By: William D. Belanger
`Pepper Hamilton LLP
`125 High Street
`19th Floor, High Street Tower
`Boston, MA 02110
`(617) 204-5100 (telephone)
`(617) 204-5150 (facsimile)
`belangerw@pepperlaw.com
`
`UNITED STATES PATENT AND TRADEMARK OFFICE
`___________________
`
`BEFORE THE PATENT TRIAL AND APPEAL BOARD
`___________________
`
`APPLE INC.
`Petitioner
`
`v.
`
`ANDREA ELECTRONICS CORPORATION
`Patent Owner
`___________________
`
`Case No. IPR2017-00626
`U.S. Patent 6,363,345
`___________________
`
`
`PATENT OWNER’S RESPONSE
`
`
`
`
`
`
`
`
`
`
`
`

`

`TABLE OF CONTENTS
`
`IPR2017-00626
`Patent 6,363,345
`
`
`Page(s)
`
`
`Table of Authorities ................................................................................................. iii
`
`Table of Exhibits .........................................................................................................
`
`I.
`
`INTRODUCTION .......................................................................................... 3
`A.
`Related Proceedings ............................................................................. 3
`B.
`Institution Decision .............................................................................. 4
`
`II. OVERVIEW OF THE ’345 PATENT ........................................................... 6
`
`III. CLAIM CONSTRUCTION AND LEVEL OF ORDINARY SKILL
`IN THE ART .................................................................................................. 7
`A. A Person Having Ordinary Skill In The Art ........................................ 8
`B.
`Claim Construction............................................................................... 8
`
`3.
`
`IV. CLAIMS 4-20, 22, 24-25, and 39-47ARE NOT OBVIOUS OVER
`HIRSCH IN VIEW OF ANY SECONDARY REFERENCE ..................... 12
`A. Grounds Based on the Combination of Hirsch and Martin ................ 13
`1.
`Summary of Hirsch .................................................................. 13
`2.
`Summary of Martin .................................................................. 15
`a. Martin’s Techniques Allegedly Reduce Delay ..............19
`b.
`The Role of Subwindows in Martin’s Algorithm ...........21
`c. Martin’s SNR Computation ............................................22
`Claims 4-11 and 39-42 are not obvious over Hirsch and
`Martin ....................................................................................... 23
`a.
`The combination of Hirsch and Martin does not
`teach or disclose a “current minimum” and a
`“future minimum” ...........................................................23
`PMmin is Not a “Future Minimum” in the
`(1)
`Rapidly Varying Noise Power Case (i.e., for
`monotonically increasing signals) ........................24
`(2) There is No “Future Minimum” in the
`Slowly Varying Noise Power Case (i.e., for
`non-monotonically increasing signals) ................26
`The Alleged “Current Minimum,” Pn(i) is Not Set
`to the Alleged “Future Minimum,” PMmin
`“Periodically” .................................................................28
`
`a)
`
`i
`
`

`

`IPR2017-00626
`Patent 6,363,345
`
`
`b.
`
`b.
`
`c.
`
`The Combination of Hirsch and Martin Fails to
`disclose the “Current Magnitude” of Claim 10 ..............30
`4. Method Claims ......................................................................... 33
`5.
`A Skilled Artisan Would Not Have Been Motivated To
`Combine Hirsch And Martin ................................................... 33
`a.
`Hirsch’s Citation Of Martin Would Not Have
`Motivated A Skilled Artisan To Combine The
`References .......................................................................34
`Apple’s Assertion That Hirsch’s Performance In
`Non-Stationary Noise Environments Could Be
`Improved By Martin’s Algorithm Lacks Rational
`Underpinnings.................................................................36
`Apple Fails To Explain How A Skilled Artisan
`Would Have Combined Elements From Hirsch
`And Martin In The Manner Arranged In The
`Claims .............................................................................40
`Apple’s Validity Positions are Based on Dr. Hochwald’s
`Incomplete Analysis and His Incorrect Understanding of
`the Martin System .................................................................... 43
`a.
`Apple’s Attempt to Eliminate Subwindows Is
`Contrary to Martin’s Express Disclosure .......................44
`Dr. Hochwald Failed to Analyze the Monotonic
`Decision Block................................................................46
`Dr. Hochwald Has an Erroneous Understanding of
`the Sample Counter and the Update of the
`Min_Vec Array ...............................................................47
`B. Grounds Based On The Combination Of Hirsch And The Other
`Relied-Upon Secondary References................................................... 51
`1.
`Summary Of The Additional Relied-Upon Secondary
`References ................................................................................ 51
`Apple Fails to Establish That A Skilled Artisan Would
`Have Been Motivated to Combine Hirsch and Boll ................ 52
`Ground Based on the Combination of Hirsch, Martin, and
`Boll ........................................................................................... 54
`Ground Based on the Combination of Hirsch, Boll, and
`Arslan ....................................................................................... 56
`Grounds Based on the Combinations of Hirsch and
`Uesugi, and Hirsch, Martin, and Uesugi .................................. 59
`
`b.
`
`c.
`
`6.
`
`2.
`
`3.
`
`4.
`
`5.
`
`V.
`
`CONCLUSION ............................................................................................. 61
`
`ii
`
`

