[00:00:01] Speaker 02: Our next case is number 242207, Apple Inc. v. Zentian Limited. [00:00:08] Speaker 02: Mr. Matsui. [00:00:12] Speaker 02: Stay at the same council table. [00:00:37] Speaker 01: May it please the court, Brian Matsui for Apple. I'd like to address the single integrated circuit issue and then claim 29. [00:00:45] Speaker 01: For the single integrated circuit claim requirement, the dispute really comes down to whether the prior art teaches or suggests placing a processor like the ones in Schmidt or elsewhere with an acoustic memory, what Zentian calls a dictionary of sounds. The board could only conclude that it didn't by imposing unclaimed requirements on how big an acoustic model must be. [00:01:11] Speaker 02: So they say it had to be a large language model, right? [00:01:14] Speaker 01: They did, yes, Your Honor. [00:01:16] Speaker 02: And you're arguing, among other things... [00:01:19] Speaker 02: that a digital recognition model would satisfy the claim, right? [00:01:26] Speaker 01: That's correct, Your Honor. [00:01:27] Speaker 02: So where do we find that the digital recognition model would take Missouri, for example, would fit within the Ford Celebrite embodiment of Missouri? [00:01:47] Speaker 01: I think that we would find that just based upon the fact that you're talking about a digit recognizer of a simple model of 1, 2, 3, 4. [00:01:55] Speaker 01: And so that's going to be much smaller than... Well, their experts said it was very small. Everybody said that this was very small and that they said that this was a very... The board basically didn't address this in the context of Moser for two reasons. One was because... It said that there was this fabrication requirement for experts, and two, because it said that we didn't raise this abbreviated model in a timely way. We didn't raise it in our petition, in our reply, which we did clearly at Appendix 367. [00:02:27] Speaker 01: So this was clearly before the board, but I think it's important to take a look at this from the perspective of This isn't bodily incorporation of one component with another component. You're talking about an obviousness analysis where you have these acoustic models, which can be very small, and you have integrated circuits, which are ubiquitous. And so a person of ordinary skill in the art would understand that you could combine these when there's no memory requirement for... [00:02:56] Speaker 01: the claims as far as how much memory must be on the integrated circuit or how large the acoustic model must be. So if the entire English language is 4.7 kilobytes, all the sounds in the English language are 4.7 kilobytes, which is what the board relied upon, then that's going to be much smaller when you're talking about just 1, 2, 3, 4. [00:03:16] Speaker 01: And under a preponderance of the evidence standard, that's something that clearly would show that you would be able to fit that small digit recognizer onto Moser's 4 kilobytes. [00:03:27] Speaker 00: How do we know everything that you're saying? How do we know that? [00:03:33] Speaker 00: Did that all come in through expert testimony? [00:03:35] Speaker 01: Yeah, I mean, it came in both from Smith. If we turn to Appendix 117, it talks about the digit recognizer. This is at column 9, and it says line 25, and it says allowing the recognition of a three-digit string, the recognizable digits being 1, 2, 3, 4. [00:03:58] Speaker 01: So Smith is talking about right here having a very small acoustic model. And then we have at Appendix 30, where the board found that Moser has 4 kilobytes of memory. We have the board's finding that the large acoustic model was 4.7 kilobytes, which is all the sounds of the English language. [00:04:22] Speaker 01: Now, there might be a dispute over whether it should be larger or smaller, but the board relied upon 4.7 kilobytes. And so then it's not a great leap. Had the board applied the right obviousness analysis to say that this small digit recognizer, 1, 2, 3, 4, would fit on the 4 kilobytes. And that's where the board went wrong. Because what it was requiring was... specific numbers and required specific calculations for acoustic models when there's nothing in Smith that dictates that you have to have an acoustic model of a particular size. [00:05:00] Speaker 01: And then there's nothing in the claims that require that you have to have a model of a certain size. It just says an acoustic model memory. [00:05:11] Speaker 02: And then there's nothing... What does the claim mean? It talks about a plurality. I mean, does that really mean that having two states would be sufficient? [00:05:23] Speaker 01: I mean, it says a plurality of states. I don't think what's at issue here is really the number of states that would be required. [00:05:29] Speaker 02: I think that it is in a sense. [00:05:32] Speaker 01: I think that it's not in the sense that I think that. [00:05:36] Speaker 01: Any digit recognizer would be more than two states because I think that a state would probably be either around an allophone or a phone. And so what you're talking about here is certainly maybe if you had a very, very simple acoustic model that was maybe speaking and not speaking or yes and no, that you might get down to two states. But