Paper No. 33
`
`Trials@uspto.gov
`571-272-7822
`UNITED STATES PATENT AND TRADEMARK OFFICE
`
`
`BEFORE THE PATENT TRIAL AND APPEAL BOARD
`
`
`APPLE INC.,
`Petitioner,
`
`v.
`
`ZENTIAN LIMITED,
`Patent Owner.
`
`
` IPR2023-00036
`Patent 10,839,789 B2
`
`
`Record of Oral Hearing
`Held: March 12, 2024
`
`
`
`
`
`Before: KEVIN F. TURNER, JEFFREY S. SMITH, and
`CHRISTOPHER L. OGDEN, Administrative Patent Judges.
`
`
`
`
`
`
`
`
`
`
`Trials@uspto.gov
`571-272-7822
`UNITED STATES PATENT AND TRADEMARK OFFICE
`
`
`BEFORE THE PATENT TRIAL AND APPEAL BOARD
`
`
`APPLE INC.,
`Petitioner,
`
`v.
`
`ZENTIAN LIMITED,
`Patent Owner.
`
`
` IPR2023-00036
`Patent 10,839,789 B2
`
`
`Record of Oral Hearing
`Held: March 12, 2024
`
`
`
`
`
`Before: KEVIN F. TURNER, JEFFREY S. SMITH, and
`CHRISTOPHER L. OGDEN, Administrative Patent Judges.
`
`
`
`
`
`
`
`
`
`
`IPR2023-00036
`Patent 10,839,789 B2
`
`
`
`APPEARANCES:
`
`ON BEHALF OF THE PETITIONER:
`
`
`JENNIFER BAILEY, ESQUIRE
`Erise IP
`7015 College Blvd. Ste. 700
`Overland Park, KS 66211
`Jennifer.bailey@eriseip.com
`(913) 777-5600
`
`
`ON BEHALF OF THE PATENT OWNER:
`
`
`KAYVAN NOROOZI, ESQUIRE
`Noroozi PC
`11601 Wilshire Blvd., Ste. 2170
`Los Angeles, CA 90025
`Kayvan@noroozipc.com
`(310) 972-7074
`
`PETER KNOPS, ESQUIRE
`peter@noroozipc.com
`
`
`The above-entitled matter came on for hearing on March 12, 2024,
`commencing at 4:25 p.m., via video teleconference.
`
`
`
`
`
`
`
`
`
`
`2
`
`(404) 684-6008
`
`Jamison Professional Services
`East Pointe, GA
`
`www.jps-online.com
`
`
`
`IPR2023-00036
`Patent 10,839,789 B2
`
`
`P R O C E E D I N G S
`- - - - -
`JUDGE TURNER: Good afternoon. This is an oral hearing for
`IPR2023-00036 involving U.S. Patent 10839789. I am Judge Turner, joined
`by Judges Ogden and Smith. All the provisos that I provided in the earlier
`hearings still apply to this hearing. Petitioner will have a total of 45 minutes
`to present its argument, and Patent Owner will have 45 minutes to present its
`opposition. Petitioner will go first to present its case. Thereafter, Patent
`Owner will present its opposition to Petitioner’s case.
`If there’s any rebuttal from Petitioner, we will hear it after Patent
`Owner’s opposition, and finally we will hear Patent Owner’s surrebuttal, if
`requested. Petitioner, if you’d please -- if you will go to the podium,
`indicate your appearances, and also provide a time for rebuttal you may wish
`to reserve?
` MS. BAILEY: Good afternoon, Your Honors. My name is
`Jennifer Bailey from the law firm of Erise IP, representing Petitioner, Apple,
`Inc. Here with me today is my co-counsel, Cristina Canino, and in-house
`counsel from Apple, Jenny Liu.
`
`JUDGE TURNER: And did you want to reserve any time?
` MS. BAILEY: So sorry, Your Honors. I reserve 15 minutes.
`Thank you.
`JUDGE TURNER: Patent Owner’s counsel, please?
`
` MR. NOROOZI: Yes, Your Honors. Kayvon Noroozi from
`Noroozi PC for Patent Owner, Zentian. Along with me, Mr. Peter Knops, as
`well as Ms. Jessica Bernhardt, from the law firm of Bartlit Beck.
