Trials@uspto.gov
`571-272-7822
`
`
` Paper No. 33
`Entered: April 10, 2024
`
`
`
`
`UNITED STATES PATENT AND TRADEMARK OFFICE
`
`
`
`BEFORE THE PATENT TRIAL AND APPEAL BOARD
`
`
`
`
`
`
`
`APPLE INC.,
`Petitioner
`v.
`ZENTIAN LIMITED,
`Patent Owner.
`
`
`
`
`
`IPR2023-00033
`Patent 7,587,319 B2
`
`
`
`Record of Oral Hearing
`Held: March 11, 2024
`
`__________
`
`Before KEVIN F. TURNER, JEFFREY S. SMITH, and
`CHRISTOPHER L. OGDEN, Administrative Patent Judges.
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`

`

`
`IPR2023-00033
`Patent 7,587,319 B2
`
`
`
`
`APPEARANCES:
`
`
`ON BEHALF OF THE PETITIONER:
`
`
`JENNIFER BAILEY, ESQ.
`Erise IP, P.A.
`7015 College Boulevard
`Suite 700
`Overland Park, KS 66211
`(913) 777-5600
`Jennifer.bailey@eriseip.com
`
`CHRISTINA CANINO, ESQ.
`Erise IP, P.A.
`5299 DTC Boulevard
`Suite 1340
`Greenwood Village, CO 80111
`(720) 307-8382
`Christina.Canino@eriseip.com
`
`JENNY LIU, ESQ.
`Apple Inc.
`1 Apple Park Way
`Cupertino, CA 95014
`(408) 996-1010
`jianing_liu@apple.com
`
`
`
`
`
`
`
` 2
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`

`

`
`IPR2023-00033
`Patent 7,587,319 B2
`
`
`
`ON BEHALF OF THE PATENT OWNER:
`
`
`KAYVAN NOROOZI, ESQ.
`PETER KNOPS, ESQ.
`Noroozi PC
`11601 Wilshire Boulevard
`Suite 2170
`Los Angeles, CA 90025
`(310) 975-7074 ex. 1
`kayvan@noroozipc.com
`peter@noroozipc.com
`
`JESSICA R. BERNHARDT, ESQ.
`Bartlit Beck LLP
`Courthouse Place
`54 West Hubbard Street
`Chicago, IL 60654
`(312) 494-4460
`Jessica.bernhardt@bartlitbeck.com
`
`DARGAYE CHURNET, ESQ.
`Fenwick & West LLP
`555 California St. #12
`San Francisco, CA 94104
`(415) 875-2460
`dchurnet@fenwick.com
`
`
`
`
`The above-entitled matter came on for hearing Monday,
`
`March 11, 2023, commencing at 1:03 p.m. EDT, via Video-conference.
`
`
`
`
`
` 3
`
`
`
`
`
`

`

`IPR 2023-00033
`Patent 7,587,319 B2
`
`P-R-O-C-E-E-D-I-N-G-S
`
`1:03 p.m.
`
`
`
`
`
`JUDGE OGDEN: Good afternoon. Welcome to the Patent
`
`Trial and Appeal Board. This is the final hearing in IPR2023-0033 between
`
`Petitioners Apple Inc., Amazon Web Services Inc., Amazon.com Services,
`
`LLC, and Patent Owner, Zentian Limited.
`
`
`
`
`
`And I am Judge Ogden, and today with me on the panel are
`
`Judges Turner and Smith. And Judge Smith is appearing remotely, joining
`
`us remotely.
`
`
`
`
`
`So let's start with counsel introductions. Who will present
`
`arguments for Petitioner, and who's at the Petitioner's counsel table today?
`
`
`
`
`
`MR. TOROSSIAN: This is Charle Torossian. I'm sorry for the
`
`interruption. We don't hear the audio. I don't know if the mic is on on their
`
`desk. Sorry.
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`MS. BAILEY: Can you hear me now, Charle? Sorry.
`
`MR. TOROSSIAN: Yes.
`
`MS. BAILEY: I'll repeat. I'll start over.
`
`Good morning. My name is Jennifer Bailey. I am from the law
`
`firm of Erise IP. I have with me today my co-counsel, Christina Canino, and
`
`I also have in-house counsel from Petitioner Apple Inc., Jenny Liu. Thank
`
`you.
`
`
`
`
`
`
`
`
`
`JUDGE OGDEN: Okay. Go ahead.
`
`MR. NOROOZI: Your Honor, Kayvan Noroozi on behalf of
`
`Patent Owner from Noroozi PC. I have with me Mr. Peter Knops, as well as
`
`Ms. Jessica Bernhardt.
`
`1
`
`2
`
`3
`
`4
`
`5
`
`6
`
`7
`
`8
`
`9
`
`10
`
`11
`
`12
`
`13
`
`14
`
`15
`
`16
`
`17
`
`18
`
`19
`
`20
`
`21
`
`22
`
`23
`
`24
`
`25
`
`
`
`
`
` 4
`
`
`
`
`
`