`

`TABLE OF AUTHORITIES
`
`
`CASES
`
`IPR2017-00626
`Patent 6,363,345
`
`
`Page(s)
`
`CFMT, Inc. v. Yieldup Int’l Corp., 349 F.3d 1333 (Fed. Cir. 2003) ..............................................10
`
`Cuozzo Speech Techs., LLC v. Lee, 136 S. Ct. 2131 (2016) ............................................................6
`
`Dish Network Corp. v. Customedia Tech., LLC, IPR2017-00936, Paper 13 (Aug.
`24, 2017) ..................................................................................................................................29
`
`Intelligent Bio-Systems, Inc. v. Illumina Cambridge, Ltd., 821 F.3d 1359 (Fed.
`Cir. 2016) ...........................................................................................................................10, 11
`
`In re Kahn, 441 F.3d 977 (Fed. Cir. 2006) ....................................................................................32
`
`KSR Int’l Co. v. Teleflex Inc., 550 U.S. 398 (2007).................................................................10, 39
`
`LG Display, Ltd. v. Innovative Display Technologies LLC., IPR2014-01094,
`Paper 10 (Jan. 13, 2015) ..........................................................................................................59
`
`PersonalWeb Techs., LLC v. Apple, Inc., 848 F.3d 987 (Fed. Cir. 2017) .........................51, 55, 57
`
`In re: Smith Int’l, Inc., 2017 U.S. App. LEXIS 18526 (Fed. Cir. Sept. 26, 2017) ..........................8
`
`Unique Concepts, Inc. v. Brown, 939 F.2d 1558 (Fed. Cir. 1991) ................................................29
`
`WMS Gaming, Inc. v. Int’l Game Tech., 184 F.3d 1339 (Fed. Cir. 1999) .....................................10
`
`STATUTES
`
`35 U.S.C. §311-319 ..........................................................................................................................1
`
`OTHER AUTHORITIES
`
`37 C.F.R. § 42.65(a).......................................................................................................................59
`
`37 C.F.R. § 42.120 ...........................................................................................................................1
`
`
`
`
`
`
`
`
`1
`
`
`
`

`

`PATENT OWNER’S TABLE OF EXHIBITS
`
`IPR2017-00626
`Patent 6,363,345
`
`
`Exhibit Description
`
`Reserved
` Declaration of Scott Douglas, Ph.D.
`Notice of Initial Determination on Violation of Section 337 from
`Inv. No. 337-TA-1026
`Reserved
`Bertrand Hochwald Deposition Transcript
`
`
`
`
`Exhibit No.
`2001
`2002
`2003
`2004
`2005
`
`
`
`
`2
`
`

