IN THE UNITED STATES DISTRICT COURT
`WESTERN DISTRICT OF TEXAS
`WACO DIVISION
`
`
`
`Plaintiff,
`
`C.A. No. 6:22-cv-00123-ADA
`
`JURY TRIAL DEMAND
`
`ZENTIAN LTD.,
`
`
`
`
`v.
`
`
`AMAZON.COM SERVICES LLC and
`AMAZON WEB SERVICES, INC.,
`
`
`Defendants.
`
`
`
`
`DEFENDANTS’ FIRST AMENDED INVALIDITY CONTENTIONS
`
`I.
`
`INTRODUCTION
`
`Pursuant to the Court’s Scheduling Order (Dkt. 33) and Order Governing Proceedings,
`
`defendants Amazon.com, Inc., Amazon Web Services, Inc., and Amazon.com Services LLC
`
`(collectively “Amazon” or “Defendants”) provide these amended invalidity contentions to plaintiff
`
`Zentian Ltd. (“Zentian” or “Plaintiff”), regarding the currently asserted claims of U.S. Patent Nos.
`
`7,979,277 (“the ’277 patent”), 7,587,319 (“the ’319 patent”), 10,062,377 (“the ’377 patent”),
`
`10,839,789 (“the ’789 patent”), and 10,971,140 (“the ’140 patent”) (collectively the “Asserted
`
`Patents”). Defendants reserve the right to provide supplemental contentions in view of additional
`
`information or documents received from third parties or that Defendants locate from ongoing
`
`investigations.
`
`Zentian asserts the claims identified in the table below (collectively, the “Asserted Claims”).
`
`Patent
`U.S. Patent No. 7,979,277
`U.S. Patent No. 10,062,377
`
`Asserted Claims
`1, 4, 5, 7, 9, 10, 12, 14-16
`1-6
`
`
`
`
`
`

`

`Patent
`U.S. Patent No. 10,839,789
`
`U.S. Patent No. 7,587,319
`U.S. Patent No. 10,971,140
`
`Asserted Claims
`1, 2, 4, 6-14, 16-18, 20-29, 32,
`34, 35, 37-39, 44, 45
`46-51, 53, 54, 58, 64, 65, 67
`1-6
`
`Defendants are providing invalidity contentions for the above-identified Asserted Claims.
`
`To the extent Zentian later attempts to assert additional claims from other patents, Defendants will
`
`amend and supplement these invalidity contentions to address any additional claims. Nothing in
`
`these contentions constitutes an admission of validity as to any claims of the Asserted Patents.
`
`Pursuant to the Court’s Scheduling Order (Dkt. 33) and based on Defendants’ investigation
`
`and knowledge developed to date, Defendants serve invalidity contentions in the form of (1) a
`
`chart setting forth where in the prior art references each element of the Asserted Claims is found,
`
`and (2) an identification of any limitations Defendants contend are indefinite or lack written
`
`description under 35 U.S.C. § 112. Defendants reserve the right to show that claims are directed
`
`to ineligible subject matter under 35 U.S.C. § 101. Consistent with the requirements of the
`
`Scheduling Order, these contentions do not include the bases for any unenforceability arguments
`
`Defendants may advance, or any
`
`invalidity positions other
`
`than
`
`those based on
`
`anticipation/obviousness in view of prior art or based on 35 U.S.C. § 112. 1 Defendants are
`
`concurrently producing the prior art referenced herein.
`
`Defendants incorporate by reference all prior art references, charts, theories, and
`
`disclosures served on Zentian in any prior or pending court action or proceeding before the Patent
`
`Trial and Appeal Board, including but not limited to, Zentian Ltd. v. Apple Inc., No. 6:21-cv-815-
`
`ADA (W.D. Tex.).
`
`
`1 References to 35 U.S.C. refer to the pre-AIA version of the statute.
`
`2
`
`