`

`IPR 2023-00033
`Patent 7,587,319 B2
`
`
`
`
`
`JUDGE OGDEN: Thank you. Do we have an appearance
`
`today for Amazon?
`
`
`
`
`
`MR. CHURNET: Hello. Dargaye Churnet from Fenwick &
`
`West representing Amazon.
`
`
`
`
`
`JUDGE OGDEN: Thank you. Just a reminder, when you're
`
`speaking today, please come closer to the microphone so that people
`
`remotely can hear us.
`
`
`
`
`
`So before we begin today, just a couple of things. I just remind
`
`everybody to speak clearly into the microphone. We have a court reporter
`
`that's working remotely, so please be sure that what you're saying in the
`
`microphone is clear, so the reporter can hear.
`
`
`
`
`
`And just a reminder, there's no objections during opposing
`
`arguments. You can save any concerns or objections until your time to
`
`speak. And this is a public hearing, so keep that in mind. I don't believe that
`
`there is any confidential information to be discussed today, but just keep that
`
`in mind.
`
`
`
`
`
`So according to our hearing order, each side has a total of 45
`
`minutes, but Petitioner has a LEAP practitioner, so that will include an
`
`additional 15 minutes for that. And each side will have the opportunity to
`
`reserve time for rebuttal or surrebuttal at the end, and so if you could tell us
`
`at the beginning how much time you'd like to reserve for that. We have a
`
`timer and we'll set the timer so that for principal arguments you'll have a
`
`five-minute warning. The light will go yellow, and the amount of time left
`
`for your arguments will appear up on the timer.
`
`
`
`
`
`Sure. Do I need to repeat anything? Okay. Okay. Thank you.
`
`
`
`
`
` 5
`
`
`
`Okay. So let's --
`
`
`
`1
`
`2
`
`3
`
`4
`
`5
`
`6
`
`7
`
`8
`
`9
`
`10
`
`11
`
`12
`
`13
`
`14
`
`15
`
`16
`
`17
`
`18
`
`19
`
`20
`
`21
`
`22
`
`23
`
`24
`
`25
`
`26
`
`

`

`IPR 2023-00033
`Patent 7,587,319 B2
`
`
`
`
`
`JUDGE TURNER: I'm sorry. This is Judge Turner. Is there a
`
`question?
`
`
`
`
`
`
`
`
`
`
`
`
`
`MR. NOROOZI: Yes, Your Honor.
`
`JUDGE TURNER: You certainly may. Yes.
`
`MR. NOROOZI: With respect to the LEAP additional time
`
`issue, it's my understanding that Petitioner is only having the LEAP
`
`practitioner argue one of the five proceedings, and that will not be in this
`
`proceeding.
`
`
`
`
`
`Does the additional 15 minutes apply, even if a LEAP
`
`practitioner is not arguing in the proceeding?
`
`
`
`
`
`JUDGE OGDEN: No. No. Okay, let me confirm that. Is that
`
`the case, Petitioner?
`
`
`
`
`
`
`
`
`
`
`
`
`
`MS. BAILEY: Yes, Your Honor.
`
`JUDGE OGDEN: Okay.
`
`MS. BAILEY: The LEAP practitioner is only participating in
`
`one of the proceedings tomorrow, so we do not need the 15 extra minutes for
`
`today's proceedings.
`
`
`
`
`
`JUDGE OGDEN: Okay. Which proceeding will that be
`
`tomorrow?
`
`
`
`
`
`MS. BAILEY: It will be on the 377 patent, and I don't have the
`
`IPR number off the top of my head. It's the 35.
`
`
`
`that.
`
`
`
`
`
`
`
`
`
`
`
`JUDGE OGDEN: 35. Okay. Okay. Thank you for clarifying
`
`MS. BAILEY: Yes.
`
`JUDGE TURNER: Just for clarification of the record, the
`
`response was actually entered in all of these cases. So I think that's maybe
`
`
`
`
` 6
`
`
`
`
`
`1
`
`2
`
`3
`
`4
`
`5
`
`6
`
`7
`
`8
`
`9
`
`10
`
`11
`
`12
`
`13
`
`14
`
`15
`
`16
`
`17
`
`18
`
`19
`
`20
`
`21
`
`22
`
`23
`
`24
`
`25
`
`26
`
`