`3
`
`
`
`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
`
`(404) 684-6008
`
`Jamison Professional Services
`East Pointe, GA
`
`www.jps-online.com
`
`
`
`IPR2023-00036
`Patent 10,839,789 B2
`
`JUDGE TURNER: And did you want to indicate if you wanted
`to reserve any time?
` MR. NOROOZI: Yes, Your Honor. Fifteen minutes as well.
`Thank you.
`JUDGE TURNER: Understood. And last but not least?
`
` MR. CHURNET: Hello, Your Honors. Dargaye Churnet from
`Fenwick & West, representing Amazon.
`
`JUDGE TURNER: Thank you very much. We are fully on the
`record. Please, Petitioner, you may start whenever you wish.
` MS. BAILEY: May it please the Board. Thank you, Your
`Honors. Let’s turn first to Petitioner’s demonstratives, DX2. So the claims
`for the `789 Patent, and specifically Claim 1, are quite broad here. We
`simply recite a -- excuse me. The claim simply recites a calculating
`apparatus which calculates the distances and the acoustic model memory,
`and both are fabricated in a single integrated circuit. Turning to DX3, ICs,
`at the time of the `789 Patent, were extremely well known and ubiquitous. I
`refer the Board to our Petitioner reply, pages 2, citing Exhibit #1003,
`paragraph 70 through 76, and the petition, pages 42 through 44. I’ve listed
`here a couple of the references that were discussed, specifically at Exhibit
`#1003, paragraphs 70 through 76.
`Turning to DX4, I want to talk about the motivations to combine.
`So once again, Zentian does not dispute that the combination of
`Smyth/Mozer teaches all of the limitations. There is overwhelming evidence
`of the benefits of a single integrated circuit having a processor and memory,
`and Zentian does not actually dispute this, either. First of all, we have a
`
`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
`
`(404) 684-6008
`
`Jamison Professional Services
`East Pointe, GA
`
`www.jps-online.com
`
`
`
`IPR2023-00036
`Patent 10,839,789 B2
`known technique, which under the Intel v. Pact case, a known technique and
`a suitable option is sufficient for a motivation to combine. And the known
`technique, just for clarity, is a processor and a memory on a single IC. And
`here the memory is simply an acoustic model memory, so it’s storing a
`particular type of data.
`Zentian’s rebuttal is that none of the art shows an acoustic model
`memory and a processor on a single IC, and this is incorrect. Mozer of
`course discloses it on the basis of the combination, but there are background
`references of record that also disclose it. Specifically, Dr. Anderson actually
`admits that Nguyen, which is Exhibit #1047, teaches an integrated circuit
`that has an acoustic model memory, storing part of the acoustic model and a
`processor.
`I’ll refer the Board to the sur reply, Zentian’s sur reply, page 11,
`citing Exhibit #1069, which is page 92, lines 12 through 21. Also, the
`Toyoda reference discussed at Exhibit #1003, paragraphs 73 through 74,
`also disclosed an acoustic model memory, and an acoustic processor or a
`calculating apparatus on the same IC.
`
`JUDGE SMITH: For the Toyoda reference, I think Patent
`Owner’s point is that it’s not an IC. Toyoda doesn’t say it’s an IC. How are
`you coming to the conclusion that it’s on an IC?
` MS. BAILEY: Mr. Schmandt, in his declaration at paragraphs
`73 and 74, opined that he would understand as a skilled person that it was on
`a single IC, and Zentian does not provide any rebuttal evidence to that.
`Turning to DX5, Dr. Anderson admitted the known advantages of a memory
`and a processor on a single IC. Dr. Anderson admitted that the Motorola
`
`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
`
`
`
`5
`
`(404) 684-6008
`
`Jamison Professional Services
`East Pointe, GA
`
`www.jps-online.com
`
`
`
`IPR2023-00036
`Patent 10,839,789 B2
`DSP that’s discussed in Smyth had memory on chip, “so could do operations
`more quickly.” And we have the citation from his transcript there on DX5.
`
`JUDGE SMITH: I think just a couple of points. Just real
`quickly, for the testimony about Toyoda.
` MS. BAILEY: Yes?