`

`Should Zentian be permitted to assert more or different claims from the Asserted Patents, or
`
`any other patents, than it does now, Defendants will identify additional/alternative invalidity grounds
`
`and prior art as appropriate and will amend or supplement these invalidity contentions accordingly.
`
`Defendants’ invalidity contentions are based on information reasonably available at this time
`
`and may require subsequent amendment, modification, or supplementation. Full fact discovery has
`
`not yet opened, and Defendants have not obtained third-party discovery, including discovery related
`
`to system prior art. Defendants also reserve the right to present additional knowledge and/or
`
`information regarding prior art located during the course of discovery or further investigation, for
`
`example from consultation with experts, prior art authors, and others. Defendants have not yet
`
`obtained deposition testimony from any named inventor of the Asserted Patents. Defendants expect
`
`that discovery will reveal not only additional prior art, but also related disclosures and evidence for
`
`many of the prior art references identified herein. As such, Defendants have not yet completed their
`
`investigation, discovery, or analysis of matters relating to the validity or enforceability of the
`
`Asserted Claims. Accordingly, nothing in these invalidity contentions should be construed as a
`
`statement that no other persons have discoverable information, that no other documents, data
`
`compilations, and/or tangible things exist that Defendants may use to support their claims or defenses,
`
`or that no other legal theories or factual bases will be pursued. Defendants reserve the right to amend,
`
`modify and/or supplement these invalidity contentions as additional information is discovered,
`
`identified or otherwise appreciated, including testimony about the scope and content of the prior art
`
`and the claimed inventions.
`
`Further, these invalidity contentions are based on Defendants’ present understanding of
`
`Zentian’s Preliminary Infringement Contentions served on September 1, 2022 (“preliminary
`
`infringement contentions”). Nothing in these invalidity contentions should be thought to
`
`concede that Zentian’s preliminary infringement contentions are legally or factually adequate or
`
`3
`
`

`

`as necessarily reflecting the proper interpretation of the claims or an interpretation of the claims
`
`that Defendants agree with or propose. Indeed, Zentian’s preliminary infringement contentions
`
`are largely incoherent and lack the detail required by the Court’s rules. Defendants reserve the
`
`right to amend these contentions should Zentian raise any new infringement theories in its final
`
`infringement contentions or provide any new applications and interpretation of the scope of the
`
`claims in those final infringement contentions.
`
`Nothing in these invalidity contentions should be treated as an admission that any of
`
`Defendants’ accused technology meets any limitation of the Asserted Claims. Defendants do not
`
`infringe any of the Asserted Claims. To the extent that any prior art reference identified by
`
`Defendants contains a claim element that is the same as or similar to an element in Defendants’
`
`accused technology, inclusion of that reference in Defendants’ invalidity contentions shall not be
`
`deemed a waiver of any claim construction or non-infringement position. Any use of these
`
`invalidity contentions to support any allegation of infringement would be misleading, false, and
`
`wrong as a matter of law and fact.
`
`Further, nothing in these contentions constitutes an admission concerning the priority dates
`
`of the Asserted Claims that Zentian asserts. These invalidity contentions, unless otherwise
`
`specified, are based on the following priority dates asserted by Zentian: (1) September 14, 2004
`
`for the ’277, ’377, and ’789 patents, and (2) February 4, 2002 for the ’319 and ’140 patents.
`
`II.
`
`IDENTIFICATION OF PRIOR ART
`
`The concepts disclosed and claimed in each Asserted Patent are not new, and had been
`
`disclosed, used, offered for sale, sold, and practiced by others before the claimed priority date of
`
`the patents. The prior art identified herein and in Exhibits A-F, individually or in combination,
`
`invalidates the Asserted Claims under 35 U.S.C. §§ 102 (a), (b), (e), (g) and 103. Because full fact
`
`discovery has yet to open, Defendants expect to gather more information about the identified prior
`
`4
`
`