what we're talking about here is is a digit recognizer in Smith. [00:06:04] Speaker 02: But what I'm asking you is how to construe the claim. I mean, your expert said that two states would be sufficient to satisfy the plurality requirement. Is that your position on appeal? [00:06:18] Speaker 01: It is, but I think that it is sufficient. [00:06:22] Speaker 01: But I think that we can show that more than two states would also work in integrated circuits, and I think that that's the argument that we're really pressing right here before the court, that we don't need to rely upon it just being two states. The fact that we can sort of meet the claim requirement more significantly than what the claim requires at the floor doesn't make our obviousness case any weaker. It just makes our obviousness case stronger. And that's basically the situation that we have here with respect to what the board did, where it basically didn't want to consider the small model because it said that it wasn't raised when it clearly was raised. [00:07:05] Speaker 01: It said that it was going to focus on one integrated circuit, the DSP 56,000. And given all that, this was a situation where the board was focusing on unclaimed features in the prior art and the obviousness analysis. And that's why it's something that would have to go back to the board at the very least to decide whether or not the small model would fit on something like Moser, which it recognized was for tool bytes under the preponderance of the evidence standard. [00:07:36] Speaker 01: And I'll just note also that one of the other rationales, one of the reasons why the board really didn't consider the Moser reference was because it imposed this fabrication requirement that a person of ordinary skill in the art had to be the one to actually fabricate the claims. And it did so based upon... It itself describes the fabrication. No, the claim doesn't describe anything about fabrication. Moser. Huh? Moser. What was that? Moser. Just... [00:08:06] Speaker 01: Discloses a fabricated chip, right? Yes. This is a fabricated chip. [00:08:12] Speaker 02: Everybody agreed that persons with ordinary skills... And the processor on the same chip, right? Integrated chip. [00:08:19] Speaker 01: Bozer discloses that, yes. [00:08:22] Speaker 01: So I don't – I think the board – Yes, I understand. And so the board, I think, was completely off base here by focusing on tenses. If I could turn quickly to claim 29 unless the court has any other questions. [00:08:36] Speaker 00: No, but when you follow up on 29, I mean you asked for reversal, I guess, and or remand, and you just mentioned – In your colloquy with Judge Dyke, why this feels more like a necessary remand than an outright reversal. So you can address that, and that same question would apply to Claim 29. [00:08:54] Speaker 01: certainly your honor i i think that when you look at the board's analysis here on claim 29 it basically imposed a memory requirement of an acoustic model memory that doesn't exist simply because we pointed to the classifier 34 in smith which has the processor but under axonics and other precedent from this court When you're pointing to different components that are going to be combined, you don't try to preserve things like the acoustic model memory that would necessarily be on that classifier. [00:09:26] Speaker 01: There's nothing in the claim requirement that says that you have to have the acoustic model memory, and that's what the board effectively imposed here. I think that given the fact that... [00:09:39] Speaker 01: Given the fact that Zentschian didn't even challenge the obviousness of Claim 29, once the board's rationale goes away and all the claims upon which Claim 29 depends upon... Wait, wait, wait. [00:09:52] Speaker 02: How does the rationale go away? I mean, it seemed to me there are three grounds on Claim 29 that the board adopted. One is the same as for the other claims, and it stands or falls with that. But it also said that there was a requirement for a results memory on the chip and a buffer memory on the chip. And you argue, and perhaps that's correct, that there's no requirement to claim that the buffer memory be integrated. But there is a requirement that the results memory be integrated, correct? [00:10:23] Speaker 01: Yes, Your Honor. But all those dependent claims were found unpatentable by the board. And so given that, there's nothing that's you know, there's nothing that would sort of make these claims patentable except for Claim 29. [00:10:41] Speaker 01: And what you're talking about with results memory and I think the small bites in Claim 25 are, like, incredibly small. So something in Boykie would clearly have the ability to say that. Now, certainly I would understand that the board – The board didn't provide us any opportunity to respond. [00:11:00] Speaker 02: If we went that direction, we don't have to get into the whole FanDuel issue and what that means, right? [00:11:05] Speaker 01: That's true. I mean, I think that our main FanDuel issue is that the board came up with this novel argument with respect to the acoustic model memory. And so that would make it – we didn't have any opportunity to respond under the APA to that issue. to that sort of rationale