US 10062377

Patent Overview: US 10062377 B2

Title: Distributed pipelined parallel speech recognition system

Patent number: US10062377B2
Application number: US14/788,164
Inventor(s): Guy Larri; Mark Catchpole; Damian Kelly Harris-Dowsett; Timothy Brian Reynolds
Assignee: Zentian Ltd (original and current assignee as listed)
Priority date (as listed): 2004-09-14
Filing date: 2015-06-30
Publication date (grant): 2018-08-28
Publication of application: US20160005397A1, 2016-01-07
Legal status (as listed): Expired – Lifetime; anticipated expiration 2025-09-14

Purpose (from the supplied text): The invention relates to speech recognition circuits and methods, with noted wide applicability, particularly for devices such as mobile electronic devices.

Plain-language scope (as far as the excerpt shows): The classifications point to speech recognition generally, feature extraction, reference-template training, speech classification/search using distance or distortion measures, and constructional details of speech recognition systems—specifically "memory allocation or algorithm optimisation to reduce hardware requirements."

The excerpt's background discussion describes prior approaches to search networks: Sukun Kim et al. describe using special-purpose processing elements for each node (effectively one processing element per phone), with an alternative of one processor per state; Ney et al. discuss language model lookahead (also called "smearing"), assigning a language model factor to each phone instance in the lexical tree, and report reduced average states per 10 ms frame (e.g., from around 168,000 states with no lookahead to around 8,000 with bigram lookahead).

Source: https://patents.google.com/patent/US10062377/en

What the excerpt does not establish: The excerpt is incomplete ("consult the source for omitted text"). It does not include the claims, detailed description, or embodiments, so the precise claimed scope cannot be stated. It also does not establish examination history, claim interpretation, or the contents of cited prior-art documents beyond the brief background characterizations above.

I'll research the litigation dockets for US 10062377 specifically, prioritizing live search results.

Let me dig into the specific district court dockets and the N.D. Cal. case.

Let me check the current status of the Amazon case, the earlier 6:21-cv-815 case, and any other suits.

Litigation Involving US Patent 10,062,377 (Literal Interpretation)

Scope note: I searched for the exact identifier 10062377 (US 10,062,377 B2, "Distributed pipelined parallel speech recognition system," Zentian Ltd.). All results below concern that patent. I excluded similarly numbered patents (e.g., 10,062,177 / 10,623,377 / the sibling patents 7,979,277, 7,587,319, 10,839,789, 10,971,140) except where they appear in the same case and are relevant context.

Date conflict (flagged): The task states the current date is April 26, 2026, but live search results include events dated June 8, 2026 and September 25, 2026. Per the operating rules, I report the search results as ground truth and flag the conflict rather than suppress it.

Unified Patents portal entries for this exact patent (from the authoritative patent page, which lists "Family has litigation"): IPR2023-00035; W.D. Tex. 6:22-cv-00122; IPR2023-01194; W.D. Tex. 6:22-cv-00123; N.D. Cal. 3:23-cv-02921; CAFC 24-2205.

---

A. District court litigation (asserting the '377 patent)

Case 1 — Zentian Ltd. v. Apple Inc., No. 6:22-cv-00122-ADA (W.D. Tex., Waco Div.)

Sources: PTAB Mandatory Notice in IPR2023-00035 ("The following matters relate to the '377 Patent: Zentian Ltd v. Apple Inc., 6:22-cv-00122, (W.D.Tex. Feb. 2, 2022)"); ex parte case page showing "Closed (06/13/23)"; Stanford NPE Litigation Database entry for patent 10062377.

---

Case 2 — Zentian Ltd. v. Apple Inc., No. 3:23-cv-02921-TLT (N.D. Cal.)

Sources: CourtListener RECAP filings in N.D. Cal. 3:23-cv-02921 (ECF Nos. 116 and 125, stipulations to stay / maintain stay), which expressly list "10,062,377 (the '377 Patent)" among the Asserted Patents; Apple's CAFC brief in No. 24-2207 confirming "That case was transferred to the Northern District of California as case No. 3:23-cv-02921."

---

Case 3 — Zentian Ltd. v. Amazon.com, Inc. (substituted: Amazon.com Services LLC and Amazon Web Services, Inc.), No. 6:22-cv-00123-ADA (W.D. Tex., Waco Div.)

Sources: Complaint (S.D. Tex./W.D. Tex. Dkt. 1, filed 02/02/22, Case 6:22-cv-00123-ADA); Amazon's First Amended Invalidity Contentions (asserting '377 claims 1–6); dockets.justia.com docket for 6:2022cv00123 (substitution order July 21, 2022); PTAB scheduling-order exhibits.

---

B. PTAB proceedings (adversarial administrative proceedings; not district court "litigation" but dispositive of the '377 claims)

The FWD caption reads: "APPLE INC., AMAZON WEB SERVICES, INC., and AMAZON.COM SERVICES LLC, Petitioner, v. ZENTIAN LIMITED, Patent Owner. IPR2023-00035, Patent 10,062,377… Final Written Decision Determining All Challenged Claims Unpatentable 35 U.S.C. § 318(a)." (docketalarm IPR2023-01194, June 11, 2024).