`

`IPR 2023-00033
`Patent 7,587,319 B2
`
`why we got confused. So it was acknowledged in all of the IPR
`
`proceedings, but you only sought a LEAP practitioner in a single case.
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`MS. BAILEY: Correct.
`
`JUDGE TURNER: Okay.
`
`MS. BAILEY: Our request was only for one specific case.
`
`JUDGE TURNER: Okay.
`
`MS. BAILEY: Yes.
`
`JUDGE TURNER: So every party will have 45 minutes per
`
`case, per side.
`
`
`
`
`
`
`
`
`
`
`
`
`
`MS. BAILEY: Thank you, Your Honor.
`
`JUDGE TURNER: Thank you.
`
`JUDGE OGDEN: Thank you. Okay. We'll start with
`
`Petitioner then. So yeah, Petitioner will have a total of 45 minutes then.
`
`
`
`
`
`If you could approach the podium, and how much time would
`
`you like to reserve for rebuttal?
`
`
`
`
`
`MS. BAILEY: I'd like to reserve 15 minutes for rebuttal,
`
`please.
`
`
`
`
`
`JUDGE OGDEN: Okay. We've put 30 minutes on the clock,
`
`and you'll get a five-minute warning when the light goes yellow. You can
`
`begin when you're ready.
`
`
`
`
`
`MS. BAILEY: Thank you. May it please the board. Thank
`
`you, Your Honors.
`
`
`
`
`
`We don't have our slides on screen here today, but I'll be sure to
`
`refer to the slides by their number. So let's start with Petitioner's
`
`Demonstratives DX-3, please.
`
`1
`
`2
`
`3
`
`4
`
`5
`
`6
`
`7
`
`8
`
`9
`
`10
`
`11
`
`12
`
`13
`
`14
`
`15
`
`16
`
`17
`
`18
`
`19
`
`20
`
`21
`
`22
`
`23
`
`24
`
`25
`
`
`
`
`
` 7
`
`
`
`
`
`

`

`IPR 2023-00033
`Patent 7,587,319 B2
`
`
`
`
`
`I want to quickly go through what the math combination is.
`
`The combination is the Thelen, Bailey, and Chen, where Thelen's hierarchal
`
`division of acoustic models, according to context, are put on the Chen
`
`clustered processor architecture.
`
`
`
`
`
`So, for example, with Thelen and its hierarchal divisions,
`
`there's going to be one context of the lexicon specific to sports, one specific
`
`to health, perhaps one specific to cooking. Bailey teaches us that multiple
`
`speech recognition processors are connected to a single shared memory.
`
`And then Chen, of course, teaches the clustered processor architecture.
`
`
`
`
`
`So the combination is that Thelen's hierarchal arrangement of
`
`context are divided up amongst the clusters of Chen, according to the
`
`context. So, for example, Cluster 1 of Chen's clustered processor
`
`architecture would be devoted to a sports context. Cluster 2 would be
`
`devoted to health and so forth.
`
`
`
`
`
`Turning to DX-4, this slide is just here to illustrate the
`
`similarity between Chen's clustered processor architecture and the clustered
`
`processor architecture described and claimed in the '319 patent. And as you
`
`can see, both Chen and the '319 patent illustrate multiple processors
`
`connected to a single shared memory. Each one of that arrangement is a
`
`cluster, and then there are multiple clusters.
`
`
`
`
`
`Chen also teaches that, I will note, that its architecture is, quote,
`
`equally applicable to any type of parallel processing computer system. And
`
`I'm quoting from Column 11, Lines 33 through 49 of Chen. So
`
`understanding that Chen teaches the same clustered processing architecture
`
`is the '319 patent, let's look at the petition's motivations to combine, turning
`
`
`
`
`
` 8
`
`
`
`to DX-5.
`
`
`
`1
`
`2
`
`3
`
`4
`
`5
`
`6
`
`7
`
`8
`
`9
`
`10
`
`11
`
`12
`
`13
`
`14
`
`15
`
`16
`
`17
`
`18
`
`19
`
`20
`
`21
`
`22
`
`23
`
`24
`
`25
`
`26
`
`

`

`IPR 2023-00033
`Patent 7,587,319 B2
`
`
`
`
`
`So Zentian does not dispute that the claim limitations are
`
`taught. Zentian's sole argument in this IPR is a lack of motivation to
`
`combine for two related reasons. First, that the combination would create
`
`tasks for which there would not be a reasonable expectation of success. And
`
`then related to that, that the skilled person would not know how to address
`
`these tasks when putting together the combination.
`
`
`
`
`
`Neither of these arguments address the petition's actual
`
`proffered motivations to combine, and Zentian actually has a fairly limited
`
`response to the petition's motivations. Notably, Zentian has only a single
`
`sentence at the end of its sur-reply responsive to what is arguably the
`
`petition's primary motivation of reducing the duplication of lexical data
`
`amongst the multiple memories. And I refer the board to Patent Owner's
`
`sur-reply at Page 26. It's at the very bottom spanning to Page 27 for that
`
`single sentence.
`
`
`
`
`
`I want to quickly kind of go through the combination and why
`
`we receive or have the benefit of the reduced duplication of lexical data. In
`
`the combination, each cluster stores a portion of the acoustic model that is
`
`for that context.
`
`
`
`
`
`We've already discussed the Cluster 1, which stores the sports
`
`context, Cluster 2 may store the health. Because similar contexts, for
`
`example, all the sports-related terms are clustered together, an acoustic
`
`model specific to sports-related words does not have to be saved in all of the
`
`clusters' memories. This ends up reducing the duplication of lexical data
`
`amongst all of the clusters' memories.
`
`
`
`
`
`So let me give you an actual example. Let's take the word,
`
`offsides. Offsides is a word that is specific to football and hockey and
`
`
`
`
` 9
`
`
`
`
`
`1
`
`2
`
`3
`
`4
`
`5
`
`6
`
`7
`
`8
`
`9
`
`10
`
`11
`
`12
`
`13
`
`14
`
`15
`
`16
`
`17
`
`18
`
`19
`
`20
`
`21
`
`22
`
`23
`
`24
`
`25
`
`26
`
`