`

`Pursuant to 37 C.F.R. § 42.120, Andrea Electronics Corporation (“Andrea”
`
`IPR2017-00626
`Patent 6,363,345
`
`
`or “Patent Owner”) hereby submits this Response to Petitioner, Apple, Inc.’s
`
`(“Petitioner” or “Apple”) Petition (the “’626 Petition”) seeking inter partes review
`
`of U.S. Patent No. 6,363,345 (“the ’345 Patent”), and to the grounds for which a
`
`trial was instituted pursuant to the Institution Decision dated July 24, 2017 (Paper
`
`7, “Institution Decision”). This filing is timely under 35 U.S.C. §311-319 and 37
`
`C.F.R. § 42.120 in accordance with the parties’ Joint Notice of Stipulation to
`
`Amend Scheduling Order (Paper 10).
`
`I.
`
`INTRODUCTION
`
`Related Proceedings
`
`A.
`As noted in the Petition, the ‘345 patent was and is subject to several other
`
`proceedings. Petition at viii-x. With the exception of the Investigation currently
`
`pending before the U.S. International Trade Commission, 337-TA-1026, and the
`
`co-pending inter partes review of the ’345 patent (IPR2017-000627),all of the
`
`other proceedings identified in the Petition have been terminated or stayed.
`
`In the ITC Investigation, following an evidentiary hearing, the
`
`Administrative Law Judge (“ALJ”) issued a Notice of Initial Determination and an
`
`Initial Determination on October 26, 2017. The Notice of Initial Determination,
`
`attached as Ex. 2003, is a public document and states that the ALJ found the
`
`relevant ‘345 patent claims not invalid, not unenforceable, and not infringed by
`
`3
`
`

`

`Apple. The full Initial Determination is a confidential document, but a public
`
`IPR2017-00626
`Patent 6,363,345
`
`
`version of the Initial Determination is expected to be available in the next few
`
`weeks. Patent Owner will seek to submit the public version of the Initial
`
`Determination in the instant proceeding as soon as it becomes available.
`
`Institution Decision
`
`B.
`The Board instituted this trial as to certain claims of the ’345 Patent on the
`
`following grounds:
`
`1.
`
`Claims 1-3, 12, 13, 21, 23, and 38 as being anticipated by an article by
`
`H. Hirsch et al., entitled “Noise estimation techniques for robust speech
`
`recognition” (Ex. 1005, “Hirsch”);
`
`2.
`
`Claims 4-11, 25, 39-42, and 46 as being obvious over Hirsch in view
`
`of an article by R. Martin, entitled “An efficient algorithm to estimate
`
`instantaneous SNR of speech signals” (Ex. 1006, “Martin”);
`
`3.
`
`Claims 13, 14, 17-21, 23, and 47 as being obvious over Hirsch in view
`
`of an article by S. Boll, entitled “Suppression of acoustic noise in speech using
`
`spectral subtraction” (Ex. 1009, “Boll”);
`
`4.
`
`5.
`
`Claim 43 as being obvious over Hirsch, Martin, and Boll;
`
`Claims 15, 16, and 24 as being obvious over Hirsch and Boll and
`
`further in view of U.S. Patent No. 5,706,395 of Arslan et al. (Ex. 1011, “Arslan”);
`
`4
`
`

`

`Claim 22 as being obvious over Hirsch in view of U.S. Patent No.
`
`IPR2017-00626
`Patent 6,363,345
`
`
`6.
`
`5,459,683 of Uesugi et al. (Ex. 1015, “Uesugi”); and
`
`7.
`
`Claims 44 and 45 as being obvious over Hirsch and Martin and further
`
`in view of Uesugi.
`
`As discussed in detail below, the combination of Hirsch and Martin fails to
`
`teach or disclose at least the “current minimum” and “future minimum” values
`
`recited in claims 4-11 and 39-42. Further, the combination of Hirsch and Martin
`
`also fails to disclose at least the “periodic” limitations of claims 6 and 9, and the
`
`“current minimum value [being] determined as the minimum value of the
`
`magnitude of the corresponding frequency bin within a predetermined period of
`
`time” limitation of claim 10.
`
`Additionally, as explained below, Apple has not sufficiently demonstrated
`
`that a skilled artisan would have been motivated to combine Hirsch with any of the
`
`relied-upon secondary references to render claims 4-11, 14-20, 22, 24-25, and 39-
`
`47 obvious.
`
`Accordingly, Petitioner has failed to demonstrate by a preponderance of the
`
`evidence that claims 4-11, 14-20, 22, 24-25, and 39-47 of the ’345 Patent are
`
`rendered obvious in view of the relevant prior art.
`
`5
`
`