`
`JUDGE SMITH: I’m looking at paragraph 74 of the
`declaration, and I see where he’s quoting from Toyoda, and then I see where
`he comes to the conclusion. He basically quotes what Toyoda says. Toyoda
`has the memory and the processor, and then he says Figure 7 shows this.
`And then he just comes to this conclusion, therefore it’s on an integrated
`circuit. But looking at his testimony, I’m not clear how, you know, just
`Figure 7 of Toyoda means it’s on an integrated -- I mean, Toyoda itself
`doesn’t say it’s on an integrated circuit, and the evidence that he’s relying on
`doesn’t -- at least in my mind, doesn’t show it’s on an integrated circuit.
`Can you help me out here?
` MS. BAILEY: Sure. So Toyoda discloses the sensor signal
`processor 31, and from that -- which, by the way has -- excuse me --
`includes the processor for performing the distance calculations, along with
`the acoustic model memory. Toyoda does teach that. And from that, Mr.
`Schmandt opined that he would understand that the sensor signal processor,
`because it includes both the memory and the, well, processor, that it would
`be on a single IC. And Judge Smith, I appreciate your questions. I do
`remind the Board that they can take into account the opinions of a skilled
`person.
`
`
`
`JUDGE SMITH: Okay. So I guess you’re basing it on the fact
`
`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
`
`
`
`6
`
`(404) 684-6008
`
`Jamison Professional Services
`East Pointe, GA
`
`www.jps-online.com
`
`
`
`IPR2023-00036
`Patent 10,839,789 B2
`that it’s called a signal processor, and so signal processor means an
`integrated circuit? Can you have a signal processor in an integrated circuit?
` MS. BAILEY: In addition to the fact that the single processor
`includes the memory, so at that point it would be more akin to a DSP
`processor, where you have both the memory and the processor on a single
`IC.
`
`JUDGE SMITH: Okay. Thank you. And then I guess to the
`
`point about, you know, Mr. Anderson’s, or Dr. Anderson’s testimony, you
`know, Patent Owner is saying Dr. Anderson is getting these advantages from
`the challenged Patent. You know, to the extent that the motivation to
`combine comes from the challenged Patent, that’s not something that we can
`really rely on. Is there evidence of these increased efficiencies that come
`from somewhere other than the challenged Patent?
` MS. BAILEY: Yes, Your Honor, and yes, I would agree that
`you can’t get the motivations from the challenged Patent itself. I want to
`address Dr. Anderson’s testimony, and then I also want to point you again to
`Exhibit #1003, paragraphs 70 through 76, that discuss ICs, and then I’m
`going to also bring up some additional. But going back to Dr. Anderson’s
`testimony itself, both on DX5 and DX6, we have his admissions about the
`advantages.
`Now, Zentian’s rebuttal to Dr. Anderson’s testimony is that he was
`referring to the challenged Patent, but if you look at DX6, you will see that
`the lead up to the questioning, Dr. Anderson actually said himself in his own
`answer that he was referring to the prior art. So referring to DX6, he was
`asked, was it known to make more efficient, having an on-chip memory for
`
`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
`
`(404) 684-6008
`
`Jamison Professional Services
`East Pointe, GA
`
`www.jps-online.com
`
`
`
`IPR2023-00036
`Patent 10,839,789 B2
`memory transfers? And he says, so in the prior art that I’ve seen presented
`here and goes on to talk about that we just have to make a distance calculator
`that has all the memory on chip. And then he has the last paragraph, where
`he makes the admission that it was more efficient to reduce the off chip
`memory access, so Dr. Anderson in this admission was discussing the prior
`art.
`
`I want to now move to DX7, and I want to discuss the ground of
`Smyth and Mozer, specifically the Claim 1E limitation that is in dispute. So
`the petition presented two mappings for Claim 1E. The first was that Smyth
`alone teaches or renders obvious Claim 1E limitation, and the second was
`that Smyth, in combination with Mozer, renders obvious, and I’ll refer -- go
`ahead, Your Honor. Oh. Sorry. And I refer the Board to Exhibit #1003,
`paragraphs 167 through 168.
`So in Mozer -- and we have Figure 4 from Mozer -- Mozer teaches
`an audio recognition peripheral IC 400. That’s the big orange box on Fig 4.
