[00:00:01] Speaker 01:
The next case for argument is 23-1454, Andrea Electronics versus Adam.

[00:00:43] Speaker 00:
Good morning.

[00:00:45] Speaker 00:
May it please the court?

[00:00:46] Speaker 00:
My name is Gwen Turisi.

[00:00:47] Speaker 00:
I'm here on behalf of the patent owner, Andrea Electronics.

[00:00:52] Speaker 00:
This court remanded in the previous decision with specific instructions to make findings of fact regarding why a person of ordinary skill in the art would have combined Hirsch with Martin.

[00:01:01] Speaker 00:
Rather than doing so, the board assumed away the central dispute, finding Andrea agreed that Hirsch identifies Martin as a solution to the problem of non-stationary noise.

[00:01:11] Speaker 00:
Andrea did not agree because the record doesn't support it.

[00:01:15] Speaker 00:
The board then combined Andrea's alleged agreement with Hersh's silence to find the existence of a motivation to combine.

[00:01:22] Speaker 00:
Silence is not a motivation to combine.

[00:01:23] Speaker 00:
That would invite the exact kind of hindsight bias and using the claims as a roadmap that KSR and this court's cases have counseled against.

[00:01:32] Speaker 00:
It's even more telling in this case that

[00:01:35] Speaker 00:
The specific combination alleged by the petitioner takes only certain pieces of Hirsch and only certain pieces of Martin to combine them as claimed.

[00:01:43] Speaker 00:
There's no explanation for why to do that in silence.

[00:01:46] Speaker 00:
The only explanation is that the claims are used as a roadmap.

[00:01:49] Speaker 03:
I thought the whole idea is you just rip out Hirsch's algorithm for setting up these adaptive thresholds and you just do a wholesale replacement, use Martin's, and then you apply the spectral subtraction.

[00:02:03] Speaker 00:
Your Honor, both Hirsch and Martin provide a complete algorithm for noise cancellation.

[00:02:08] Speaker 00:
So what the petitioner has advocated is taking out the improvement that Hirsch provides.

[00:02:13] Speaker 00:
the simplification that HRSH provides and replacing it with only those pieces of Martin that tend to be more complicated than what HRSH already discloses.

[00:02:20] Speaker 03:
As I understand the board's reasoning, it's that HRSH doesn't say anything about how to deal with non-stationary noise signals.

[00:02:30] Speaker 03:
Martin, however, does.

[00:02:31] Speaker 03:
And so at least, at the very least, in the context of having to confront and deal with non-stationary noise signals,

[00:02:41] Speaker 03:
an obvious thing to do would be to use Martin's algorithm.

[00:02:47] Speaker 00:
Well, Your Honor, the board is wrong.

[00:02:51] Speaker 00:
Martin doesn't actually disclose an improvement for non-stationary noise.

[00:02:55] Speaker 03:
I'd like to walk you through... I guess the idea is we know one of these two references deals with non-stationary noise.

[00:03:02] Speaker 03:
That's Martin.

[00:03:03] Speaker 03:
We don't know that about Hirsch.

[00:03:06] Speaker 03:
So, as I understand, the board kept saying this idea about

[00:03:10] Speaker 03:
when you're looking at Hirsch, which is unproven with respect to non-stationary noise, and we know about Martin, which definitely is effective for non-stationary noise, why not, in those instances when you're sensitive to non-stationary noise, go ahead and do the combination with Martin's algorithm?

[00:03:29] Speaker 00:
Your Honor, again, I have to disagree with the board's finding that Martin is proven for non-stationary noise.

[00:03:35] Speaker 00:
And I heard your caution earlier to not re-argue the facts.

[00:03:38] Speaker 00:
but the substantial evidence doesn't allow the board's decision to be affirmed when it's demonstrably wrong.

[00:03:43] Speaker 00:
So if I may, I'd like to walk you through why the board's decision is demonstrably wrong and contradicted by Martin itself.

[00:03:51] Speaker 00:
If you look at the board's decision at pages five through six, it says that Martin provides good results in both stationary and non-stationary noise, and it cites to exhibit 1006 at 1095 through 1096.

[00:04:05] Speaker 00:
The next paragraph starts by quoting

[00:04:08] Speaker 00:
what Martin discloses, which is that it provides a computationally inexpensive and effective mean to cope with the problem.