`

`II. OVERVIEW OF THE ’345 PATENT
`The ’345 patent is generally directed to systems, methods, and apparatuses
`
`IPR2017-00626
`Patent 6,363,345
`
`
`for processing audio signals, and more specifically to the processing of audio
`
`signals so as to cancel noise components. See Ex. 1001 at 1:19-21; Ex. 2002 at ¶
`
`37.
`
`The ’345 patent breaks the signal into its constituent frequency components,
`
`called “frequency bins,” using a “frequency spectrum generator,” for example, an
`
`FFT. Ex. 1001 at 2:11-19. The claimed invention has the ability to detect and
`
`cancel noise on a frequency bin-by-frequency bin basis by setting an adaptive
`
`threshold for each frequency bin. Ex. 1001 at 3:24-45; Ex. 2002 at ¶ 38.
`
`Once the frequency bins are generated, the ’345 patent teaches that the
`
`sample values within each frequency bin can be optionally smoothed. Ex. 1001 at
`
`5: 45-58. After the smoothing process, there are still a collection of sample values
`
`in each bin. Ex. 2002 at ¶ 39.
`
`An adaptive threshold is set for each frequency bin based on the magnitude
`
`characteristics of the (optionally smoothed) samples within the bin. Ex. 1001 at
`
`6:10-22. The ’345 patent sets the threshold using a cascading minimum
`
`determination process. First, a “future minimum” is determined as the minimum
`
`magnitude of the frequency bin for a given period of time, for example, five
`
`seconds. Ex. 1001 at 6:23-32. After this five-second period elapses, the “future
`
`6
`
`

`

`minimum” value is used to initiate the “current minimum” parameter. Ex. 1001 at
`
`IPR2017-00626
`Patent 6,363,345
`
`
`6:33-41. The initiated “current minimum” is then compared to each sample in the
`
`frequency bin over the next five second period, and its value is updated whenever
`
`the magnitude of a sample in the frequency bin is less than the “current
`
`minimum’s” value. Id. Thus, the “current minimum” is ultimately determined as
`
`the smallest magnitude value within the frequency bin. After the five second
`
`period has elapsed, the “current minimum” value is then used to set the threshold.
`
`In the preferred embodiment, the minimum is scaled by a factor of 4, and this
`
`scaled minimum “4*Min” serves as the threshold. Ex. 1001 at 6:46-57; see also,
`
`id. at Fig 3; Ex. 2002 at ¶ 40.
`
`Once the threshold is set (e.g., 4*Min), the threshold detector compares the
`
`threshold to the magnitude of the signal. Id. Where the magnitude of the signal is
`
`less than the threshold, the noise estimate is updated using the magnitude value
`
`(“New Data”). See, e.g., id. at Fig. 3. The updated noise estimate is then used in
`
`the subtraction process to cancel noise from the signal. Ex. 1001 at 6:58-7:27; see
`
`also Ex. 2002 at ¶ 41.
`
`III. CLAIM CONSTRUCTION AND LEVEL OF ORDINARY SKILL IN
`THE ART
`In an inter partes review, claim terms in an unexpired patent are interpreted
`
`according to their broadest reasonable construction in light of the specification of
`
`7
`
`

`

`the patent in which they appear. Cuozzo Speech Techs., LLC v. Lee, 136 S. Ct.
`
`IPR2017-00626
`Patent 6,363,345
`
`
`2131, 2142 (2016).
`
`A Person Having Ordinary Skill In The Art
`
`A.
`Apple alleges that a hypothetical person of ordinary skill in the field of the
`
`’345 patent would have had “a good working knowledge of digital signal
`
`processing techniques and their applications” gained through “an undergraduate
`
`education in electrical engineering or a comparable field, in combination with
`
`either a graduate degree (or two years of graduate work) in electrical engineering
`
`or a comparable field, or through two years of practical work experience, where
`
`such graduate education or work experience focused on or involved the use of
`
`digital signal processing techniques.” Petition at 12.
`
`For the purposes of this proceeding, Patent Owner applies Apple’s proposed
`
`standard without prejudice, but reserves its rights to present an alternative
`
`definition in any other proceeding.
`
`Claim Construction
`
`B.
`In its Petition, Apple proposed constructions for the following claim terms:
`
`“magnitude” (independent claims 1 and 38); “frequency spectrum generator”/
`
`“generating the frequency spectrum” (independent claims 1 and 38); “threshold
`
`detector for setting a threshold … and for detecting” (independent claim 1); and
`
`“generating a noise canceling signal for canceling noise” (independent claim 38).
`
`8
`
`