`This audio peripheral IC includes the vector processor 423, which would be
`akin to the calculating apparatus as recited in the claims, and vector
`processor 423 calculates the distances. There’s also an on chip acoustic
`model memory 460, that stores a portion of the acoustic model.
`So in the mapping, Mozer is only used to show having a
`calculating apparatus and an acoustic model fabricated on an IC, and the
`mapping says that it would be obvious to fabricate Smyth’s vector processor,
`or processor 341 -- Smyth refers to it as a classifying processor, to be
`precise. It would be obvious to fabricate Smyth’s classifying processor 341
`and acoustic model memory 342 on a single IC. And this is discussed at the
`
`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
`
`
`
`8
`
`(404) 684-6008
`
`Jamison Professional Services
`East Pointe, GA
`
`www.jps-online.com
`
`
`
`IPR2023-00036
`Patent 10,839,789 B2
`petition, pages 41 through 44.
`
`JUDGE SMITH: I want to ask you about the memory in
`Mozer. So you’re calling it an acoustic model memory, but the claim itself
`says that the acoustic model memory defines the plurality of acoustic states.
`In Mozer, the memory in Mozer, it just loads -- it’s loading a first vector
`representing the audio signal, and a second vector representing the template
`into the memory. So I mean, to the extent that the claim says a plurality of
`acoustic states, in the memory of Mozer, there are two vectors. One is from
`the input speech, and the other is from the model memory.
` MS. BAILEY: Yes, Your Honor.
`
`JUDGE SMITH: How does the one vector from the model
`meet the limitation of a plurality of acoustic states?
` MS. BAILEY: Yes, Your Honor. We are not relying on
`Mozer’s acoustic model memory to teach Claim 1B of an acoustic model
`memory for storing an acoustic model. The petition makes clear that it is
`relying on Smith’s memory 342 to teach the acoustic model memory. As I
`previously said, we are only relying on Mozer for the obviousness of putting
`an acoustic processor, or a calculating apparatus and an acoustic model
`memory on a single IC.
`
`JUDGE SMITH: But I guess it doesn’t have an acoustic model
`memory, according to the claim. I guess that’s what I’m struggling with.
`Where is the acoustic model memory, if the acoustic model memory requires
`a plurality of acoustic states?
` MS. BAILEY: I think --
`
`JUDGE SMITH: Mozer doesn’t have that.
`
`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
`
`
`
`9
`
`(404) 684-6008
`
`Jamison Professional Services
`East Pointe, GA
`
`www.jps-online.com
`
`
`
`IPR2023-00036
`Patent 10,839,789 B2
` MS. BAILEY: Mozer does have an acoustic model memory.
`
`JUDGE SMITH: Where?
` MS. BAILEY: A plurality of states are not stored on it. I agree
`with that. But that’s where --
`
`JUDGE SMITH: Okay. Okay.
` MS. BAILEY: -- Smyth comes in. Smyth does have an
`acoustic model memory that has a plurality of states. And so that is why the
`combination relies on Smyth’s acoustic model memory to be stored on a
`single IC, per Mozer’s teachings.
`
`JUDGE SMITH: But for Mozer’s teachings --
` MS. BAILEY: (CROSSTALK) -- oh, I’m so sorry.
`
`JUDGE SMITH: Oh, go ahead. Go ahead. Go ahead.
` MS. BAILEY: Because if Mozer taught an acoustic model
`memory that stored the entirety of the acoustic model or a plurality of states,
`then Mozer would be an anticipating reference.
`
`JUDGE SMITH: Okay. So --
` MS. BAILEY: And this is an obviousness combination.
`
`JUDGE SMITH: I understand. So what is the teaching of
`Mozer that suggests putting the entire memory of -- the entire acoustic
`model memory of Smyth on to the chip?
` MS. BAILEY: Do you want to -- are you asking what is the
`motivation to combine? Because I can move to that, if that would be
`helpful. I’m sorry.
`
`JUDGE SMITH: If you’re going to get to it later -- if you’re
`going to get to it later, that’s fine. I’ll wait.
`
`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
`
`
`
`10
`
`(404) 684-6008
`
`Jamison Professional Services
`East Pointe, GA
`
`www.jps-online.com
`
`
`
`IPR2023-00036
`Patent 10,839,789 B2
` MS. BAILEY: I do plan on -- I’m actually getting to it in the
`next slide.