the board had, and that's the APA violation. And FanDuel itself recognized that if the board comes up with sort of a novel non-obviousness argument, then that could be an APA violation, and I think that that's what we have here, because it required us, with respect to the acoustic model memory, [00:11:38] Speaker 02: But to get a reversal, your APA violation is not going to help you. [00:11:42] Speaker 01: Our APA violation won't help us there, but I think that it certainly, to the extent the court has concerns about the results memory. [00:11:50] Speaker 00: Another reason for remand, while we're talking about remand and reversal, did the other side preserve the argument? My understanding that their response to everything in this case, including claim 29, but it may be wrong, only dealt with the first issues that he applied to Claim 1 as well as Claim 29. But I didn't recall that they made a separate argument with respect to the other stuff in Claim 29. Am I right about that? [00:12:16] Speaker 01: My recollection is that they did not make any argument with respect to Claim 29 except the acoustic model memory and responding to the APA argument, that they didn't address anything from Claim 18 or Claim 25, which Claim 29 depends from. So those were issues that I don't believe they raised in this court. [00:12:37] Speaker 01: And they certainly didn't raise them before the board since they didn't contest unpatentability there. [00:12:46] Speaker 02: You're saying that they didn't raise the issue about the results memory in this court in their brief? [00:12:53] Speaker 01: When I looked at it, I don't recall seeing that. I can check again and correct myself in rebuttal if I'm wrong about that. [00:13:00] Speaker 02: Okay. [00:13:12] UNKNOWN: Ms. [00:13:13] Speaker 02: Rose? [00:13:23] Speaker 03: May it please the court, Catherine Rhodes for Zentschian. [00:13:27] Speaker 02: The Patent Trial and Appeal Board did not... In that last point before we forget about it, does the red brief argue that there's a results memory requirement in Claim 29 that's not satisfied? [00:13:40] Speaker 03: No, it's that claim 29 rises and falls for the same reason as claim 1, in which claim 29, Apple proposed a six-reference combination for claim 29. That claim requires an acoustic coprocessor and an additional processor on an integrated circuit. The coprocessor has the acoustic model memory and calculating apparatus. That's found in claim 10. [00:14:05] Speaker 02: So 29 stands or falls with claim 1? aside from the APA argument? [00:14:11] Speaker 03: Yes, Your Honor. [00:14:11] Speaker 02: Okay. [00:14:13] Speaker 03: The board here carefully considered each and every one of Apple's obviousness arguments. The issue here is that the board's findings are a direct response to Apple's own framing and its failure to provide evidence to support that framing. I want to start with the unclaimed memory requirement. [00:14:32] Speaker 03: The claim here requires that there is an acoustic model memory for storing an acoustic model defining a plurality of states. The board here said I think at least two states are required. We don't dispute that, and Apple doesn't challenge the board's findings. [00:14:49] Speaker 02: Where does the large language model requirement come from, which the board seemed to impose here? I find it hard to see where that is in the claim. [00:14:59] Speaker 03: There is no large language requirement, and the board did not impose one. The issue here is the claim requires that there is an acoustic model with a plurality of states for doing these complex distance calculations and speech recognition. And mind you, these states are multidimensional. It could be 39-dimensional states for a variety, even for a small-digit recognizer, of a variety of sounds that make up the different numbers. That's so interesting. [00:15:27] Speaker 02: digit recognizer would fall within the claims. [00:15:31] Speaker 03: Yes, if it meets the other requirements. And that's exactly what Apple and its expert point to is the digit recognizer. [00:15:38] Speaker 03: The board addressed both their large and their abbreviated model distinction, but Smith doesn't have a distinction there. What Apple points to is the, quote, digit recognizer in its petition and its expert report as to what is the acoustic model it's relying on. [00:15:53] Speaker 02: Why wouldn't a digit recognizer fall within Mazer's 4K memory? [00:16:00] Speaker 03: That's an interesting question, but Apple didn't present any evidence on that. Apple didn't present evidence as to how that state memory, that digit recognizer acoustic model, could be stored in the memory on Moser's integrated circuit. And that is a claim element here. It's not just memory in the abstract. It's memory with an acoustic model with this functionality for doing speech recognition stored. And I think it's important to remember here we're talking about a 2004 priority date. [00:16:32] Speaker 03: This predates iPhone and Siri. At this time, these types of chips were very small. We're talking about digital signal processors with a very small amount of memory, and the board agreed with Zenshin's expert that none of the integrated circuits of chips that Apple pointed to in the record