`

`In the Institution Decision, the Board determined that no term of the ‘345 patent
`
`IPR2017-00626
`Patent 6,363,345
`
`
`claims requires an express construction. ’626 Institution Decision at p. 5. For
`
`purposes of this proceeding, Andrea applies Apple’s proposed constructions
`
`without prejudice, but reserves its rights to present evidence and arguments as to
`
`the proper construction of any claim terms within the meaning of the ’345 Patent in
`
`any other proceeding.
`
`Additionally, Patent Owner proposes a construction for the claim term
`
`“periodically,” as described below.
`
`In the ‘626 Petition, Apple did not propose a specific claim construction for
`
`the term “periodically.” See Petition at 13-16.
`
`In Apple’s other Petition directed to the ’345 patent (IPR2017-00627),
`
`Apple relied on two dictionary definitions of “periodically.” Specifically, Apple
`
`relied on a 1993 edition of Webster’s Third New International Dictionary that
`
`defined “periodically” as: “1: at regular intervals of time; 2: from time to time.”
`
`(See 627 Petition at 35, Ex. 1013). Apple appears to take the position that either of
`
`these two definitions apply to the term periodically in the context of the ‘345
`
`patent. See 627 Petition at 36 (“Helf also shows the ‘future minimum’ is replaced
`
`when it is greater than the current magnitude, and thus is replaced ‘periodically.’”);
`
`see also Ex. 1013 (defining “periodically” as “1: at regular intervals of time; 2:
`
`from time to time” (emphasis in original)).
`
`9
`
`

`

`Patent Owner asserts that the term “periodically” should be construed as
`
`IPR2017-00626
`Patent 6,363,345
`
`
`“occurring at regular intervals of time.” In contrast to Apple’s position,
`
`“periodically” should not be construed to also encompass something that occurs
`
`from “time to time.” The Federal Circuit recently elaborated on the inquiry
`
`concerning the determination of the broadest reasonable construction of a claim
`
`term:
`
`The correct inquiry in giving a claim term its broadest reasonable
`interpretation in light of the specification is not whether the
`specification proscribes or precludes some broad reading of the claim
`term adopted by the examiner. And it is not simply an interpretation
`that is not inconsistent with the specification. It is an interpretation
`that corresponds with what and how the inventor describes his
`invention in the specification, i.e., an interpretation that is “consistent
`with the specification.”
`
`In re: Smith Int’l, Inc., 2017 U.S. App. LEXIS 18526, *12-13 (Fed. Cir. Sept. 26,
`
`2017) (internal quotations omitted).
`
`Patent Owner’s proposed construction adopts the first definition provided in
`
`the dictionary relied upon by Apple and this construction is consistent with the
`
`’345 patent specification’s only discussion of “periodically.” The ’345 patent
`
`describes how the invention updates (or initiates or sets) the current minimum
`
`value with the future minimum value at regular intervals, for example, every 5
`
`seconds. See Ex. 1001, 6:23-41; 8:36-40. This is precisely how the term
`
`10
`
`

`

`“periodically” is used in claim 6. In the only place that the specification uses the
`
`IPR2017-00626
`Patent 6,363,345
`
`
`term periodically, it states that “[t]he future and current minimum values are
`
`calculated continuously and initiated periodically, for example, every 5 seconds as
`
`determined in step 724 and control is advanced to steps 722 and 726 wherein the
`
`new future and current minimum are calculated.” Ex. 1001 8:36-40.
`
`Accordingly, the ’345 patent’s discussion of “periodically” supports, and is
`
`consistent with, “periodically” being construed to mean “occurring at regular
`
`intervals of time.” In contrast, the ’345 patent specification provides no disclosure
`
`that would support, or is consistent with, the overly-broad construction utilized by
`
`Apple that would define periodically to also include something that simply
`
`happens “from time to time.”
`
`Apple’s own expert confirmed at his deposition that outside of the context of
`
`the ’345 patent a skilled artisan would understand that the term “periodically” to
`
`mean at regular time intervals and that that same definition would apply in the
`
`context of the ’345 patent. See Ex. 2005 at 99:9-100:7. Given the evidence of
`
`record, the term “periodically” should not be construed to further include the
`
`alternative, overly-broad definition “from time to time.”
`
`
`
`11
`
`