`

`art, and other prior art, through third party discovery or other discovery, and may thus amend and
`
`supplement these invalidity contentions once they obtain that discovery and have reasonable time
`
`to analyze it in a meaningful way.
`
`As described in more detail below, claims of the Asserted Patents are also invalid under 35
`
`U.S.C. § 112.
`
`A.
`
`THE ’277 PATENT
`
`1.
`
`Identification of Invalidity Grounds
`
`Based on the reasons identified herein and the accompanying charts and references, a
`
`person having ordinary skill in the art would have found the Asserted Claims of the ’277 patent
`
`invalid under 35 U.S.C. §§ 102 and/or 103. The ’277 patent invalidity charts attached as Exhibit
`
`A set forth herein, for corresponding limitations of each asserted claim, exemplary disclosures
`
`from each identified reference in an invalidity ground below where applicable. Defendants reserve
`
`the right to amend their invalidity grounds or charts, or provide supplemental grounds or charts,
`
`based on any amendments to Plaintiff’s infringement contentions, or information or documents
`
`that may be obtained as discovery begins including third party discovery.
`
`Defendants provide the following identification of exemplary invalidity grounds for the
`
`currently Asserted Claims of the ’277 patent:
`
`
`
`
`
`
`
`5
`
`

`

`
`
`Claim No. Exemplary Invalidity Grounds2, 3 Under 35 U.S.C. §§ 102 and/or 1034
`1
`Mathew alone or in combination with Tran
`Mathew alone or in combination with Bailey
`Mathew alone or in combination with Lund
`TMS320C alone or in combination with Tran
`TMS320C alone or in combination with Bailey
`TMS320C alone or in combination with Mathew
`TMS320C alone or in combination with Melnikoff
`TMS320C alone or in combination with Melnikoff II
`TMS320C alone or in combination with Stogiannos
`TMS320C alone or in combination with Lund
`Bailey alone or in combination with Tran
`Bailey alone or in combination with Mathew
`Bailey alone or in combination with Melnikoff
`Bailey alone or in combination with Melnikoff II
`Bailey alone or in combination with Stogiannos
`Bailey alone or in combination with Lund
`Stogiannos alone or in combination with Tran
`Stogiannos alone or in combination with Bailey
`Stogiannos alone or in combination with TMS320C
`Stogiannos alone or in combination with Lund
`Melnikoff alone or in combination with Tran
`
`
`2 For brevity, each reference in this table is identified by its short cite, which is defined later in
`subsections II.A.2 and II.A.3 below.
`3 To the extent the grounds identified in the below table for a particular asserted claim refer back
`to the grounds identified for another claim, the grounds disclose the particular asserted claim for
`the reasons identified in the corresponding claim charts for the particular asserted claim.
`4 Defendants contend that any combination of references identified in this table as a ground
`invalidates the relevant asserted claim under 35 U.S.C. § 103. In addition, each reference identified
`by itself as a ground invalidates the relevant asserted claim under 35 U.S.C. §§ 102 and 103, as “it
`is well settled that a disclosure that anticipates under § 102 also renders the claim invalid under
`§ 103, for anticipation is the epitome of obviousness.” Realtime Data, LLC v. Iancu, 912 F.3d
`1368, 1373 (Fed. Cir. 2019) (citations and quotations omitted) (finding no error in § 103 invalidity
`determination based on a single reference).
`
`6
`
`

`

`Claim No. Exemplary Invalidity Grounds2, 3 Under 35 U.S.C. §§ 102 and/or 1034
`Melnikoff alone or in combination with Bailey
`Melnikoff alone or in combination with TMS320C
`Melnikoff alone or in combination with Lund
`Melnikoff II alone or in combination with Tran
`Melnikoff II alone or in combination with Bailey
`Melnikoff II alone or in combination with TMS320C
`Melnikoff II alone or in combination with Lund
`Melnikoff II alone or in combination with Nguyen
`Tran in combination with Bailey
`Tran in combination with Mathew
`Tran in combination with Melnikoff
`Tran in combination with Melnikoff II
`Tran in combination with Stogiannos
`Stolzle in combination with Tran
`Stolzle in combination with Bailey
`Stolzle in combination with TMS320C
`Stolzle in combination with Lund
`See grounds of claim 1.
`See grounds of claim 1.
`See grounds of claim 1.
`See grounds of claim 1.
`See grounds of claim 1.
`See grounds of claim 1.
`See grounds of claim 1.
`See grounds of claim 1.
`See grounds of claim 1.
`
`4
`5
`7
`9
`10
`12
`14
`15
`16
`
`2.
`
`Identification of Invalidity Due to Anticipation
`
`The table below lists the prior art references that anticipate one or more of the Asserted
`
`Claims of the ’277 patent under 35 U.S.C. § 102. The attached claim charts in Exhibit A demonstrate
`
`7
`
`