`JUDGE SMITH: Okay.
`
` MS. BAILEY: So how about we move to that? Okay. So
`moving to DX8, let’s talk about the motivations to combine of Smyth and
`Mozer. So in Mozer, the audio recognition peripheral 400 includes the
`acoustic model memory on the embodiment -- excuse me. Includes an
`acoustic model memory on chip in the embodiments at Figs 3 and 4. At the
`top of column 9, which is at the top of our DX8 slide, Mozer says that the
`embodiments are illustrative of advantages of providing the audio
`recognition peripheral on one IC, so on one integrated circuit. And
`remember, the audio recognition peripheral includes both the calculating
`apparatus, which Mozer calls a vector processor, and the acoustic model
`memory.
`
`Mozer also teaches -- and this is the second section of text on
`DX8 -- of offloading burdensome vector processing as another stated
`advantage, and this was relied on in the petition and the declaration. Now,
`with respect to this particular advantage, Zentian has argued that Smyth
`already achieves offloading of burdensome vector processing, because
`Smyth already has a separate DSP 341 for performing distance calculations.
`So I have two responses to that. First, we actually addressed this
`in the petition somewhat, at page 42. We said the combination would have
`likewise captured the advantages stated in Mozer. Specifically, having a
`distinct processor perform the distance calculations achieves a benefit of
`offloading burdensome vector processing, in the same way as the benefits of
`
`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
`
`(404) 684-6008
`
`Jamison Professional Services
`East Pointe, GA
`
`www.jps-online.com
`
`
`
`IPR2023-00036
`Patent 10,839,789 B2
`Mozer. But in addition, having the processor and the memory on chip
`achieves the known technique with a predictable solution of more efficient
`memory transfers.
`So we have a second benefit of being able to offload burdensome
`vector processing when you have the memory and the process on chip. And
`I again refer back to Mozer’s -- the two Zs caught me -- Mozer’s teaching at
`column 9, line 15 through 21, that the embodiments, which includes the
`embodiments of the processor and memory on a single IC, of providing a
`recognition peripheral system.
`Additionally, Mozer talks about the advantage of avoiding a CPU
`hold, and this is at DX9. Mozer teaches that there is an advantage of
`avoiding a CPU hold when the memory and the processor are on a single IC.
`In response to the petitioner reply arguments, which were provided at pages
`17 through 19, Zentian is not responding to those arguments. And I kind of
`want to go through this, because Zentian is responding to a different
`architecture than what we are discussing relative to the no CPU hold
`advantage.
`So Zentian argues that there needs to be a CPU hold when there’s
`two processors accessing a single memory, which is what the architecture is
`of Mozer. And we agree with that, but that’s not the only architecture in
`which a CPU hold, or avoiding a CPU hold would be advantageous. A CPU
`hold would also be advantageous -- I should say a no CPU hold would also
`be advantageous when a first processor is accessing two memories, and then
`you have a second processor pulling data and instructing the first processor
`and pulling from a second memory.
`
`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
`
`
`
`12
`
`(404) 684-6008
`
`Jamison Professional Services
`East Pointe, GA
`
`www.jps-online.com
`
`
`
`IPR2023-00036
`Patent 10,839,789 B2
`So here we have the figures of Smyth to show the Smyth
`architecture, and why the Smyth architecture would benefit from avoiding a
`CPU hold. Looking on the left, the distance calculations are performed at
`the processor 341, and then we have the acoustic model memory at 342.
`The parsing processor, which is in the right hand fig 351, is mapped as a
`CPU, per Claim 6 -- excuse me, per Claim 9, and the parsing processor
`instructs processor 341. The parsing processor is sending instructions to
`341, which is a DSP, to write distance calculations to memory 353. So DSP
`341 is handling a lot of tasks. It’s pulling from acoustic model memory. It’s
`writing to memory 353. It’s performing distance calculations, and it’s
`responding to the instructions from 351.
`DSP 341 will face a resource contention issue, given all of those
`steps that it must perform, and we don’t want a situation where DSP 341 is
`interrupted in the middle of a memory operation by CPU 351. Otherwise,
`the system ends up with a half-completed operation and a potentially
`unstable condition. So DSP 341 needs to be able to tell CPU 351 that 351
`cannot write to memory 353 while it is accessing from memory 342 or
`performing distance calculations. In contrast, when memory 342 is on chip,
`then we only have processor 341 accessing one memory, and therefore a
`CPU hold would not be needed.