`

`IV. CLAIMS 4-20, 22, 24-25, AND 39-47ARE NOT OBVIOUS OVER
`HIRSCH IN VIEW OF ANY SECONDARY REFERENCE
`The Board has instituted this proceeding on the grounds that Hirsch
`
`IPR2017-00626
`Patent 6,363,345
`
`
`anticipates claims 1-3, 12-13, 21, 23, and 38 and grounds that claims 4-11, 13-25,
`
`38-42, and 43-47 are obvious over Hirsch in combination with one or more of
`
`Martin, Boll, Arslan, and Uesugi. In this Response, Patent Owner does not address
`
`the anticipation grounds with respect to the Hirsch reference.
`
`“To establish obviousness of a claimed invention, all the claim limitations
`
`must be taught or suggested by the prior art.” CFMT, Inc. v. Yieldup Int’l Corp.,
`
`349 F.3d 1333 (Fed. Cir. 2003). “[A] patent claim composed of several elements,
`
`however, is not proved obvious merely by demonstrating that each of its elements
`
`was known, independently, in the prior art.” KSR Int’l Co. v. Teleflex Inc., 550
`
`U.S. 398, *419 (2007).
`
`“When an obviousness determination relies on the combination of two or
`
`more references, there must be some suggestion or motivation to combine the
`
`references.” Intelligent Bio-Systems, Inc. v. Illumina Cambridge, Ltd., 821 F.3d
`
`1359, 1367-8 (Fed. Cir. 2016) (quoting WMS Gaming, Inc. v. Int’l Game Tech.,
`
`184 F.3d 1339, 1355 (Fed. Cir. 1999). Thus, it is “[petitioner]’s burden to
`
`demonstrate both that a skilled artisan would have been motivated to combine the
`
`teachings of the prior art references to achieve the claimed invention, and that the
`
`skilled artisan would have had a reasonable expectation of success in doing so.”
`
`12
`
`

`

`Intelligent Bio-Systems, 821 F.3d 1359, 1363-64 (Fed. Cir. 2016) (internal
`
`IPR2017-00626
`Patent 6,363,345
`
`
`quotations removed).
`
`Apple fails to establish that the instituted claims are rendered obvious
`
`because none of the relied upon prior art, alone or in combination, discloses the
`
`“current minimum” and “future minimum” values recited in claims 4-11 and 39-
`
`42. Further, Apple fails to establish that the relied upon prior art discloses
`
`updating the “current minimum” and “future minimum” values “periodically” as
`
`recited by claims 6 and 9, or the “minimum. . . magnitude” value recited in claim
`
`10.
`
`Finally, challenged claims 4-20, 22, 24-25, and 39-47 are not obvious
`
`because Apple has failed to establish a motivation to combine the Hirsch with the
`
`relied-upon secondary references that is supported by rational underpinnings.
`
`A. Grounds Based on the Combination of Hirsch and Martin
`As noted above, Apple alleges that claims 4-11, 25, 39-42, and 46 are
`
`rendered obvious by Hirsch in view of Martin. As discussed below, Petitioner fails
`
`to demonstrate by a preponderance of the evidence that claims 4-11 and 39-42 are
`
`unpatentable as alleged.
`
`Summary of Hirsch
`
`1.
`Hirsch is directed to two techniques to estimate noise characteristics for
`
`noisy speech signals. Ex. 1005 at 153, Abstract. These techniques were designed
`
`13
`
`