`

`1
`
`2
`
`3
`
`4
`
`5
`
`6
`
`7
`
`8
`
`9
`
`10
`
`11
`
`12
`
`13
`
`14
`
`15
`
`16
`
`17
`
`18
`
`19
`
`20
`
`21
`
`22
`
`23
`
`24
`
`25
`
`26
`
`IPR 2023-00033
`Patent 7,587,319 B2
`
`soccer, but it's also a word that doesn't really show up in our lexicon outside
`
`of sports. So the word offsides now does not have to be duplicated in all of
`
`the memories of all of the clusters, because now we are clustering the sports
`
`contexts together in a single shared memory. And because the word offsides
`
`is common to multiple sports, we would also be obtaining increased
`
`efficiency in the speech recognition by grouping all the sports contexts
`
`together.
`
`
`
`
`
`And I just want to hit home that all of that motivation that I just
`
`described, that is described at length in the petition and the declaration, is
`
`addressed by Zentian in that single sentence in their sur-reply at Page 26.
`
`And Zentian understood this motivation. They asked Mr. Schmandt about it
`
`during his deposition. I refer the board to Exhibit 2017, which is Mr.
`
`Schmandt's depo transcript. Page 122, Line 13 through Page 125, 7.
`
`
`
`
`
`So Zentian understood what this motivation was. It just chose
`
`not to respond to it. That's not the only motivation that was provided in the
`
`petition. There's also reduced financial costs because with the reduced
`
`duplication of data, there is now less memory to operate. So therefore,
`
`there's reduced financial costs there.
`
`
`
`
`
`The combination also takes advantage of Chen's commercially
`
`available processors. Cluster processing architecture was known to have
`
`flexibility and scalability because you could continually add more
`
`processors. Hon teaches that. And these motivations are provided in
`
`Exhibit 1003, Paragraph 84.
`
`
`
`
`
`We also have increased processing power for a given cost
`
`because now we're using Chen's commercially available processors that are
`
`inexpensive. We're using less memory. There are less memory transfers
`
`
`
`
`10
`
`
`
`

`

`IPR 2023-00033
`Patent 7,587,319 B2
`
`from the huge data store. And so the overall result is the increased
`
`processing power while not extending the raising of the financial costs.
`
`
`
`
`
`I also want to note that the motivation to combine is not
`
`contingent on adding additional processors to the combination. That's been
`
`something that's discussed by Zentian. And I do want to make clear, we also
`
`said this in our Petitioner reply, I believe at Pages 24 through 25, that it's not
`
`contingent on that. Certainly more processors could be added, but it's not
`
`contingent on that.
`
`
`
`
`
`Turning now to DX-5, I want to address the task that Zentian
`
`raises. So instead of substantively responding to the petition's mapping and
`
`motivations, Zentian raises --
`
`
`
`
`
`JUDGE SMITH: Yeah, Counsel, before we move on to that,
`
`can I ask you about these advantages that you're talking about? I just have
`
`some questions. It's not clear to me what advantages are provided by Chen
`
`that aren't already in the primary reference Thelen.
`
`
`
`
`
`
`
`
`
`MS. BAILEY: Yeah.
`
`JUDGE SMITH: For example, if you have the first cluster-
`
`shared memory devoted to sports with the sports vocabulary, Thelen would
`
`already have that, wouldn't it?
`
`
`
`
`
`MS. BAILEY: That's a good question, Your Honor. It's
`
`actually one that I asked Mr. Schmandt when we were putting together the
`
`combination. Let me address that.
`
`
`
`
`
`So in Thelen, it talks about generic context and specific context.
`
`So a generic context would be sports. The specific context would be
`
`football or baseball. And in Thelen, it would be understood that each
`
`1
`
`2
`
`3
`
`4
`
`5
`
`6
`
`7
`
`8
`
`9
`
`10
`
`11
`
`12
`
`13
`
`14
`
`15
`
`16
`
`17
`
`18
`
`19
`
`20
`
`21
`
`22
`
`23
`
`24
`
`25
`
`
`
`
`
`11
`
`
`
`

