that Martin sets Pn(i)
`
`(“current minimum”) equal to PMmin (“future minimum”) only “sporadically.” See
`
`Resp., 28. According to Andrea, Martin sets Pn(i) to PMmin only in the
`
`monotonically increasing scenario, and thus, because the signal sometimes
`
`monotonically increases and sometimes does not, this assignment occurs randomly
`
`not periodically. Id.
`
`
`
`11
`
`
`
`IPR2017-00626
`
`
`
`Petitioner’s Reply
`
`First, as with claim 4, Andrea’s argument is irrelevant to the scenario where
`
`Martin is configured to use 1 sub-window (W=1), and it can be rejected for that
`
`reason alone. Second, even where Martin is configured to use multiple sub-
`
`windows (e.g., W=4), Andrea’s argument must be rejected because it rests on
`
`Andrea’s flawed position that, in the non-monotonically increasing case, Martin’s
`
`use of min_vec to determine the smallest PMmin values in a data window means that
`
`the noise floor Pn(i) is not set equal to a PMmin value. As explained above, Andrea’s
`
`distinction between min_vec and PMmin is illusory, as Dr. Douglas admitted. See
`
`§IV.A.2, above; Ex. 1030, 177:5-16 (“Q So this assignment in Figure 2 of Martin
`
`assigns Pn(i) to the smallest PMmin in the data window, correct? A Yes.”); Ex.
`
`1026, 87:15-20, 103:7-104:12.
`
`Martin updates the value Pn(i) after every M samples, as Dr. Douglas
`
`admitted. Ex. 1026, 77:16-80:19. Thus, irrespective of whether the signal is
`
`monotonically increasing or not, Pn(i) is set equal to PMmin at the end of every sub-
`
`window of M samples, as shown in the blue box of the figure below.5 Ex. 1023,
`
`¶20.
`
`
`
`5 Because Martin sets Pn(i) equal to PMmin every M samples, it does so “at regular
`
`intervals of time.” Thus, the Board need not construe “periodically,” because
`
`Martin teaches it even under Andrea’s narrow proposed construction.
`
`
`
`12
`
`
`
`IPR2017-00626
`
`
`
`Petitioner’s Reply
`
`
`C. Martin Discloses the “Current Minimum Value” of Claim 10
`Claim 10 depends from claim 4 and specifies that the “current minimum” is
`
`determined as “the minimum value of the magnitude” of the frequency bin “within
`
`a predetermined period of time.” Andrea argues Martin does not teach this
`
`limitation by mischaracterizing both the claim language and Apple’s position.
`
`First, Andrea erroneously asserts that claims 4 and 10 each recite setting the
`
`“current minimum” to a distinct minimum value. Resp., 32-33. But no such
`
`limitations appear in the claims. Claim 4 specifies “deriving” the “current
`
`minimum” in accordance with a “future minimum value of the magnitude of the
`
`corresponding frequency bin.” Claim 10 then specifies that the “current minimum
`
`value is determined as the minimum value of the magnitude of the corresponding
`
`frequency bin within a predetermined period of time.” Thus, claim 10 simply
`
`requires the “current minimum” of claim 4 to be determined over a “predetermined
`
`
`
`13
`
`
`
`IPR2017-00626
`
`
`
`Petitioner’s Reply
`
`period of time.” If the “current minimum” is set equal to the “future minimum”,
`
`and the “future minimum” is determined as the minimum value of the signal within
`
`a predetermined time period, then the “current minimum” necessarily is as well.
`
`Nothing in claim 10 requires setting the “current minimum” to be a different value
`
`than the “future minimum” as Andrea contends, and even Dr. Douglas has admitted
`
`that they can be set to the same value. Ex. 1026, 47:13-15, 50:13-16.
`
`Next, relying on its misinterpretation of the claims, Andrea asserts that
`
`Apple “relies on the same parameter, Px(i), to satisfy both ‘minimum magnitude’
`
`values recited in claims 4 and 10,” which allegedly improperly maps one element
`
`in Martin to two different claim limitations. Resp., 31. But Apple does not rely on
`
`Px(i) to satisfy claim 4’s “future minimum” nor the requirements of claim 10.