were capable of storing an acoustic model at this time, because they were very small, and these acoustic models are large and powerful. No, no, no. [00:17:00] Speaker 02: Wait, your own expert said that the digit recognizer was a very small, had a very small memory requirement. [00:17:07] Speaker 03: He said it would be smaller than a large language. He called very small, I think was exactly the words he used. He did not agree, though, that it was capable of being stored in any of the integrated circuits that Apple points to, though, and the board agreed that Apple's suggestion that our expert conceded that was mischaracterizing his testimony. I think it's telling they're pointing to our expert to try to support their case because they do not have anything in the petition in their own expert's declaration showing that you could put an acoustic model, the only one they point to in the record is the Smith's state memory, you know, that has a certain size and functionality. [00:17:47] Speaker 03: They don't point to any integrated circuit at the time that you could have stored that on, and they offer no arguments that you would be able to modify it and do something to get it to fit. [00:17:58] Speaker 00: Let me ask you, if there's some technological difficulty, as you say, in 2004 with putting this processor and the memory in the same integrated circuit, Where does your patent itself provide enabling disclosure or otherwise kind of speak to how the invention would overcome these difficulties? Because I'm not seeing it. [00:18:16] Speaker 03: Well, our patent, for example, figure 17 is a detailed architecture diagram showing how the acoustic model, the calculating apparatus could be configured such that these distance calculations could be done. Do you want to give me a line? Figure 17, which is appendix... [00:18:44] Speaker 00: 89, that's the figure. [00:18:47] Speaker 03: 89, appendix 89. So, you know, Apple's suggestion, there's nothing in here on this limitation. [00:18:52] Speaker 00: But where's the disclosure in what column, what line? [00:18:56] Speaker 03: So, for example, at column 35, for example, lines 5 through 16... [00:19:11] Speaker 03: This is talking about for on-chip RAM memory here. That's at line 5. This is expensive to have a lot of memory on a chip. And so towards the end of this paragraph, it talks about one way to be able to get this acoustic model in an on-chip memory at this time is to compress it. [00:19:28] Speaker 03: And that would be a way to get that acoustic model into memory on a chip, similar to what's shown in Figure 17. In Apple's, I think, opening brief, they suggest that a person of skill could have compressed the state memory in Smith to get it on the DSP, but that is not an argument they raised below. And I think the real issue here is there was just a failure for Apple to prove their case. [00:19:59] Speaker 03: There was no evidence of, for example, it's Smith alone theory. Apple has one line in there where it says, Smith teaches it would be obvious. The board made express findings that Smith does not teach this because it only talks about putting the classifier processor on the DSP and it never talks about putting the external state memory onto the DSP. [00:20:22] Speaker 02: Mazer discloses this. It discloses having a memory on the chip together with the processor and with a 4K memory. [00:20:29] Speaker 03: But it was not an acoustic model memory. It was undisputed that the acoustic model in Smith was stored... I'm sorry, in Moser was stored externally. It was not stored on Moser's chip. And, for example, Apple... disclaimed any reliance on Moser's teaching an acoustic model memory on that chip at a hearing transcript at Appendix 528. An expert also conceded that the, quote, vector memory that's on that chip is not an acoustic model memory. [00:20:59] Speaker 03: It's not capable of storing a full acoustic model at Appendix 3906. [00:21:04] Speaker 02: But that's a question, again, of what you define an acoustic model as being, which the board seemed to think was a large language model. And there seems not to be support for that in the past. [00:21:19] Speaker 03: No, the board said that it believed it could be a two-state model, but that Apple had not pointed to any model in the record that had two states and could function and do the functionalities that is in the claim, such as doing distance calculations. So it's an absence of proof here, on top of the fact that Apple did not raise the abbreviated, this quote, abbreviated model till its reply. [00:21:47] Speaker 02: Where in the oral hearing did they disclaim relying on Mazer for the acoustic model memory disclosure? [00:21:55] Speaker 03: Sure. It's at 528. 