`

`1
`
`2
`
`3
`
`4
`
`5
`
`6
`
`7
`
`8
`
`9
`
`10
`
`11
`
`12
`
`13
`
`14
`
`15
`
`16
`
`17
`
`18
`
`19
`
`20
`
`21
`
`22
`
`23
`
`24
`
`25
`
`26
`
`IPR 2023-00033
`Patent 7,587,319 B2
`
`recognizer would be associated with a specific context. So you would have
`
`one for baseball, one for football.
`
`
`
`
`
`Let's go back to the example that I gave earlier of offsides and
`
`that it's used in multiple different sports. So now you would have the
`
`duplication of the word offsides in multiple recognizers because these sports
`
`recognizers were not grouped together. Mr. Schmandt said with respect to
`
`Thelen that it is not a, quote, optimally efficient recognizer. And I refer the
`
`board to Exhibit 2017, Page 122, Line 22 through Page 123, Line 19.
`
`
`
`
`
`
`
`
`
`Does that answer your question, Your Honor?
`
`JUDGE SMITH: Let me make sure I understand. So you're
`
`saying the word offsides would be in the shared memory, and it would be
`
`kind of to the memory model for soccer and hockey and what other sport
`
`uses offsides. Whereas Thelen duplicates it for each of those sports? Is that
`
`what you're saying?
`
`
`
`
`
`MS. BAILEY: Yes, that's correct. And we could also come up
`
`with terms that are common to sports, say, score or point. And I appreciate
`
`that those words would show up in other contexts, but within the sports
`
`context, those are very common words. And so these common vocabulary
`
`within a particular generic or genus context would have to be duplicated
`
`each time within Thelen's specific recognizers because Thelen has no
`
`grouping of the clusters or of the context as we propose in the combination.
`
`
`
`
`
`JUDGE SMITH: Okay. And then I have another question
`
`about in Chen, these four different clusters, as far as I understand Chen, are
`
`all operating at the same time, which in terms of dividing a task and
`
`separating it among these multiple processes would provide an advantage in
`
`terms of processing power.
`
`
`
`
`
`12
`
`
`
`

`

`1
`
`2
`
`3
`
`4
`
`5
`
`6
`
`7
`
`8
`
`9
`
`10
`
`11
`
`12
`
`13
`
`14
`
`15
`
`16
`
`17
`
`18
`
`19
`
`20
`
`21
`
`22
`
`23
`
`24
`
`IPR 2023-00033
`Patent 7,587,319 B2
`
`
`
`
`
`But from what I understand, if you have these clusters for
`
`sports in the first cluster, these other three clusters in Chen, those processors
`
`would be idle; is that correct?
`
`
`
`
`
`MS. BAILEY: No, that's not correct, Your Honor. Let me kind
`
`of go further in what the combination would explain or would propose.
`
`
`
`
`
`In Thelen, it teaches that it is going to determine what context is
`
`being spoken by recognizing a few words. So a few words are spoken. It
`
`recognizes, oh, they're talking about sports here.
`
`
`
`
`
`And so what Thelen teaches and what we would propose in the
`
`combination is that we have a large data store where it was holding the
`
`entire acoustic model for the entire lexicon. And from that large data store,
`
`you would pull the sports context. And at that point, you would pull sports
`
`into all of the four clustered processors to gain the efficiency because we
`
`know that the speech is specific to sports.
`
`
`
`
`
`That is also, recall that I just mentioned, transferring the
`
`acoustic models from the large data store, but also has increased efficiency
`
`because you're now minimizing how much of the acoustic model has to be
`
`transferred from the large data store for any particular recognition instance.
`
`
`
`
`
`JUDGE SMITH: So in your combination, all of the clusters
`
`would be accessing the shared memory that's colored red for the sports; is
`
`that correct?
`
`
`
`
`
`MS. BAILEY: Let me pull up the Chen. So let me change that
`
`understanding just a little bit, Your Honor. I don't think it's quite what we're
`
`proposing.
`
`
`
`
`
`
`
`13
`
`