`

`where each limitation of the Asserted Claims is found in the references listed below, either expressly
`
`or inherently in the larger context of the passage, as understood by a person having ordinary skill in
`
`the art. The following references are prior art under at least 35 U.S.C. §§ 102(a), (b), or (e).
`
`a.
`
`Prior Art Patents and Patent Publications that Anticipate the
`Asserted Claims of the ’277 Patent
`
`Patent No.
`
`Filing Date /
`Priority Date
`U.S. Patent No. 5,459,798 March 19, 1993
`
`Date of Issue or
`Publication
`October 17, 1995
`
`Short Cite
`
`Bailey
`
`b.
`
`Prior Art Publications that Anticipate the Asserted Claims of
`the ’277 Patent
`
`Date of
`Publication
`2003
`
`Author/Publisher
`
`Short Cite
`
`Binu Mathew et al.
`
`Mathew
`
`1994
`
`Texas Instruments
`
`TMS320C
`
`2000
`
`2002
`
`Panagiotis Stogiannos
`et al.
`
`Stogiannos
`
`Stephen Melnikoff et
`al.
`
`Melnikoff
`
`January 12, 2002 Stephen Melnikoff et
`al.
`
`Melnikoff II
`
`Prior Art Publications
`Title
`A Low-Power Accelerator
`for the SPHINX 3 Speech
`Recognition System
`TMS320C5x DSPs -
`Telecommunications
`Applications with the
`TMS320C5x DSPs
`A Configurable Logic
`Based Architecture for
`Real-Time Continuous
`Speech Recognition Using
`Hidden Markov Models
`Speech Recognition on an
`FPGA Using Discrete and
`Continuous Hidden
`Markov Models
`Performing Speech
`Recognition on Multiple
`Parallel Files Using
`Continuous Hidden
`Markov Models on an
`FPGA
`
`8
`
`

`

`3.
`
`Identification of Invalidity Due to Obviousness
`
`The tables below list prior art references that render one or more of the Asserted Claims of
`
`the ’277 patent invalid as obvious under 35 U.S.C. § 103, alone or in combination with the
`
`knowledge of a person having ordinary skill in the art and/or other prior art. The attached claim
`
`charts in Exhibit A demonstrate where each limitation of the Asserted Claims is found in the
`
`references listed below, either expressly or inherently in the larger context of the passage, as
`
`understood by a person having ordinary skill in the art. The following references are prior art
`
`under at least 35 U.S.C. §§ 102(a), (b), or (e).
`
`a.
`
`Prior Art Patents and Patent Publications that Render
`Obvious the Asserted Claims of the ’277 Patent
`
`Date of Issue or
`Publication
`October 17, 1995
`December 3, 2002
`October 23, 2003
`
`November 4, 1998
`April 1, 1999
`
`June 27, 2000
`October 12, 2000
`
`Filing Date /
`Patent or Patent
`Priority Date
`Publication No.
`U.S. Patent No. 5,459,798 March 19, 1993
`U.S. Patent No. 6,490,559
`October 10, 1997
`U.S. Patent Publication No.
`April 22, 2002
`2003/0200085
`U.S. Patent No. 6,080,140
`International Patent
`Application Publication No.
`WO 2000060577
`Chinese Patent No. 1503220 November 20, 2002 May 11, 2005
`U.S. Publication No.
`August 2, 2001
`April 25, 2002
`2002/0049582
`February 20, 1998
`U.S. Patent No. 6,374,219
`April 22, 2002
`U.S. Patent No. 6,879,954
`U.S. Patent No. 5,819,222 March 31, 1994
`U.S. Patent No. 5,699,456
`January 21, 1994
`
`April 16, 2002
`April 12, 2005
`October 6, 1998
`December 16, 1997
`
`9
`
`Short Cite
`
`Bailey
`Budde
`Nguyen
`
`Tran
`Lund
`
`Jiang
`Baumgartner
`
`Jiang II
`Nguyen II
`Smyth
`Brown
`
`