[00:04:18] Speaker 00:
If we look at exhibit 1006, which is now appendix 455, the problem that Martin is discussing is clearly varying noise levels, not non-stationary noise.

[00:04:31] Speaker 00:
The second sentence of the conclusion says, the algorithm proposed in this paper provides a computationally inexpensive and effective means to cope with this problem.

[00:04:38] Speaker 00:
That's what the board quoted.

[00:04:39] Speaker 00:
The sentence before that says, varying noise levels have a significant impact on the performance of many speech processing algorithms.

[00:04:47] Speaker 00:
The only way to read that is that the problem that Martin is solving is varying noise levels

[00:04:52] Speaker 00:
not non-stationary noise.

[00:04:53] Speaker 03:
What about where Martin's abstract talks about Martin's algorithm is capable to track non-stationary noise signals and has a low computational complexity?

[00:05:04] Speaker 03:
So that's the second sentence of this entire article introducing what Martin's going to be trying to do here.

[00:05:13] Speaker 03:
So I guess in that sense, I'm left with

[00:05:20] Speaker 03:
for the board to read this Martin reference as contemplating how to deal with non-stationary noise signals when that's in fact what it leads with in the abstract.

[00:05:34] Speaker 00:
Well, Your Honor, it's capable to track doesn't mean that it provides good results.

[00:05:38] Speaker 00:
Again, everyone agrees that Martin is tested for non-stationary noise.

[00:05:43] Speaker 00:
The question is whether or not the board's finding that it provides good results for non-stationary noise is supported.

[00:05:48] Speaker 00:
And what the boards relied on is not supported.

[00:05:50] Speaker 00:
Moreover, at page 43 of its reply brief, the petitioner stated, acknowledging that the board had confused non-stationary noise and varying noise levels, that the board's interpretation of varying noise levels as at least inclusive of non-stationary noise is certainly reasonable.

[00:06:06] Speaker 00:
But the issue with that is if the two are the same, HRSS doesn't need to be improved, because what HRSS says is that it presents two methods that are capable to adapt to varying noise levels, and one is

[00:06:18] Speaker 00:
low in computational complexity.

[00:06:20] Speaker 00:
That's at Appendix 451.

[00:06:23] Speaker 00:
So either the board's statement that Martin provides good results because it's computationally inexpensive and effective is unsupported by the record, or Hirsch doesn't need to be improved because varying noise levels and non-stationary noise are one in the same.

[00:06:38] Speaker 03:
Either way... This reference to varying noise levels in Hirsch that you're citing, did you argue that below the board?

[00:06:46] Speaker 00:
Your Honor, we didn't argue that specifically to the Board because, again, the Board assumed away the central dispute.

[00:06:52] Speaker 03:
What we did argue to the Board is that... My point is I didn't see this pop up until your gray brief.

[00:06:59] Speaker 00:
Your Honor, that's not correct.

[00:07:02] Speaker 00:
We argued it in both our opening brief in this case and in the gray brief.

[00:07:06] Speaker 00:
And what we told the Board below was specifically that HERSH doesn't need to be improved.

[00:07:11] Speaker 00:
There's no indication in HERSH itself that it doesn't already work for non-stationary noise levels.

[00:07:16] Speaker 00:
In fact, Hirsch says that it provides the same type of good results that Martin does.

[00:07:22] Speaker 00:
We've argued that below.

[00:07:23] Speaker 00:
We argued that.

[00:07:23] Speaker 00:
I can give you the site to our opening brief as well.

[00:07:27] Speaker 00:
I believe it starts on page 39 of our opening brief.

[00:07:35] Speaker 00:
Now, we didn't rely on the petitioner's admission that the board confused non-stationary noise and varying noise levels in our opening brief because we didn't have that until they responded.

[00:07:44] Speaker 00:
But that argument is laid out in several places in our opening brief.

[00:07:48] Speaker 00:
Page 39 is one of them.

[00:07:50] Speaker 00:
And it's also explained pages 44 through 49 of our opening brief with citations to the record.

[00:08:01] Speaker 00:
And your honor, I also want to go to the second legal error that we've identified, which is that the board failed to consider Hirsch as a whole.