`

`1
`
`2
`
`3
`
`4
`
`5
`
`6
`
`7
`
`8
`
`9
`
`10
`
`11
`
`12
`
`13
`
`14
`
`15
`
`16
`
`17
`
`18
`
`19
`
`20
`
`21
`
`22
`
`23
`
`24
`
`25
`
`IPR 2023-00033
`Patent 7,587,319 B2
`
`
`
`
`
`So there would be a large data store that will hold the entire
`
`acoustic model. That's not one of the memories of the clustered processors
`
`that we're looking at.
`
`
`
`
`
`If, per Thelen, it recognizes that sports is being discussed, then
`
`it will pull the sports context into each of the shared cluster memories of the
`
`four clusters so that all four clusters can run the sports context. But then
`
`once it moves to the next recognition task, the next instance, it recognizes,
`
`for example, that health is being discussed, at which point it would transfer
`
`out the acoustic models from the shared cluster memory and transfer in from
`
`the large data store, the health model.
`
`
`
`
`
`JUDGE SMITH: Okay. So one cluster would be doing
`
`baseball, one would be doing football. Or they would all be doing
`
`everything in sports?
`
`
`
`
`
`
`
`
`
`MS. BAILEY: They would all be doing everything in sports.
`
`JUDGE SMITH: And why would you want to duplicate that
`
`processing four times?
`
`
`
`
`
`MS. BAILEY: You would be then dividing up from the
`
`acoustic model, and this is actually how most acoustic models are divided
`
`up. And so then you would, say, have every phoneme that begins with, say,
`
`A through F would be one cluster. The next cluster would be G through K.
`
`
`
`
`
`So you would take your sports model and then divide it up
`
`amongst the different phonemes. So it actually would not be duplicated.
`
`You would have multiple processors operating for one cluster on one portion
`
`of the sports acoustic model.
`
`
`
`
`
`
`
`JUDGE SMITH: Okay. Thank you.
`
`
`
`
`
`14
`
`

`

`1
`
`2
`
`3
`
`4
`
`5
`
`6
`
`7
`
`8
`
`9
`
`10
`
`11
`
`12
`
`13
`
`14
`
`15
`
`16
`
`17
`
`18
`
`19
`
`20
`
`21
`
`22
`
`23
`
`24
`
`25
`
`26
`
`IPR 2023-00033
`Patent 7,587,319 B2
`
`
`
`
`
`MS. BAILEY: Okay. Let's turn back to DX-6. Oh, sorry. Go
`
`ahead.
`
`
`
`
`
`JUDGE OGDEN: No. Hold on. Hold on for a second. Patent
`
`Owner, do you have a question?
`
`
`
`
`
`MR. NOROOZI: Your Honor, it's just as a matter of procedure.
`
`If I have an objection on the basis that an argument is new or was not
`
`disclosed before, should I keep that for my argument?
`
`
`
`
`
`
`
`
`
`JUDGE OGDEN: Yes, please keep that. Yeah.
`
`JUDGE TURNER: This is Judge Turner. While you're
`
`stopped, let me ask maybe, hopefully, an edifying question. Is Petitioner
`
`relying on all of the references of all of the teachings of Chen et al?
`
`
`
`
`
`MS. BAILEY: No, Your Honor. We are only relying on the
`
`portions of the reference needed to teach each of the limitations in Claim 46.
`
`
`
`
`
`JUDGE TURNER: And I ask that because we recently had a
`
`case also, coincidentally, involving a reference Chen where we were
`
`reversed where we said there was no reasonable expectation of success. And
`
`that was in Netflix we did it, where we said, well, we take the whole context
`
`of the reference, and we think it does. And basically on appeal, in that case,
`
`Petitioner said, oh, no, but I'm not relying on anything else. Just this one
`
`section of Chen is what I'm working on.
`
`
`
`
`
`So, do we have that situation here where we need to be very
`
`cautious? Or can we take Chen for all that it teaches, for example?
`
`
`
`
`
`MS. BAILEY: We rely on Chen for certain teachings. I'm not
`
`familiar with the Netflix case, but I would say that the way that the
`
`combination is put together, that any ground is put together, is that we start
`
`with our primary, and we pull from the secondary and tertiary references as
`
`
`
`
`15
`
`
`
`

`

`1
`
`2
`
`3
`
`4
`
`5
`
`6
`
`7
`
`8
`
`9
`
`10
`
`11
`
`12
`
`13
`
`14
`
`15
`
`16
`
`17
`
`18
`
`19
`
`20
`
`21
`
`22
`
`23
`
`24
`
`25
`
`IPR 2023-00033
`Patent 7,587,319 B2
`
`needed. But we need to be mindful of not bodily incorporating elements
`
`from the secondary and tertiary references because, of course, bodily
`
`incorporation is impermissible.
`
`
`
`
`
`Now there is case law that says that we look at all of the
`
`references together and see what the combination would do to put together.
`
`But in our proposed ground, you do not need to consider anything from the
`
`secondary and tertiary references other than what we have mapped in the
`
`limitations and have put into our combination.
`
`
`
`
`
`
`
`
`
`Does that answer your question?
`
`JUDGE TURNER: You said we don't need to, but I guess can
`
`we, is sort of maybe the question that's begged there.
`
`
`
`
`
`MS. BAILEY: I would actually disagree with that from a
`
`matter of kind of academic point of view. Now there is case law, and I will
`
`be citing it actually in a different one of these matters, that if you have a
`
`primary reference, you can't modify that primary reference or say that it is
`
`inoperable to modify the primary reference for reasonable expectation of
`
`success if the context of that primary reference is not pertinent to the claim
`
`limitations.
`
`
`
`
`
`So let me know if I'm not answering your question, Your
`
`Honor, but I do think that it would be inappropriate for you to take
`
`something from Chen, which is a tertiary reference that is not relied on in the
`
`combination and say that this particular feature in Chen would then
`
`somehow impact the reasonable expectation of success argument. I am not
`
`aware of any case law that does that, but I will be going back and reading the
`
`Netflix case.
`
`
`
`
`
`
`
`16
`
`