`

`b.
`
`Prior Art Publications that Render Obvious the Asserted
`Claims of the ’277 Patent
`
`Title
`
`Integrated Circuits for a Real-Time
`Large-Vocabulary Continuous
`Speech Recognition System
`A Low-Power Accelerator for the
`SPHINX 3 Speech Recognition
`System
`A Low-Power VLSI Design of a
`HMM Based Speech Recognition
`System
`A High‐Speed HMM VLSI
`Module with Block Parallel
`Processing
`Speech Recognition on an FPGA
`Using Discrete and Continuous
`Hidden Markov Models
`Architectural Optimizations for
`Low-Power, Real-Time Speech
`Recognition
`Performing Speech Recognition on
`Multiple Parallel Files Using
`Continuous Hidden Markov
`Models on an FPGA
`A Configurable Logic Based
`Architecture for Real Time
`Continuous Speech Recognition
`Using Hidden Markov Models
`A VLSI Implementation of Pdf
`Computations in HMM Based
`Speech Recognition
`TMS320C5x DSPs -
`Telecommunications Applications
`with the TMS320C5x DSPs
`Efficient Algorithms for Speech
`Recognition
`
`Date of
`Publication
`January
`1991
`
`Author/Publisher
`
`Short Cite
`
`Anton Stolzle et al.
`
`Stolzle
`
`2003
`
`Binu Mathew et al. Mathew
`
`2002
`
`S. Yoshizawa et al.
`
`Yoshizawa
`
`December
`2002
`
`Shingo Yoshizawa et
`al.
`
`Yoshizawa II
`
`2002
`
`Stephen Melnikoff et
`al.
`
`Melnikoff
`
`2003
`
`Rajeev Krishna et al. Krishna
`
`January 12,
`2002
`
`Stephen Melnikoff et
`al.
`
`Melnikoff II
`
`2000
`
`Panagiotis
`Stogiannos et al.
`
`Stogiannos
`
`1996
`
`Johnny Pihl et al.
`
`Pihl
`
`1994
`
`Texas Instruments
`
`TMS320C
`
`May 15,
`1996
`
`Mosur K.
`Ravishankar
`
`Ravishankar
`
`10
`
`

`

`Title
`
`A Low-Power VLSI Design of an
`HMM Based Speech Recognition
`System
`
`Date of
`Publication
`2002
`
`Author/Publisher
`
`Short Cite
`
`S. Yoshizawa, et all. Yoshizawa
`
`c.
`
`Prior Art Systems/Services that Render Obvious the Asserted
`Claims of the ’277 Patent
`
`System/Service
`
`TMS320C5x DSPs -
`Telecommunications
`Applications with the
`TMS320C5x DSPs
`Sphinx II
`
`Relevant Dates Persons/Entities Involved
`in Prior Use, Sale, and/or
`Offers for Sale
`At least by 1994 Texas Instruments
`
`Short Cite
`
`TMS320C
`
`At least by 1996 The Sphinx Group at
`Carnegie Mellon
`University
`
`Sphinx II
`
`d.
`
`Motivations for Combining Prior Art
`
`The Asserted Patents are invalid in view of the combinations of references disclosed above.
`
`A person having ordinary skill in the art (“POSITA”) at the effective filing date of the Asserted
`
`Patents would have naturally consulted each of these references and combined the teachings, for at
`
`least the reasons provided below. Accordingly, a POSITA would have been motivated to combine
`
`elements of the references identified above, recognizing that the combination would be a predictable
`
`use of elements known in the art to solve known problems by using known techniques, resulting in
`
`expected outcomes. Defendants’ expert(s) may elaborate on the motivation to combine prior art and
`
`why the asserted claims of the asserted patents are invalid as obvious in accordance with the case
`
`schedule. Nothing herein is intended to limit or preclude such future expert opinion.
`
`11
`
`