`

`IPR2017-00626
`Patent 6,363,345
`
`to separate unwanted background noise from a speech signal, enhancing the speech
`
`signal and improving the performance of speech recognition systems.
`
`The first technique utilizes an algorithm using a first order recursive system
`
`that calculates the noise magnitude level in signal subbands using a weighted
`
`average of past spectral magnitudes. Ex. 1005 at 153. The first technique utilizes
`
`adaptive thresholds to calculate the noise magnitude level. Ex. 1005 at 153. The
`
`algorithm estimates the noise magnitude level by taking a weighted sum of past
`
`spectral magnitude values in each subband and multiplying that sum by an
`
`overestimation factor to derive a threshold. The noise is estimated when a spectral
`
`component first exceeds the noise threshold previously calculated, thus stopping
`
`the recursive accumulation of past spectral magnitude values, resulting in an
`
`estimate of the magnitude level of noise. Ex. 1005 at 153. When a spectral value
`
`is less than the noise threshold, speech is not detected, and said values are set to
`
`zero. Ex. 1005 at 153.
`
`The second technique disclosed in Hirsch employs an algorithm that
`
`evaluates the histograms of past spectral magnitude values corresponding to noise
`
`segments in signal subbands, taking the maximum as a noise magnitude level
`
`estimation. Ex. 1005 at 154. The noise threshold calculated in accordance with
`
`the first technique described above is used to evaluate past spectral values that fall
`
`below said threshold. Ex. 1005 at 154. Past values identified as noise segments
`
`14
`
`

`

`are evaluated to determine the noise distribution in roughly forty frequency bins.
`
`IPR2017-00626
`Patent 6,363,345
`
`
`Ex. 1005 at 154. The maximum of the noise distribution in each subband is used
`
`to estimate the noise magnitude level. Ex. 1005 at 154. These estimated noise
`
`magnitude values are smoothed over time. Ex. 1005 at 154.
`
`Summary of Martin
`
`2.
`Martin is directed to an algorithm for estimating the instantaneous signal-to-
`
`noise-ratio (“SNR”) of a noisy speech signal. Ex. 1006 at 1093, Abstract.
`
`Accordingly, the Martin algorithm uses its noise estimation process to compute
`
`SNR values for a given signal, and does not set an adaptive threshold for canceling
`
`noise. Ex. 1006 at 1094.
`
`Martin teaches acquiring noise statistics by tracking varying noise power
`
`levels during speech activity. Ex. 1006 at 1093. Martin achieves its stated
`
`improvement by distinguishing between conditions where a signal’s power is not
`
`monotonically increasing and conditions where a signal’s power is monotonically
`
`increasing.1 Ex. 1006 at 1094. Martin determines whether a signal is
`
`
`1 Martin’s algorithm includes a decision block where a determination is
`
`made as to whether the signal’s power is monotonically increasing over a window
`
`of length L. Ex. 1006 at Fig. 2. Martin’s description of this determination is also
`
`refers to the case of a monotonically increasing signal as “rapidly varying noise
`
`
`
`15
`
`

`

`monotonically increasing by observing subwindow minimum values across a
`
`IPR2017-00626
`Patent 6,363,345
`
`
`window of length L. Id. The following annotated Figure 2 from Martin shows the
`
`monotonically increasing power decision block (annotated in orange):
`
`Ex. 1006 at Fig. 2 (annotated). A skilled artisan would recognize that Martin’s
`
`system aims to calculate SNR values despite a signal’s unpredictable behavior
`
`
`
`
`power” and the case of a non-monotonically increasing signal as “slowly varying
`
`noise power.” Ex. 1006 at 1094.
`
`16
`
`