`And I want to also turn to DX10 and note testimony by Mr.
`Schmandt during his deposition where he pointed out that there’s only one
`set of memory lines. And as discussed in our Petitioner reply and the
`briefing, we also discuss that there is a bus in between 341 and 351. So I’m
`going to pause there. I know I just went through a lot of numbers. The
`
`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
`
`
`
`13
`
`(404) 684-6008
`
`Jamison Professional Services
`East Pointe, GA
`
`www.jps-online.com
`
`
`
`IPR2023-00036
`Patent 10,839,789 B2
`takeaway from the Smyth architecture is that there would be an advantage of
`avoiding a CPU hold given the Smyth architecture, if memory 342 was on
`chip with 341, the DSP.
`
`JUDGE SMITH: And just real quickly again, I know you
`mentioned offloading. Can you explain one more time real quickly why the
`advantage of offloading from Mozer would apply to Smyth, given that
`Smyth already offloads the --
` MS. BAILEY: Yes, Your Honor. You would have the
`additional advantage that when the acoustic model memory of Smyth 342 is
`on chip with DSP 341, that you then avoid the memory transfers or memory
`access that you would need to have from 341 to memory 342. It is well
`known that on chip memory access is going to be more efficient than having
`to go off chip for memory access.
`
`JUDGE SMITH: Okay. Thank you.
` MS. BAILEY: I want to next turn to DX11, and let’s talk about
`the Smyth alone teaching or rendering obvious Limitation 1E. So Zentian’s
`argument is that Smyth alone cannot render obvious Claim 1E, because
`Smyth’s memory size for the Motorola DSP is not large enough to store a
`large vocabulary model. So Smyth teaches that processor 341 can be a
`specific Motorola DSP that was commercially available at the time.
`So if you would like, we can turn back to DX2 with the claim
`language. Claim 1 only recites an acoustic model defining a plurality of the
`acoustic space, which Judge Smith, you just discussed is arguably met by
`two states. There is no memory size required in this limitation. There’s no
`boundaries or definition for the size of the acoustic model being stored. And
`
`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
`
`
`
`14
`
`(404) 684-6008
`
`Jamison Professional Services
`East Pointe, GA
`
`www.jps-online.com
`
`
`
`IPR2023-00036
`Patent 10,839,789 B2
`Dr. Anderson, turning back to DX11, admitted that the claims do not recite a
`minimum memory size for the acoustic model.
`
`JUDGE SMITH: Just real quickly, to the point about its met by
`two states, two acoustic states, I actually put that in the decision institute.
`But from reading the subsequent briefing from the parties, it seemed like
`neither party actually said one way or the other whether they agree with that.
`It seems like the Patent Owner, from what I understand, is saying that the
`acoustic memory has to store a large model memory, and from what I
`understand, your briefs, the acoustic model memory has to store maybe a
`limited vocabulary to recognize digits. You know, what we said in the DI
`about these two states, did either party address that, or agree with it or
`disagree with it?
` MS. BAILEY: Your Honor, neither party addressed it. And
`the fact of whether plurality is met by two states or requires three states or
`four states is really not at issue here, because of the preface to your question,
`and it’s what I’m going to get to with the memory size. Smyth teaches --
`and I think my answer will answer your question, but please, Judge Smith,
`let me know if it does not.
`Smyth teaches a digit’s recognizer that’s recognizing digits like
`one, two, three, four. Dr. Anderson admitted that he assumed the claims
`recite a large language model. That’s how he based his opinions. But then
`he also admitted on DX12 that Smyth had an abbreviated model. I would
`also point out that Counsel just admitted in the last IPR that we had a
`hearing on for the `277, that Smyth has an “extremely small vocabulary,”
`and “not much computation to run.” And that’s because it’s a digit’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
`
`
`
`15
`
`(404) 684-6008
`
`Jamison Professional Services
`East Pointe, GA
`
`www.jps-online.com
`
`
`
`IPR2023-00036
`Patent 10,839,789 B2
`recognizer. So to Zentian’s argument that the Motorola DSP would not have
`a sufficient memory size to hold an acoustic model of a digit’s recognizer,
`Zentian has no support for that.