`

`A POSITA would have understood the references listed above, alone or in combination, to
`
`contain explicit and/or implicit teaching, suggestion, and/or rationales to combine them for at least
`
`the following exemplary reasons.5
`
`(1)
`
`Speech recognition systems
`
`A POSITA would have known that “speech recognition systems,” were known in the art,
`
`as the specification admits, and that “prior art schemes have been proposed to increase the
`
`efficiency” of them. ’277 patent at 1:36-38; see also id. at 1:39-2:6 (identifying speech recognition
`
`prior art). The specification explains that prior art speech recognition systems typically used one
`
`of “a wide variety of probability distributions that can be used for the distance calculation stage”
`
`and a “lexical tree . . . because it greatly reduces the amount of storage space required to store a
`
`word vocabulary, by avoiding duplication of common word beginnings.” Id. at 3:57-60, 10:38-40.
`
`The patent therefore admits that speech recognition systems were known in the prior art.
`
`The prior art identified in these invalidity contentions describe such speech recognition
`
`systems. For example, Bailey describes a “high performance real-time pattern recognition within
`
`a general purpose computer system that may be utilized for . . . voice recognition applications.”
`
`Bailey at 5:28-30. The system “includes a specially optimized multiprocessing hardware unit
`
`capable of performing, in parallel, a multitude of steps required for pattern recognition procedures,
`
`such as . . . Hidden Markov Models.” Id. at 5:31-35. Mathew likewise describes “a special-
`
`purpose coprocessor architecture which improves the performance of Sphinx 3 while
`
`
`5 In KSR Int’l Co. v. Teleflex Inc., 550 U.S. 398 (2007), the Supreme Court held that prior art need
`not disclose the precise teachings of a patented invention to render it obvious, because a court “can
`take account of the inferences and creative steps that a person of ordinary skill in the art would
`employ.” Id. at 418. Under KSR, an explanation for why a combination of prior art items renders
`a claim obvious may be found in the “interrelated teachings of multiple patents; the effects of
`demands known to the design community or present in the marketplace; and the background
`knowledge possessed by a person having ordinary skill in the art . . . .” Id. at 418.
`
`12
`
`

`

`simultaneously reducing the energy requirements to the point where real-time, speaker-
`
`independent speech recognition is viable on embedded systems.” Mathew at 2. Mathew explains
`
`that “real-time large vocabulary speech recognition can be done within a power budget
`
`commensurate with embedded processing using” existing technologies. Id. at 1. Melnikoff is also
`
`directed to “a speech recognition system.” Melnikoff at 3. And Stogiannos describes “[a]n
`
`architecture . . . for real-time continuous speech recognition based on a modified hidden Markov
`
`model.” Stogiannos at Abstract. Stogiannos describes an algorithm that is “adapted to the needs
`
`of continuous speech recognition by efficient encoding of the state space.” Id. Yoshizawa II
`
`“discusses speed improvement by parallel processing and VLSI design, aiming at the realization
`
`of a high-speed speech recognition LSI.” Yoshizawa at 1.
`
`The POSITA would thus have been familiar with speech recognition systems. There were
`
`many such systems, and the POSITA would have found such systems highly relevant in designing
`
`a system similar to that of the Asserted Patents because they are analogous. Therefore, the POSITA
`
`would have engaged in the practice of learning about, consulting the teachings of, and adopting
`
`and combining improvements from known systems and other references related to speech
`
`recognition, as further detailed below. The rationale to combine or modify prior art references is
`
`strong when references seek to solve similar problems, come from the same field, and correspond
`
`well. In re Inland Steel Co., 265 F.3d 1354, 1362 (Fed. Cir. 2001).
`
`(2)
`
`Feature vector extraction
`
`The POSITA, already familiar with prior art speech recognition technologies would have
`
`also been familiar with the concept of extracting feature vectors for speech recognition. For
`
`example, Melnikoff explains that “[a] typical speech recognition system starts with a pre-
`
`processing stage, which takes a speech waveform as its input, and extracts from it feature vectors
`
`or observations which represent the information required to perform recognition.” Melnikoff at 2.
`
`13
`
`