528. [00:22:08] Speaker 03: around line 13, we are not relying on Moser's acoustic model memory to teach claim B of an acoustic model memory for storing an acoustic model. They are simply relying on it as evidence of a chip that has processor and memory. Not memory that is capable of storing an acoustic model, but just as evidence of this is a configuration of a chip. [00:22:36] Speaker 03: So turning to their Smith Plus – so even putting – the board's findings about Smith Plus Moser just addressed whether Apple could – their combination could meet the elements of the claim in the way that the claim says. But that was just one of Apple's own failings. The panel doesn't even need to reach that question because it failed for multiple reasons, including no motivation to modify Smith to get Moser's configuration, but more importantly, that there was no reasonable expectation of success. [00:23:15] Speaker 03: And the only thing the panel needs to address here is Apple's argument that its theory for Smith plus Mosher was that you could go out and procure and purchase an off-the-shelf piece of hardware, which the board found was an untimely new argument raised for the first time in reply. And that finding... [00:23:40] Speaker 03: at Pandex 28 to 29 is reviewed for an abuse of discretion. Apple does not point to anything in the petition where it raised that. And in its reply brief, it abandons any notion that it raised a motivation to modify Smith to put its... state memory and its calculating apparatus on an integrated circuit together. So the court does not even need to address whether there is an improper bodily incorporation or unclaimed memory requirement. If you affirm the board's finding that Apple's theory that it could just go and purchase an off-the-shelf piece of hardware, if you find that that was not an abuse of discretion because it was an untimely raised argument. [00:24:23] Speaker 03: The board addressed that argument as well, though, and found it was unsupported because there was no evidence in the record, again, Apple's burden here, that any such integrated circuit at this time in 2004 existed to store an acoustic model with the functionalities to meet the claim. [00:24:39] Speaker 02: Because of the size of the memory required? [00:24:43] Speaker 03: No, because the only acoustic model Apple points to is Smith's state memory. [00:24:48] Speaker 03: It's not as if Apple pointed to – Apple didn't make any arguments that, oh, well, there are smaller acoustic models out there. [00:24:55] Speaker 02: When Apple relied on Smith, it didn't say Smith has a large language model in it, and that's what we're proposing it for. [00:25:09] Speaker 03: No, it points – that's correct, Your Honor. It points to Figure 3, the state memory, the digit recognizer. That's what it points to. And that's what the board made findings that Apple did not provide any evidence to support what that model – like what its size is. Can it fit on the DSP? Because Smith itself – does not ever say or suggest you could put it on the DSP. It only talks about having that state memory as external to the DSP. And Apple did not fill in the blanks with its expert to say, oh, well, a person of skill could have, would have done this. [00:25:43] Speaker 03: Like, there's just an absence of evidence in the record here. This is not a case where, you know, Apple put forth a full-fledged obviousness analysis and the board... There's no numerical value. [00:25:54] Speaker 00: There's no size requirement in these claims, right? That's correct, Your Honor. [00:25:59] Speaker 03: The only requirement is that there is an acoustic model stored in memory that is fabricated on an integrated circuit with the calculating apparatus such that you can do all of the distance calculations needed for speech recognition on this one chip. [00:26:14] Speaker 03: And the fabricating everything on one circuit... [00:26:19] Speaker 02: Those two statements that you made don't seem to be consistent. There's no size requirement, but you're saying that it has to function with a particular size. [00:26:30] Speaker 03: Well, it has to function because the claim requires... [00:26:36] Speaker 03: One of the aspects of the claim is a calculating apparatus for calculating distances, and those distances are calculated by comparing a feature vector, a multidimensional feature vector, to acoustic states of the acoustic model. So if you have a digit recognizer, say it has 100 different states for the different types of sounds you might have for 1, 2, 3, 4, all the different ways the letters and sounds could be represented, that acoustic model stored in memory, and then you also have the calculating apparatus that is taking feature vectors in, these multidimensional vectors, comparing them to every single state in the model, because that's how this works. [00:27:19] Speaker 03: You're not comparing it to one. You're doing these distance calculations to every single state in the model and computing which one is it most similar to. And so it's a pretty powerful calculation here. [00:27:30] Speaker 02: But if I understand what the board said with respect to Mazer, is it sort of dismissed... [00:27:37] Speaker 02: that reference as being irrelevant because Apple didn't show that someone skilled in the art would have the ability to fabricate the chip. Let's assume for the moment that that's wrong. That seems to be very questionable. What's left of the board's decision about Mazer [00:27:59] Speaker 03: Well, first, Apple did not appeal to finding that it was beyond the skill of a person of skill in the art to fabricate. [00:28:06] Speaker 02: Okay, but I'm just asking to put that aside as to whether there's any such requirement in there. Let's assume there is not, and that the board is mistaken