`
`JUDGE SMITH: But I think it’s important that the available
`memory is about one and a half kilobytes, and the minimum memory
`needed, even accepting that Mr. Schmandt’s testimony is somewhere on the
`order of four or five kilobytes.
` MS. BAILEY: Yes, but that four or five kilobytes by Mr.
`Schmandt was for a large vocabulary model. It was not for a digit’s
`recognizer. And we disagree -- oh, sorry. Go ahead.
`
`JUDGE SMITH: Did Mr. Schmandt provide testimony about
`the memory requirements of a digit’s recognizer?
` MS. BAILEY: Mr. Schmandt did testify that it would be far
`less, I believe, in his deposition, and I can get you that citation on rebuttal.
`
`JUDGE SMITH: Okay.
` MS. BAILEY: Far less than would be for a large vocabulary
`language model. But then I will say, Your Honors, if we start talking about
`the memory size of the acoustic model, then we’re starting to import
`limitations into the claim, because the claim simply recites a plurality of
`states.
`
`JUDGE SMITH: Well, I mean, I think it’s significant to the
`
`extent that the memory has to store a plurality of states, and even Mozer
`itself is only storing one vector, and Mozer is using 4K to do it. So I mean,
`it’s relevant, to the extent that the memory -- I understand your point about
`importing limitations into the claim, but the on chip memory has to be big
`
`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
`
`
`
`16
`
`(404) 684-6008
`
`Jamison Professional Services
`East Pointe, GA
`
`www.jps-online.com
`
`
`
`IPR2023-00036
`Patent 10,839,789 B2
`enough to store at least two states.
` MS. BAILEY: And Zentian has not disputed that 1.5 kilobytes
`would not be large enough to store a digit’s recognizer. It cannot, Your
`Honor. That would be an absurd assertion by Zentian. I would like to use
`the remaining time for my rebuttal, please, if there are no further questions.
`Thank you, Your Honors.
`
`JUDGE TURNER: Okay. You will have 19 minutes.
` MR. NOROOZI: Thank you, Your Honors. Kayvon Noroozi
`for Patent Owner and may it please the Board. I just want to state at the
`outset that there was a motivation articulated, it seemed based on an alleged
`memory contention problem that would arise in Smyth, as well as due to the
`bus placement in Smyth. Both of those are new arguments, and they were
`not raised before, and they did not appear in the briefing.
`With respect to the bus, the bus that was shown -- if we could pull
`up the figure from Smyth? It’s in our slides at Slide 4. The bus -- oh, I’m
`sorry. We’re waiting for the audio visual to become available. But I’ll note
`it for the record, the bus will be in the same place, regardless of the
`combination that they’re proposing. They’re proposing to put the memory
`342 and the processor 341 on a single IC. That won’t affect the bus. The
`bus is external. It’s next to the processor, I think as you can see from the
`figure.
`
`With respect to where we just left off in Petitioner’s argument
`regarding Smyth itself, Judge Smith, to your questions, if we could pull up --
`you’re exactly right about the point of the one and a half kilobytes versus the
`minimum memory requirement that Mr. Schmandt himself articulated for
`
`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
`
`
`
`17
`
`(404) 684-6008
`
`Jamison Professional Services
`East Pointe, GA
`
`www.jps-online.com
`
`
`
`IPR2023-00036
`Patent 10,839,789 B2
`what would need to be put into the memory, and he said that it would require
`at least four to five kilobytes, which obviously would not fit.
`But Counsel for Petitioner said, well, Mr. Schmandt was just
`talking about a large vocabulary model, not Smyth’s really simple, really
`small, digit-based model. That’s inaccurate, Your Honors. So we can look
`at Mr. Schmandt’s second declaration at paragraph seven, and right there,
`we can see right there, it says in the first sentence regarding the memory size
`requirements for Smyth’s tristate allophone model. Right?
`And then he does say if the model were for large vocabulary
`speech recognition, but if we keep going, we see that his analysis, at the end
`of the day, the number that he gives us is based off of Smyth’s model. And
`he never provides any other acoustic model size that would be adequate to
`meet the claim limitations, and would also be small enough to fit into the
`memory, the one and a half kilobytes in Kloker. So there is just no evidence
`to support Petitioner’s assertion that surely there would be something that
`would be small enough to meet the claims and fit into the one and a half
`kilobytes, and I think Judge Smith, as you pointed out, even Mozer teaches
`that a single vector is four kilobytes.