`

`Melnikoff describes a “continuous HMM implementation” that extracts “continuous observation
`
`vectors.” Id. at 9. Tran similarly notes that “[w]ith respect to the feature extractor 180, a wide
`
`range of techniques is known in the art for representing the speech signal.” Tran at 11:11-13. A
`
`POSITA would have known that these techniques “include the short time energy, the zero crossing
`
`rates, the level crossing rates, the filter-bank spectrum, the linear predictive coding (LPC), and the
`
`fractal method of analysis.” Id. at 13-16. Stogiannos discloses that “front-end processor typically
`
`performs a short-time Fourier analysis and extracts a sequence of observation vectors (or acoustic
`
`vectors).” Stogiannos at 2 (alterations omitted); see also Melnikoff at 2 (explaining that “feature
`
`vectors” are synonymous with “observations”). Nguyen discloses an “exemplary speech
`
`recognizer [that] performs the recognition process in three steps”:
`
`First, speech analysis and feature extraction 10 is performed on the input Speech.
`This Step generates a Sequence of acoustic feature vectors representing the
`temporal and Spectral behavior of the Speech input. In general, an input Speech
`Signal is partitioned into a Sequence of time Segments or frames. Spectral features
`are then extracted from each frame using a variety of well known techniques.
`
`Nguyen at [0013]. Lund also discloses a “feature extractor.” Id. at 1, 13-14.
`
`The POSITA would have been motivated to combine the teachings of any of the references
`
`describing feature vector extraction with other references describing speech recognition systems
`
`to aid in speech recognition by distilling received audio into key features that a computer can
`
`analyze. Extracting feature vectors for this reason was common in speech recognition systems as
`
`demonstrated by the references described above. To the extent a speech recognition system did
`
`not include these features, the POSITA would have found it obvious to include the features in the
`
`system because it would merely be a combination of prior art elements according to known
`
`methods to yield predictable results, a simple substitution of one known element for another for
`
`another to obtain predictable results, and/or use of a known technique to improve a similar system
`
`in the same way.
`
`14
`
`

`

`(3)
`
`Hidden Markov Models
`
`The specification acknowledges that “[a]n important class of distance computation in
`
`speech recognition is the calculation of output state probabilities in recognisers using Hidden
`
`Markov Models.” ’277 patent at 4:15-17. The prior art confirms that “[t]he most widespread and
`
`successful approach to speech recognition is based on the Hidden Markov Model (HMM), which
`
`is a probabilistic process that models spoken utterances as the outputs of finite state machines.”
`
`Melnikoff at 4; see also Stogiannos at 2 (“state-of-the-art speech recognizers are based on
`
`statistical techniques, with the hidden Markov models being the dominant approach”). A POSITA
`
`would have appreciated that the “hidden Markov model is used . . . to evaluate the probability of
`
`occurrence of a sequence of observations” using a “probabilistic function.” Tran at 22:50-57.
`
`Yoshizawa II describes an “HMM circuit is composed of four modules, namely, the control,
`
`Gaussian distribution calculation, addlog operation, and Viterbi algorithm modules.” Yoshizawa
`
`II at 132. Thus, a POSITA would have known that a hidden Markov model could have been used
`
`in speech recognition systems because an HMM was a “highly successful model in continuous
`
`speech recognition.” Nguyen at [0012].
`
`The specification admits that “[t]here are a wide variety of probability distributions that
`
`can be used for the distance calculation stage of a speech recognizer” that are “widely documented
`
`in the literature.” ’277 patent at 3:57-62. A POSITA would have known that “[a] common choice
`
`is to use Gaussian Distributions and correspondingly the Mahalanobis Distance metric.” Id. at
`
`4:62-63. The specification acknowledges that “[t]he Mahalanobis Distance (MHD) is extensively
`
`described in the literature.” Id. at 12:51-53.
`
`Indeed, a POSITA would have known that “[i]f the front-end vector quantizes the acoustic
`
`vectors, the output distributions take the form of discrete probability distributions.” Stogiannos at
`
`3. One way to calculate a probability distribution is a “Mahalanobis distance of the feature from
`
`15
`
`