`

`while minimizing the delay in the computation. Thus, depending on signal
`
`IPR2017-00626
`Patent 6,363,345
`
`
`conditions, Martin’s algorithm estimates noise in different ways.
`
`If the signal is not monotonically increasing, i.e., in “slowly varying noise
`
`power” conditions, Martin’s algorithm estimates the noise floor, or minimum noise
`
`power, Pn(i), in three steps. First, a given data window of length L samples is
`
`divided into subwindows of length M samples. Second, the minimum noise power
`
`in each subwindow is found and stored in an array, called the “min_vec” array.
`
`Third, the noise estimate is derived by determining the minimum of the
`
`subwindow minima values stored in the min_vec array, which Martin designates
`
`“Pn(i)” (annotated in purple). Ex. 1006 at 1094. The values of L, W and M that
`
`are chosen will affect the performance of Martin’s algorithm. See Ex. 1006 at
`
`1094; see also, Ex. 2002 at ¶ 50.
`
`Where the signal is monotonically increasing, i.e., in “rapidly varying noise
`
`power conditions,” the noise power estimate, Pn(i), is determined through a
`
`different process. In this case, the noise power estimate Pn(i) is taken as the power
`
`of the most recent subwindow minimum value, PMmin (annotated in red). Ex. 1006
`
`at 1094. The following example shows how Martin uses the most recent
`
`subwindow minimum in the monotonically increasing case:
`
`17
`
`

`

`IPR2017-00626
`Patent 6,363,345
`
`
`
`
`As shown in the example above, the most recent subwindow minimum in the
`
`monotonically increasing case is also always the largest subwindow minimum.
`
`Thus, the rapidly varying noise power estimate (the largest subwindow minimum)
`
`is always different than the slowly varying noise power estimate (which designates
`
`Pn(i) as the smallest subwindow minimum value). See Ex. 2002 at ¶ 51-52. A
`
`determination of whether a signal’s power is monotonically increasing over
`
`window length L requires that at least two subwindows exist, since the monotonic
`
`decision block compares the minimum power in each subwindow to make its
`
`determination. Ex. 1006 at 1094.
`
`Independent of these two methods, the Martin algorithm also uses an
`
`instantaneous noise estimate update (annotated in green in the figure below):
`
`18
`
`

`

`IPR2017-00626
`Patent 6,363,345
`
`
`
`
`Ex. 1006 at 1094 (annotated). The noise estimate Pn(i) is continuously
`
`compared with the actual smoothed power Px(i) for every sample, and Pn(i) is
`
`instantaneously updated with Px(i) if the actual smoothed power, Px(i), is smaller
`
`than the estimated noise power. Ex. 1006 at 1094; see also Ex. 2002 at ¶ 54.
`
`a. Martin’s Techniques Allegedly Reduce Delay
`Martin’s noise estimation process allows the Martin system to obtain
`
`allegedly more accurate noise estimates despite the unpredictable behavior of a
`
`given signal. Martin touts his algorithm’s ability to efficiently perform these noise
`
`estimations with minimal delay. See Ex. 1006 at 1094; see also, Ex. 2002 at ¶ 55.
`
`19
`
`

`

`Martin’s monotonic decision block allows the algorithm to adaptively update its
`
`IPR2017-00626
`Patent 6,363,345
`
`
`noise estimate based on signal characteristics (i.e., slowly varying vs. rapidly
`
`varying noise power). See Ex. 1006 at 1094; see also, Ex. 2002 at ¶ 55.
`
`As noted, the monotonic decision block, annotated in orange above,
`
`determines whether the signal’s power is monotonically increasing by comparing
`
`the subwindow minimum values stored in the min_vec array. Ex. 1006 at 1094
`
`(“If the minimum power of the last W windows with M samples each is
`
`monotonically increasing we decide on rapid noise power variation.”); see also Ex.
`
`2002 at ¶ 56. To determine whether the signal’s power is monotonically increasing
`
`over a window of length L, at least two subwindow minimum values must be
`
`available for the comparison, and therefore at least two subwindows must be used
`
`(i.e., “W” must be at least 2). See Ex. 2002 at ¶ 56.
`
`Martin clarifies that the monotonic decision block is not invoked until an
`
`entire window length of L samples has been read:
`
`In case of slowly varying noise power the update of noise estimates is
`delayed by L + M samples. If a rapid noise power increase is detected
`this delay is reduced to M samples, thus improving the noise tracking
`capability of the algorithm.
`
`Ex. 1006 at 1094.
`
`Where the signal over length L is not monotonically increasing, the
`
`minimum value Pn(i) could be any sample, including the very first sample in the L
`
`20
`
`

`

`window. In such a scenario, the noise estimate would not be updated until L + M
`
`IPR2017-00626
`Patent 6,363,345
`
`
`samples have been read, and therefore the algorithm would include a “delay” o