about that. [00:28:17] Speaker 02: Then what's left of the board's decision that you can rely on with respect to MESA? [00:28:22] Speaker 03: So there's multiple, Your Honor. The fact that the combination of... [00:28:29] Speaker 03: Smith's state memory and its calculating apparatus that it would not meet the claimed limitation of the fabricated circuit with Smith's state memory stored in that memory in the circuit. So it doesn't meet the claim. The combination does not meet the claim element. [00:28:48] Speaker 02: Where does the board say that about Mazer? [00:28:51] Speaker 03: that it does not fit. [00:28:53] Speaker 03: What Apple was not suggesting, you put the state memory in Moser. [00:28:57] Speaker 02: What Apple was saying... I'm just trying to understand what's left of the board's decision after we put aside the fabrication issue. Yeah, and let me... So I view the board's... Where does the board make other findings with respect to Maser? [00:29:12] Speaker 03: Yes, so the board found that Apple's argument in reply that you could procure a piece of hardware off the shelf and use that, that that was untimely. [00:29:23] Speaker 02: That was at... Okay, but that's the fabrication issue. [00:29:28] Speaker 03: No, that's separate, Your Honor. [00:29:31] Speaker 03: They were saying you don't need a motivation to fabricate because that's not our theory. That instead you just pull, and I understand I'm into my rebuttal time, but I want to make sure I answer your question. So the board found the claim combination does not meet the claim elements, that there was no motivation to combine, rejected all of Apple's proposed motivations. That's at, for example, like four pages, appendix 32 to 35. [00:29:58] Speaker 03: also found that Apple's reliance on an abbreviated acoustic model was untimely. That's Appendix 23 to 24, and that was unsupported. That's Appendix 24. [00:30:10] Speaker 03: And so, you know, there's like five different findings here. Apple has to unravel each and every one, including findings that two of its arguments were untimely and are subject to an abuse of discretion review. [00:30:23] Speaker 03: And so if the panel has no further questions, we ask that you affirm. [00:30:27] Speaker 02: Mr. Matsui. [00:30:42] Speaker 01: Thank you, Your Honors. Just a few quick points. We didn't need to appeal that a personality skill in the arts the level to fabricate because that wasn't our argument. Our point was that you don't need to fabricate in the claims. [00:30:54] Speaker 02: And let's assume that you prevailed on that argument. What's left of the board's decision about Mazer? [00:31:00] Speaker 01: So the board said that you need the 4.7. [00:31:04] Speaker 01: Kilobytes was required from Smith's large, or the large model, and that wouldn't fit on majors for kilobytes. And that's what the board's rationale would be left. And they didn't consider the digit recognizer. And I'll just note that if you look at Smith figure three, that's the state memory. When you look at appendix 117 of Smith's, it's referring to figure three and it's talking about the digit recognizer. So I think the board was just mistaken when it was using this large model and saying that that was coming from Smith, when Smith was really more about digit recognizers instead. [00:31:40] Speaker 01: So I think that once you get rid of the fabrication requirement that I think is just wrong, then this case has to go back because you have the small model, you have the fact that you have Moser, which is four kilobytes, and all the rationales there were either based... upon the large model. [00:31:57] Speaker 02: What they're saying, I think, is that Moser doesn't disclose this because it doesn't disclose storing acoustic model memory. [00:32:07] Speaker 01: That would be an anticipation argument. We don't need an express disclosure that you have the coprocessor and the acoustic model memory already stored in memory. And I think that that's one of the things that Zenshin has been arguing, is that you need to actually show in the art this processor and this acoustic model memory together already in a chip. but our argument is an obviousness one and that's the fact that Moser doesn't expressly disclose something doesn't mean that it wouldn't have that it would not have been obvious to a person of ordinary skill in the art we're just talking about storing what you're getting from Moser is a processor and a memory of a particular size on the same integrated chip but it's also a speech recognition it's also in the field of speech recognition it's using an acoustic model There's all these reasons why a person of ordinary skill in the art would see the advantages of putting together a processor and the acoustic memory – acoustic model on a memory because there would be advantages. [00:33:11] Speaker 01: If I could just mention one thing about claim 29, I don't agree that if we were not to prevail on claim 1 that there would be nothing left on claim 29 because – that only would go to the acoustic model memory being on the integrated circuit, which is not a claim requirement. If there are no further questions, we would ask that the court reverse or vacate. [00:33:32] Speaker 02: Okay. Thank you. Thank both counsels.