[00:08:10] Speaker 00:
What Hirsch is doing is it's referring to Martin as a straw man to set up

[00:08:14] Speaker 00:
the explanation for the improvement that Hirsch provides over Martin.

[00:08:20] Speaker 00:
The first place that the board considers the context of Hirsch as a whole is in its rejection of Andrea's rebuttal argument.

[00:08:27] Speaker 00:
And what Hirsch specifically says is that Martin's approach, not known solution, but approach, is insufficient because it has too long past segments of noisy speech required.

[00:08:40] Speaker 00:
Again, this is the first time the board addressed that entire context.

[00:08:44] Speaker 00:
In doing so, the board said, we don't believe the testimony from either expert.

[00:08:48] Speaker 00:
We're going to reject the testimony from either expert.

[00:08:50] Speaker 00:
And then it substituted its own judgment in for what should have been the petitioner's burden of proof.

[00:08:56] Speaker 00:
It cites to the Belden case, but the Belden case doesn't address a final written decision.

[00:08:59] Speaker 00:
It addresses only the institution phase, which has a much lower standard of proof.

[00:09:06] Speaker 00:
What's left after the board rejects the two parties expert evidence is just what Hirsch itself says, which is Martin is a known approach, but that known approach isn't sufficient because it requires too long a past segment of noisy speech.

[00:09:21] Speaker 00:
And then Hirsch goes on to present its own entirely complete algorithm that doesn't use Martin.

[00:09:27] Speaker 00:
Again, there's nothing from reading those two references as a whole that would tell a person of ordinary skill in the arc, one, that Hirsch needs to be improved when it provides a computationally effective means to deal with noise cancellation, or two, even if it did suggest to combine it with Martin, there's no indication that you would take only those pieces of Martin and only those pieces of Hirsch that petitioner needs for its combination.

[00:09:52] Speaker 00:
That's clear hindsight bias, and the board

[00:09:55] Speaker 00:
in its decision has sort of a last point saying that it didn't need to consider how to combine the two references.

[00:10:02] Speaker 00:
But that's really important here.

[00:10:04] Speaker 00:
It's central to a question of motivation to combine because you're not taking two complete algorithms and combining them and just adding them together.

[00:10:12] Speaker 00:
You're taking bits and pieces of each.

[00:10:14] Speaker 01:
What's wrong with that?

[00:10:15] Speaker 01:
I mean, a person of skill in the art is entitled to look at two prior art references and pick and choose to combine, right?

[00:10:23] Speaker 01:
I mean, that's what combination is all about.

[00:10:25] Speaker 00:
Absolutely, Your Honor, but there again has to be at least some suggestion, something in the references that would indicate to a person of ordinary skill in the art to select those pieces.

[00:10:35] Speaker 00:
The only evidence that petitioner provided is that a person of ordinary skill in the art.

[00:10:39] Speaker 00:
I'm sorry, can you repeat that what you said?

[00:10:41] Speaker 00:
There has to be something specific in the references?

[00:10:43] Speaker 00:
There has to be something specific to support the board's decision.

[00:10:47] Speaker 00:
We fully acknowledge that teaching doesn't have to be expressed.

[00:10:51] Speaker 00:
That's what KSR said.

[00:10:53] Speaker 00:
But the board has to point to something more than just a person of ordinary skill in the art would have done exactly what the claims say, because it would have optimized two complete algorithms.

[00:11:03] Speaker 00:
And I want to draw your attention to one of the cases that we cite on that point, which is the personal web case versus Apple, the same petitioner, where this court found that such an argument only suggests that a person could combine the references, not that it would take those specific pieces and combine them in that specific way.

[00:11:20] Speaker 00:
There's nothing in the board's decision that supports a finding of taking just those pieces of Hirsch and just those pieces of Martin and combining them in exactly the way.

[00:11:29] Speaker 01:
But this is a remand, right?

[00:11:31] Speaker 01:
And we said that on remand, we want the board to look at the why, not the how.

[00:11:39] Speaker 01:
You're arguing the how.

[00:11:41] Speaker 00:
Well, Your Honor, I think the how and the why are intricately linked in an analysis of motivation to combine.

[00:11:48] Speaker 00:
There has to be something saying why a person of ordinary skill would take those specific pieces and combine them in the way that the petitioner argues.

[00:11:56] Speaker 01:
That's the why.