`

`1
`
`2
`
`3
`
`4
`
`5
`
`6
`
`7
`
`8
`
`9
`
`10
`
`11
`
`12
`
`13
`
`14
`
`15
`
`16
`
`17
`
`18
`
`19
`
`20
`
`21
`
`22
`
`23
`
`24
`
`25
`
`26
`
`IPR 2023-00033
`Patent 7,587,319 B2
`
`
`
`
`
`JUDGE TURNER: Okay, but let's say we're pruning, just for
`
`hypothetical example, perhaps. Let's say we're pruning. I want to make sure
`
`that are we going too far afield by looking at this? Because most of what
`
`Patent Owner here is saying is, look, we don't have a reasonable expectation
`
`of success because if you have A and B, and let's say we're putting them
`
`together, that one of ordinary skill maybe wouldn't have done that, and there
`
`wouldn't have been expectation of success.
`
`
`
`
`
`So, when I sort of ask how much are we to take from your
`
`referencing, you're sort of saying, well, just what I'm telling you. Maybe
`
`that gives me some pause because I'm trying to figure out what that ground
`
`is. How much expansion can the vision support or not?
`
`
`
`
`
`MS. BAILEY: So I don't think the ground or Petitioner's
`
`arguments require expansion beyond what was put into the petition and the
`
`declaration, first of all. Second, if you analyze obviousness of taking the
`
`references and mixing and matching their teachings to the point where you
`
`could create something that had no reasonable expectation of success, I
`
`suspect that that could be done in a large variety of grounds. That is not the
`
`proper legal analysis for obviousness.
`
`
`
`
`
`
`
`
`
`JUDGE TURNER: Okay. Thank you.
`
`MS. BAILEY: So you did just ask a question about Zentian's
`
`lack of reasonable expectation of success argument, so let's look at DX-6.
`
`Zentian's argument is that there would be these tasks that are listed on DX-6
`
`that would have to be performed that would lead to the lack of reasonable
`
`expectation of success. I first note that the task sharing and the messaging
`
`strategy identified by Zentian, they don't talk about at all in their briefing.
`
`They just mention that that task would need to be performed, but there's no
`
`
`
`
`17
`
`
`
`

`

`IPR 2023-00033
`Patent 7,587,319 B2
`
`explanation of why the combination would have these tasks that would be
`
`insurmountable.
`
`
`
`
`
`With respect to bandwidth, Zentian cites the section of the '319
`
`patent copy below, but as you can see, the section says, by providing a
`
`plurality of processors in a group with a common memory, bandwidth is
`
`saved. Well, that's what's in the claimed combination. So, if anything, the
`
`combination would not have bandwidth issues.
`
`
`
`
`
`Zentian mostly discusses memory collision or memory
`
`contention issues, which we will get to. But in the sur-reply, Zentian seems
`
`to concede that these issues could be addressed. In other words, Zentian
`
`doesn't argue that these tasks are insurmountable tasks. It argues that the
`
`task just wouldn't be done by a skilled person. Zentian --
`
`
`
`
`
`
`
`
`
`
`
`
`
`JUDGE OGDEN: I'd like to --
`
`MS. BAILEY: Yes, Your Honor.
`
`JUDGE OGDEN: -- ask a question about the level of ordinary
`
`skill in the art.
`
`
`
`
`
`
`
`
`
`MS. BAILEY: Yes.
`
`JUDGE OGDEN: It's my understanding, and I hope you can
`
`just clarify this for the record, that the level of ordinary skill in the art, that
`
`Petitioner proposed in the petition, sticking with that, which is a person who
`
`has an educational degree in computer engineering, computer science,
`
`electrical engineering, or related field, and then also some experience in the
`
`field of speech recognition.
`
`
`
`
`
`But as you articulate that, it does not include any specific
`
`additional experience in high-performance computing or deep hardware
`
`1
`
`2
`
`3
`
`4
`
`5
`
`6
`
`7
`
`8
`
`9
`
`10
`
`11
`
`12
`
`13
`
`14
`
`15
`
`16
`
`17
`
`18
`
`19
`
`20
`
`21
`
`22
`
`23
`
`24
`
`25
`
`
`
`
`
`18
`
`
`
`