`

`a set of references used while training the recognizer.” Mathew at 2. Bailey also discloses a
`
`method for using a Hidden Markov model (HMM) that “first determines the probability that [an]
`
`unknown state 1 will be found with [a] probability distribution of the first state of [a] reference
`
`pattern,” and then “computes the probabilities that the unknown state 1 is within each of the other
`
`states of the test pattern (2 to n) along [a] first vertical lattice line . . . starting from the lowest state
`
`and sequentially processing lattice points.” Bailey at 8:2-12. And Tran discloses an HMM that
`
`uses “multivariate Gaussian function probability densities.” Tran at 22:6-9. A POSITA would
`
`have also known that probability density could have been calculated using “a probability density
`
`function bj(Ot) which determines the probability that state j emits a particular observation Ot at
`
`time t.” Melnikoff at 4.
`
`To the extent a reference does not disclose using an acoustic model or calculating
`
`probability distributions as performed in HMMs, the POSITA would have found it obvious to do
`
`so. The POSITA would have recognized the many benefits of using an HMM for speech
`
`recognition, including the efficiency of training the model using raw data received by the system.
`
`The POSITA would naturally have looked to existing solutions that provide this benefit such as
`
`the exemplary references described above. Such a combination is no more than a combination of
`
`prior art elements according to known methods to yield predictable results, and a simple
`
`substitution of one known element for another to obtain predictable results.
`
`(4)
`
`Pipelining
`
`The specification admits that pipelining techniques, like speech recognition systems,
`
`feature vector extraction, and hidden Markov models, were “well-known” in the art. ’277 patent
`
`at 34:6-14. The patent explains that “[p]ipelining may comprise processing data in different parts
`
`of a circuit at the same time, i.e. parallel processing.” Id. at 37:64-65.
`
`16
`
`

`

`Yoshizawa II describes a “vector pipeline” that is used to perform “parallel processing.”
`
`Yoshizawa II at 13. Mathew likewise describes a speech recognition system that includes “several
`
`levels at which [the] system may be integrated into a speech recognition task pipeline . . . .”
`
`Mathew at 8. Melnikoff also describes a speech recognition circuit that uses “a fully pipelined
`
`architecture.” Melnikoff at 8. And Bailey discloses a speech recognition system that uses two
`
`pipelines: an “arithmetic pipeline” for “executing arithmetic instructions” and a “pointer pipeline”
`
`for “executing pointer instructions . . . in parallel with the step of executing arithmetic instructions.”
`
`Bailey at 4:10-14. The POSITA would have found it obvious to include such features in a speech
`
`recognition system because it would merely be a combination of prior art elements according to
`
`known methods to yield predictable results and/or use of a known technique to improve a similar
`
`system in the same way.
`
`4.
`
`Invalidity Under § 112
`
`The Asserted Claims are invalid under 35 U.S.C. § 112. The Asserted Claims, read in
`
`light of the specification and prosecution history, fail to inform, with reasonable certainty, those
`
`skilled in the art about the scope of the purported invention and are therefore indefinite. See
`
`Nautilus, Inc. v. Biosig Instruments, Inc., 572 U.S. 898, 901 (2014). The asserted patent does not
`
`enable one of skill in the art to practice the full scope of the invention claimed without undue
`
`experimentation. It instead seeks to improperly “monopolize an entire class of things defined by
`
`their function.” Amgen Inc. v. Sanofi, No. 21-757, 2023 WL 3511533, at *2 (U.S. May 18, 2023).
`
`The following identification of claims/claim elements are only exemplary and Defendants
`
`reserve the right to supplement the identification of claims and claim elements that do not comply
`
`with the requirements of 35 U.S.C. § 112. Specifically, to the extent an element identified below,
`
`or its variation, appears in claims other than the ones specified below, it also renders those
`
`additional claims invalid under 35 U.S.C. § 112. Claims that depend on these additional claims
`
`17
`
`

`

`and on the claims identified below are also invalid under 35 U.S.C. § 112. Defendants reserve the
`
`right to identify additional claims and claim elements that do not comply with the requirements of
`
`35 U.S.C. § 112 should Zentian make further allegations based on claim scope lacking the requisite
`
`§ 112 support, including in its final infringement contentions. Furthermore, this identification of
`
`claims/claim elements should not be interpreted as reflecting Defendants’ position on the
`
`construction of any claim/claim element.
`
`At least the following claims,