[00:11:57] Speaker 01:
And then the how is, how do you go about doing it?

[00:12:00] Speaker 01:
Correct.

[00:12:00] Speaker 01:
And we were interested in the why, not the how.

[00:12:04] Speaker 00:
Yes, Your Honor.

[00:12:04] Speaker 00:
And my point is that the board hasn't adequately explained the why, because it hasn't explained why you would take those specific portions.

[00:12:12] Speaker 01:
of Hirsch and those specific portions.

[00:12:15] Speaker 01:
It doesn't have to adequately explain it.

[00:12:16] Speaker 01:
It just has to have substantial evidence supporting its decision.

[00:12:19] Speaker 00:
Yes, Your Honor.

[00:12:20] Speaker 00:
And again, substantial evidence doesn't mean that anything that the board cites to is OK.

[00:12:24] Speaker 00:
The board has to cite something that's consistent with the record.

[00:12:27] Speaker 00:
And the board's finding that Petitioner met its burden here is Andrea agreed to something that it didn't, Hirsch is silent on non-stationary noise, and Martin provides an improvement for non-stationary noise.

[00:12:37] Speaker 00:
But what the board's relying on to find that Martin provides an improvement for non-stationary noise

[00:12:41] Speaker 00:
is contradicted by the record.

[00:12:44] Speaker 00:
So what, there's two ways to deal with the board's confusion over varying noise and non-stationary noise.

[00:12:50] Speaker 00:
Either our petitioner is correct that they're the same thing and that would mean then that HRSH doesn't need to be improved because it already provides good results for varying noise levels.

[00:13:01] Speaker 00:
Or the board's finding that Martin provides an improvement for varying noise levels isn't supported by the record.

[00:13:07] Speaker 00:
Either way, there's no grounds for

[00:13:10] Speaker 01:
the board's decision to be affirmed.

[00:13:25] Speaker 02:
I want to start by level setting a bit.

[00:13:27] Speaker 02:
What remains of this case is narrow and fact bound.

[00:13:30] Speaker 02:
The only remaining question is whether substantial evidence supports the board's robust factual findings across 18 pages that a skilled artisan would have been motivated to combine Hirsh and Martin to yield a combination that indisputably meets the claim evidence.

[00:13:44] Speaker 02:
A few basic points confirm that conclusion.

[00:13:47] Speaker 02:
In the 17 years since KSR, this court has reiterated many times the flexibility of the motivation to combine the inquiry.

[00:13:54] Speaker 02:
Within that inquiry, the starting point here is powerful.

[00:13:57] Speaker 02:
There is no question that a skilled artisan would have considered Martin's teachings generally when reviewing Hirsch.

[00:14:02] Speaker 02:
This court said as much in the last decision, and Hirsch explicitly refers to Martin as a known alternative approach to noise estimation

[00:14:08] Speaker 02:
into avoiding the problem with speech pause detection.

[00:14:11] Speaker 02:
Often in motivation to combine cases, much of the action is on why the skilled artisan would even look from the primary reference to the secondary reference.

[00:14:19] Speaker 02:
Here, we're past that.

[00:14:21] Speaker 02:
The board then credited Apple's evidence that those working in the field of noise reduction would regularly add techniques addressing particular challenges or issues like

[00:14:31] Speaker 02:
like noise reduction in non-stationary environments, here that's Martin, to modify an existing scheme, here that's Hirsch.

[00:14:38] Speaker 02:
So why would a skilled artisan have done this?

[00:14:41] Speaker 02:
Because, as the board also found, Hirsch did not test its system in non-stationary noise settings and in fact recognized the increased difficulty of such a setting.

[00:14:50] Speaker 02:
Martin, by contrast, did test its setting in non-stationary noise

[00:14:54] Speaker 02:
environments and found good results.

[00:14:56] Speaker 02:
Putting them together, substantial evidence supports the board's finding that a skilled artisan would have appreciated that Martin provides a proven way of dealing with non-stationary noise environments.

[00:15:06] Speaker 02:
These are two references that relate to the same field and address the same problem combined according to their established function.

[00:15:13] Speaker 02:
a formulation that this court has said many times is sufficient to provide a motivation to combine.

[00:15:20] Speaker 02:
The central defect in Andrea's arguments is that they are purely factual and cannot be reconciled to the Supreme Court's KSR decision and this court's case law sense.