`
`Instead, Apple explained that Martin sets Pn(i) (“current minimum”) equal to PMmin
`
`(“future minimum”), and this meets claim 4’s requirement of “deriving” a “current
`
`minimum” in accordance with a “future minimum.” Pet., 41-43. Because PMmin is
`
`determined as the minimum value of the signal6 over a predetermined period of M
`
`
`
`6 While Martin discloses operating on signal power, not magnitude, Apple
`
`explained it would have been obvious to adapt Martin to use magnitude when
`
`combined with Hirsch. Pet., 34, 40; Ex. 1003, ¶23. Andrea does not contest this.
`
`
`
`14
`
`
`
`IPR2017-00626
`
`
`
`Petitioner’s Reply
`
`samples, so is Pn(i) because it is set equal to PMmin. Pet., 46-47. Thus, the
`
`assignment Pn(i)=PMmin satisfies claim 10 as well.
`
`Andrea also argues that “the future minimum and current minimum search
`
`for a minimum over two different data windows” and therefore they cannot be the
`
`same value. Resp., 32. But nothing in any of the claims requires the current and
`
`future minimums to be set over different data windows. For example, while claim
`
`10 recites determining the “current minimum” within a predetermined period of
`
`time, it says nothing about how the “future minimum” is determined, let alone
`
`specify that a different data window must be used. Nor do any claims require the
`
`current and future minimums to be set to different values. Dr. Douglas admitted
`
`that both minimums can have the same value, (Ex. 1026, 47:13-15, 50:13-16), and
`
`both can be set based on the minimum of the current period, (id., 49:16-21 (current
`
`minimum is the minimum of the current frame), 51:16-21 (future minimum is the
`
`minimum of the current frame)).
`
`Andrea’s reading of the claims also is inconsistent with the ’345
`
`specification, which shows setting the current minimum and the future minimum
`
`equal to the same value “Y(n)” calculated during the same period. See Ex. 1001,
`
`Fig. 7, blocks 716 & 718 (highlighted below); id., 8:29-36.
`
`
`
`15
`
`
`
`IPR2017-00626
`
`
`
`Petitioner’s Reply
`
`
`
`Thus, Andrea’s assertion that the current and future minimums must be set
`
`to different values using different parameters is contradicted by the ’345 patent.
`
`D. A Skilled Artisan Would Have Combined Hirsch and Martin
`In the Petition, Apple described the similarities between the Hirsch and
`
`Martin algorithms, and explained why the skilled artisan reading Hirsch would
`
`have been motivated to combine it with features of Martin. Pet., 34-38. In
`
`Response, Andrea alleges that Apple failed to set forth a rationale for this
`
`combination. Resp., 33. Andrea’s arguments are factually flawed for several
`
`reasons.
`
`First, Andrea is incorrect that Hirsch “disparages” Martin. Hirsch’s
`
`statement that “most” known noise estimation approaches have a disadvantage of
`
`requiring “relatively long past segments of noisy speech” would not have
`
`
`
`16
`
`
`
`IPR2017-00626
`
`
`
`Petitioner’s Reply
`
`discouraged a skilled artisan from combining Hirsch with Martin. Ex. 1023, ¶¶32,
`
`36-37. When Hirsch cites Martin, Hirsch also cites to references [3] and [7].
`
`Reference [7] is Hirsch’s own work, and it would be odd for an author to
`
`“disparage” his own work.
`
`To the extent Hirsch is criticizing the cited articles, that criticism would not
`
`have been directed at Martin. In Hirsch’s prior article [7], he points out that a
`
`variety of windows from 0.250 to 2.0 seconds can be considered for estimating
`
`noise in speech. Ex. 1032(Hirsch93), 10. As Dr. Hochwald explains, Hirsch’s
`
`prior article recognizes that there is a tradeoff between the accuracy of a noise
`
`estimate and the ability to adapt to the noise and that the optimum choice of
`
`window depends on the nature of the speech signals being considered. Ex. 1023,
`
`¶¶33, 37. The other article Hirsch cites describes a system that requires at least 10
`
`seconds of noisy speech. Ex. 1023, ¶35; Ex. 1033(Campernolle), 6 (“Parameter
`
`estimates are updated in block mode at regular intervals, typically 30 secs, at which
`
`time the histograms are also restarted”), 14 (“histograms are restarted every 10 secs
`
`of (presumed) speech”), 15 (discussing “effective time constants of 20 to 50 secs
`
`depending on the mixing ratio of speech and silence.”). In comparison to these
`
`other cited approaches, the difference between Hirsch’s 0.4 seconds and Martin’s
`
`0.625 seconds is trivial. Ex. 1023, ¶¶34-37.