`So with respect to Smyth alone, if we look at our slide four here,
`the evidence establishes that Smyth’s memory 342 simply cannot be already
`on board the processor 341, because if processor 341 is the Motorola
`DSP56000, it only has one and a half kilobytes of memory that could even
`theoretically be used for storing data structures. And Mr. Schmandt himself
`says that the acoustic model of Smyth would require 4,800 bytes to 5,500
`bytes, which would not fit. And then if we move on --
`
`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
`
`(404) 684-6008
`
`Jamison Professional Services
`East Pointe, GA
`
`www.jps-online.com
`
`
`
`IPR2023-00036
`Patent 10,839,789 B2
`
`JUDGE TURNER: Counsel, just a quick question. Does
`Smyth say that the processor 341 is limited to only that particular model?
` MR. NOROOZI: That’s an interesting question, Your Honor.
`
`JUDGE TURNER: Thank you.
` MR. NOROOZI: Because I say that in the sense that no one
`has -- that Petitioner has not argued that Smyth could be used to run a
`different model, so there’s no evidence to suggest that Smyth is useful for
`some other model besides what’s been discussed in Smyth.
`
`JUDGE TURNER: No, I understand, but you’re making the
`calculations on that basis, so I’m trying to -- my own recollection of whether
`it’s limited for the disclosure of Smyth, and therefore -- but if it’s not, if it’s
`just exemplary, then in theory, we could have an updated DSP to serve as --
`in 341, that maybe has an extended memory. So I’m trying to tease out if
`it’s limited, you can make the assertion that it’s limited, but I wanted to --
`you’re more than welcome to say it’s beyond what’s been discussed, and it’s
`not in the record, and you don’t want to respond. That’s fine.
` MR. NOROOZI: No. Of course, I always want to respond if I
`can, Your Honors, especially based on the record and evidence. What I
`don’t want to do is speculate.
`
`JUDGE TURNER: Mm-hmm.
` MR. NOROOZI: I misunderstood your question initially, and I
`think I understand it now. I think your question is could the processor 341
`be something different than the Motorola DSP56000? Is that the question?
`
`JUDGE TURNER: Yes, that’s my question.
` MR. NOROOZI: Okay. And the answer to that is I’m not
`
`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
`
`
`
`19
`
`(404) 684-6008
`
`Jamison Professional Services
`East Pointe, GA
`
`www.jps-online.com
`
`
`
`IPR2023-00036
`Patent 10,839,789 B2
`completely sure, but I will say that Petitioner itself relied on the document
`Kloker, Exhibit I think #1011, as the evidence for the capabilities of
Accessing this document will incur an additional charge of $.
After purchase, you can access this document again without charge.
Accept $ Charge
Still Working On It
This document is taking longer than usual to download. This can happen if we need to contact the court directly to obtain the document and their servers are running slowly.
Give it another minute or two to complete, and then try the refresh button.
A few More Minutes ... Still Working
It can take up to 5 minutes for us to download a document if the court servers are running slowly.
Thank you for your continued patience.
This document could not be displayed.
We could not find this document within its docket. Please go back to the docket page and check the link. If that does not work, go back to the docket and refresh it to pull the newest information.
Your account does not support viewing this document.
You need a Paid Account to view this document. Click here to change your account type.
Your account does not support viewing this document.
Set your membership
status to view this document.
With a Docket Alarm membership, you'll
get a whole lot more, including:
- Up-to-date information for this case.
- Email alerts whenever there is an update.
- Full text search for other cases.
- Get email alerts whenever a new case matches your search.
One Moment Please
The filing “” is large (MB) and is being downloaded.
Please refresh this page in a few minutes to see if the filing has been downloaded. The filing will also be emailed to you when the download completes.
Your document is on its way!
If you do not receive the document in five minutes, contact support at support@docketalarm.com.
Sealed Document
We are unable to display this document, it may be under a court ordered seal.
If you have proper credentials to access the file, you may proceed directly to the court's system using your government issued username and password.
Access Government Site