[00:15:29] Speaker 02:
KSR jettisoned the so-called TSM test, and Judge Proce, I think you

[00:15:34] Speaker 02:
alertly noticed that what council was starting to say was that you needed some sort of affirmative suggestion or motivation in the prior art in order to make this combination.

[00:15:44] Speaker 02:
That is gone.

[00:15:45] Speaker 02:
Here, we think that the prior art actually does provide that motivation, but what Andre is asking for is even more.

[00:15:51] Speaker 02:
The gray brief at page 10 says, for example, that nothing in HRSH states that improvement is needed or that the board needed to affirmatively say that HRSH would not work well in non-stationary environments.

[00:16:02] Speaker 02:
That's not the right inquiry.

[00:16:03] Speaker 02:
The notion that a reference would have to denigrate itself and announce that it needs to be improved or would not work well in certain settings is both unrealistic and not required.

[00:16:13] Speaker 02:
This court has said many times that it's not necessary to show that a combination is even the best option, only that it is a suitable one.

[00:16:21] Speaker 02:
I want to address briefly the point that came up.

[00:16:24] Speaker 01:
What do we find in the record evidence that Martin has good results?

[00:16:31] Speaker 02:
That is what Martin is reporting.

[00:16:33] Speaker 02:
You mean in non-stationary settings?

[00:16:35] Speaker 02:
That is what Martin is reporting, and that is what the board found, including at appendix page eight, and it credits... Page eight?

[00:16:46] Speaker 02:
Yeah, yeah.

[00:16:47] Speaker 02:
It says, you know, patent owner disagrees with the effectiveness of Martin's algorithm, but then it says this is inconsistent with the express teachings of Martin.

[00:16:57] Speaker 02:
And it goes on to quote our expert declaration which is citing Martin itself to say that the algorithm is particularly suited to identifying those in a frequency bin during active speech and has advantages that have been tested to work well with non-stationary noise.

[00:17:12] Speaker 02:
The board then goes on to say,

[00:17:14] Speaker 02:
Patent owner and Dr. Douglas focus on certain inaccuracies but don't expressly address the overall effectiveness of Martin in non-stationary noise settings.

[00:17:23] Speaker 02:
Judge Chen, you were exactly right to point to the abstract where Martin says the entire point of this article is to address the non-stationary noise setting.

[00:17:32] Speaker 02:
And the board's finding that that is what Martin addressed and that Hirsch did not is at least reasonable.

[00:17:37] Speaker 02:
This is substantial evidence review.

[00:17:40] Speaker 03:
I guess Martin, in the conclusion, says something about how the algorithm is effective to cope with this problem.

[00:17:49] Speaker 03:
But then the board has to do an interpretation of what does that mean when it's talking about the word effective.

[00:17:57] Speaker 03:
And it read the reference to say, well, that is at least in part directed to non-stationary noise signals.

[00:18:06] Speaker 02:
Right.

[00:18:07] Speaker 02:
Because that's what Martin is about.

[00:18:09] Speaker 03:
Right.

[00:18:10] Speaker 03:
But, you know, there's little ambiguity as to, in this Martin reference, what does Martin mean when it uses the term varying noise signals versus when Martin uses the term non-stationary noise signals.

[00:18:24] Speaker 02:
Right.

[00:18:24] Speaker 02:
And I do want to address that because counsel for the appellant brought that up, including with respect to the blue brief, the sequence of the briefing here.

[00:18:34] Speaker 02:
It's wrong on the facts, and I'll get to why it's wrong on the facts, but this is also why the court does not allow parties to relitigate facts in this forum.

[00:18:42] Speaker 02:
In the blue brief, Andrea at pages 39 to 40 accused the board of misunderstanding Martin, and it said, without citation, that varying noise levels and non-stationary noise are not the same thing.

[00:18:54] Speaker 02:
In the red brief, we came back and said, this is an improper attempt to disagree with the board's factual findings.

[00:19:00] Speaker 02:
But it was at least reasonable for the board to read Martin when it's talking about varying noise levels as inclusive of non-stationary noise, because that's what Martin is about.

[00:19:11] Speaker 02:
The gray brief then tries to take this statement as equating varying noise levels with non-stationary noise and turn this into a factual issue that doesn't exist.