`

`1
`
`2
`
`3
`
`4
`
`5
`
`6
`
`7
`
`8
`
`9
`
`10
`
`11
`
`12
`
`13
`
`14
`
`15
`
`16
`
`17
`
`18
`
`19
`
`20
`
`21
`
`22
`
`23
`
`24
`
`25
`
`26
`
`IPR 2023-00033
`Patent 7,587,319 B2
`
`experience beyond what would a person of ordinary skill in the art, as you
`
`define it, would have had; is that correct? That's your position today?
`
`
`
`
`
`MS. BAILEY: I have two responses to that question, Your
`
`Honor. You're correct that the field of high-performance computing is not
`
`listed here, but any skilled person with these credentials would be able to
`
`address the tasks that Zentian identifies. Every EE, every CS that I have
`
`ever met has taken a class on parallel processing architecture. But more
`
`importantly, the art of record overwhelmingly shows that speech recognition
`
`researchers at the time were investigating and addressing parallel processing
`
`architectures.
`
`
`
`
`
`So if we're going to talk about memory contention issues, which
`
`is a common task in parallel processing, we have multiple art of record
`
`showing that speech recognition researchers were working in parallel
`
`processing systems at the time. There's no indication in the record that they
`
`would not be able to address a simple memory contention issue.
`
`
`
`
`
`JUDGE OGDEN: Do we have expert testimony positively
`
`supporting that? That a person with one of these degrees would've had
`
`enough experience in the field of the hardware field in order to address these
`
`issues, such as memory collisions and tasks theory and so forth?
`
`
`
`
`
`MS. BAILEY: We have evidence of the references that already
`
`shows that. In other words, we don't need to have an expert also say what
`
`the art of record already teaches. Chen already addresses memory collision
`
`issues, Hon addresses memory bottlenecks. Ravishankar talks about parallel
`
`processing and memory bottlenecks.
`
`
`
`
`
`JUDGE OGDEN: Well, I think maybe the issue is, for
`
`example, with Chen, I think Patent Owner is arguing that maybe, and Patent
`
`
`
`
`19
`
`
`
`

`

`IPR 2023-00033
`Patent 7,587,319 B2
`
`Owner can correct me later if I'm wrong, but I think the argument is sort of
`
`maybe a person of ordinary skill in the art may be able to start with Thelen
`
`and the other reference, but the Chen reference is in a sort of a different field
`
`that maybe this person of ordinary skill in the art wouldn't have had
`
`experience with. And so maybe they wouldn't have been able to make that
`
`combination involving Chen.
`
`
`
`
`
`MS. BAILEY: So I do disagree that Chen is in a different field.
`
`It is related to parallel processing architecture, and it says that it can be used
`
`for any type of parallel processing system. But more pointedly to your
`
`question or specific to your question, Federal Circuit case law does not
`
`require that simple tasks that could be performed by the skilled artisan be
`
`explained.
`
`
`
`
`
`And I note, and this is in our briefing, that the '319 patent itself
`
`doesn't address any of these issues, other than the bandwidth that we've
`
`already talked about. So if these tasks were insurmountable that would lead
`
`to a lack of reasonable expectation of success, as in you can't do it, then the
`
`'319 patent should've addressed those issues, because at the end of the day,
`
`all Claim 46 is reciting is a clustered processor architecture that happens to
`
`have the data of the acoustic model stored. In other -- oh, sorry. Go ahead.
`
`
`
`
`
`JUDGE OGDEN: Couldn't it be that the patent at issue is just
`
`not enabled? And then likewise, Petitioner's proposed combination would
`
`not have been able to have been performed by a person of level of ordinary
`
`skill in the art. And obviously we don't decide the question of whether the
`
`challenged patent, the claims of the challenged patent are enabled, but it
`
`doesn't seem to follow necessarily that just because the challenged patent
`
`1
`
`2
`
`3
`
`4
`
`5
`
`6
`
`7
`
`8
`
`9
`
`10
`
`11
`
`12
`
`13
`
`14
`
`15
`
`16
`
`17
`
`18
`
`19
`
`20
`
`21
`
`22
`
`23
`
`24
`
`25
`
`
`
`
`
`20
`
`
`
`

`

`1
`
`2
`
`3
`
`4
`
`5
`
`6
`
`7
`
`8
`
`9
`
`10
`
`11
`
`12
`
`13
`
`14
`
`15
`
`16
`
`17
`
`18
`
`19
`
`20
`
`21
`
`22
`
`23
`
`24
`
`25
`
`26
`
`IPR 2023-00033
`Patent 7,587,319 B2
`
`doesn't describe things in that way, that necessarily that there's a reasonable
`
`expectation of success.
`
`
`
`
`
`MS. BAILEY: So o