`
`
`
`17
`
`
`
`IPR2017-00626
`
`
`
`Petitioner’s Reply
`
`Even if Hirsch’s comment is interpreted as criticizing Martin, it does not rise
`
`to the level of “teaching away” from Martin. See Ex. 1023, ¶¶32, 36-37
`
`(explaining Hirsch’s comment would not discourage a POSA from combining).
`
`As Dr. Hochwald explains, “[t]o a person knowledgeable in the arts, the difference
`
`between 0.4 seconds and 0.625 seconds is not very significant in capturing speech
`
`data. Martin’s algorithm has some advantages in that it had been tested to work
`
`well with non-stationary noise. A person of ordinary skill would have considered
`
`Martin’s benefits to outweigh the hypothetical cost described by Hirsch.” Ex.
`
`1003, ¶129. Hirsch itself describes adding his algorithm into systems that operate
`
`on 600 ms segments of speech, and thus, plainly does not believe systems using
`
`speech segments longer than 400 ms are incompatible with his algorithm. Ex.
`
`1005(Hirsch), 155.
`
`Second, Andrea argues that because Hirsch does not present experimental
`
`results showing that it performs poorly in non-stationary noise environments, a
`
`POSA would not have been motivated to improve Hirsch’s performance in such
`
`environments. Resp., 36-38. That Hirsch states it works will in stationary noise
`
`environments but is silent about non-stationary environments, suggests that it does
`
`not perform well. Ex. 1003, ¶¶131-32. Moreover, as Dr. Hochwald explains, in
`
`the field of audio signal processing, “[i]t was standard to attempt to optimize
`
`system performance by swapping algorithms or tuning parameters.” See Ex. 1003,
`
`
`
`18
`
`
`
`IPR2017-00626
`
`
`
`Petitioner’s Reply
`
`¶¶131-32. Even if Hirsch had some ability to track non-stationary noise, the
`
`skilled artisan would have been motivated to add in Martin’s algorithm to
`
`determine whether it performed better. Ex. 2005, 129:1-23. As the Supreme Court
`
`explained, “if a technique has been used to improve one de
Accessing this document will incur an additional charge of $.
After purchase, you can access this document again without charge.
Accept $ Charge
Still Working On It
This document is taking longer than usual to download. This can happen if we need to contact the court directly to obtain the document and their servers are running slowly.
Give it another minute or two to complete, and then try the refresh button.
A few More Minutes ... Still Working
It can take up to 5 minutes for us to download a document if the court servers are running slowly.
Thank you for your continued patience.
This document could not be displayed.
We could not find this document within its docket. Please go back to the docket page and check the link. If that does not work, go back to the docket and refresh it to pull the newest information.
Your account does not support viewing this document.
You need a Paid Account to view this document. Click here to change your account type.
Your account does not support viewing this document.
Set your membership
status to view this document.
With a Docket Alarm membership, you'll
get a whole lot more, including:
- Up-to-date information for this case.
- Email alerts whenever there is an update.
- Full text search for other cases.
- Get email alerts whenever a new case matches your search.
One Moment Please
The filing “” is large (MB) and is being downloaded.
Please refresh this page in a few minutes to see if the filing has been downloaded. The filing will also be emailed to you when the download completes.
Your document is on its way!
If you do not receive the document in five minutes, contact support at support@docketalarm.com.
Sealed Document
We are unable to display this document, it may be under a court ordered seal.
If you have proper credentials to access the file, you may proceed directly to the court's system using your government issued username and password.
Access Government Site