[00:19:21] Speaker 02:
The gotcha reply arguments are wrong.

[00:19:23] Speaker 02:
You can have varying noise levels in both stationary and non-stationary settings.

[00:19:28] Speaker 02:
Stationary simply means that the noise is slowly varying, so an oscillating fan, for example, sitting in this room hearing AC.

[00:19:37] Speaker 02:
Non-stationary noise means that it's more unpredictable, it's rapidly varying.

[00:19:41] Speaker 02:
When we in the red brief said that Martin is talking about varying noise levels and saying that its algorithm works well, that was in support of the board's reasonable reading that this was about non-stationary noise because that's what Martin tested and reported.

[00:19:58] Speaker 02:
Hirsch, by contrast, does not test and report on non-stationary noise.

[00:20:02] Speaker 02:
It's only stationary.

[00:20:03] Speaker 02:
That's slowly changing.

[00:20:04] Speaker 02:
So when Hirsch discusses slowly changing or varying noise,

[00:20:08] Speaker 02:
It is also perfectly reasonable and not inconsistent with Martin to say that what Hirsch is talking about is stationary noise.

[00:20:15] Speaker 02:
There's no inconsistency whatsoever, and the board's findings, again, are reasonable.

[00:20:20] Speaker 02:
I think this points up an important fact about the context of Andrea's appeal, given where we are in this case and after all this time.

[00:20:29] Speaker 02:
Andrea's position at bottom

[00:20:31] Speaker 02:
is that no reasonable fact finder could ever find a motivation to combine on this record.

[00:20:37] Speaker 02:
And respectfully, that's just implausible.

[00:20:40] Speaker 02:
It would destroy this court's post-KSR case law, but the flexibility of the motivation to combine inquiry, and it would upend the deference that this court gives to the board's factual findings, particularly in a decision that spends 18 pages on an issue that's usually one of many issues in the context of an obviousness fight.

[00:20:58] Speaker 02:
and carefully walks through each of the arguments, each of the expert declarations and makes fact findings throughout.

[00:21:05] Speaker 02:
Unless the panel has any further questions, I would urge support to affirm.

[00:21:20] Speaker 00:
Petitioner has tried to characterize Andrew's entire appeal as re-arguing the facts.

[00:21:25] Speaker 00:
And there are certainly factual errors in the board's decision, which we've discussed at length.

[00:21:29] Speaker 00:
But there are also serious legal errors.

[00:21:33] Speaker 00:
For one, the board relied on silence.

[00:21:35] Speaker 00:
Relying on silence without more, as the board did here, is a significant expansion of this court's motivation to combine case law, and an unjustified one at that.

[00:21:45] Speaker 00:
Again, this court found in the last appeal that

[00:21:48] Speaker 00:
person of ordinary skill in the art would consider Hirsch and Martin together and remanded with express instructions to make findings of fact regarding why a person of ordinary skill in the art would have combined them.

[00:21:59] Speaker 00:
The only thing the board came back with is an incorrect assumption about what Andrea agreed to and silence from one reference.

[00:22:06] Speaker 00:
That again is a significant expansion and is a legal error from the board.

[00:22:10] Speaker 00:
The second issue is that the board failed to consider the Hirsch reference as a whole.

[00:22:15] Speaker 00:
It is setting up Martin as a straw man to explain that Martin has problems that Hirsch's algorithm solves.

[00:22:23] Speaker 00:
The fact that Hirsch didn't expressly state that it tested its algorithm in non-stationary noise isn't enough for a finding to support the board's decision that a person makes.

[00:22:34] Speaker 01:
That's a factual issue there.

[00:22:36] Speaker 00:
Well, no, Your Honor.

[00:22:37] Speaker 00:
Again, the board didn't consider the reference as a whole before it found Petitioner had met its burden.

[00:22:41] Speaker 00:
The first time that the board considered Martin as a whole was in rejecting Andrea's counterarguments, where again, it rejected the testimony from both experts and then said, we as the board can fill in the gaps.

[00:22:52] Speaker 00:
That's also a legal error.

[00:22:56] Speaker 00:
Unless you have further questions?

[00:22:57] Speaker 01:
Thank you.

[00:22:57] Speaker 01:
Thank both sides.