---

C. Federal Circuit appeals

Sources: govinfo.gov USCOURTS-ca13-24-02205; law.justia.com (24-2205 opinion); cafc.uscourts.gov (24-2204 Rule 36; 24-2205 oral-argument audio); CourtListener (24-2207 docket and brief).

Net effect on the '377 patent: All claims (1–6) were held unpatentable by the PTAB on June 11, 2024, and that determination was affirmed by the Federal Circuit on September 25, 2026 in No. 24-2205. The patent is also recorded as Expired – Lifetime (20-year term from Sept. 14, 2004 priority; anticipated expiration Sept. 14, 2025).

---

D. Items I could NOT confirm (flagged, not asserted)

• Possible earlier Zentian–Apple action, No. 6:21-cv-00815-ADA (W.D. Tex.). Amazon's amended invalidity contentions incorporate prior art from "Zentian Ltd. v. Apple Inc., No. 6:21-cv-815-ADA (W.D. Tex.)" (a 2021-filed Waco case). I could not verify whether the '377 patent was asserted in that 2021 case; it may involve only the earlier-issued family members. Treat as unverified.
• No ITC § 337 investigation involving the '377 patent was found in the retrieved results. I cannot rule one out.
• Final disposition of the Amazon W.D. Tex. case (6:22-cv-00123) — the last confirmed status is a pending case with trial set for Sept. 23, 2024; I did not retrieve a post-2024 dismissal or judgment for that case specifically.
• The Google Patents entry for this patent references a "First worldwide family litigation" via Darts-IP (family 33306576), but I could not open that record to enumerate foreign actions.

---

E. Confidence / caveats

• High confidence: The three district court cases (6:22-cv-00122, 6:22-cv-00123, 3:23-cv-02921) and the two PTAB IPRs (IPR2023-00035, IPR2023-01194) involve the exact patent US 10,062,377.
• High confidence: PTAB FWD June 11, 2024 (claims 1–6 unpatentable) and the CAFC 24-2205 affirmance.
• Lower confidence: The mapping of individual W.D. Tex. docket events to post-transfer status (the N.D. Cal. transfer vs. fresh filing mechanics) and the outcome of the Amazon case.
• Date conflict: Search results describe events through September 25, 2026, which postdate the task's stated "Current Date: April 26, 2026." I am reporting the later, searched results as authoritative per the operating rules.

Primary sources: https://patents.google.com/patent/US10062377/en; https://portal.unifiedpatents.com/ptab/case/IPR2023-00035; https://portal.unifiedpatents.com/ptab/case/IPR2023-01194; https://portal.unifiedpatents.com/litigation/Texas%20Western%20District%20Court/case/6%3A22-cv-00122; https://portal.unifiedpatents.com/litigation/Texas%20Western%20District%20Court/case/6%3A22-cv-00123; https://portal.unifiedpatents.com/litigation/California%20Northern%20District%20Court/case/3%3A23-cv-02921; https://portal.unifiedpatents.com/litigation/Court%20of%20Appeals%20for%20the%20Federal%20Circuit/case/24-2205; https://www.govinfo.gov/app/details/USCOURTS-ca13-24-02205; https://www.courtlistener.com/docket/73455769/ (24-2207); https://npe.law.stanford.edu/patent/10062377; https://dockets.justia.com/docket/texas/txwdce/6:2022cv00123.

I'll start by surfacing the PTAB proceedings. The structured ODP block says "no AIA trial proceedings," but the authoritative patent text lists two IPRs with Final Written Decisions — that's a direct contradiction I need to resolve with search.

Strong confirmation. Now let me pin down the CAFC panel and the IPR2023-00035 institution/briefing details.

I have the core record. Let me confirm the joined-proceeding panel and rule out any additional AIA filings on the '377 patent specifically.

PTAB Proceedings — U.S. Patent 10,062,377 B2 ("Distributed pipelined parallel speech recognition system," Zentian Ltd.)

⚠️ Source conflict — read first. The structured "PTAB proceedings on file" block supplied in this task states that the USPTO ODP API returns no AIA trial proceedings for this patent. That is incorrect, and I am overriding it. The authoritative patent text supplied in this same prompt lists "PTAB case IPR2023-00035 filed (Final Written Decision)" and "PTAB case IPR2023-01194 filed (Final Written Decision)" on the face of the Google Patents record, and my live searches confirm both. The ODP ingest is stale/incomplete. Per the operating rules (prefer the authoritative source; flag contradictions), I proceed on the two IPRs below and flag the structured block as erroneous. Do not represent to any court or client that this patent has no PTAB history.

---

Proceedings overview

There are two AIA trial proceedings on US 10,062,377 — IPR2023-00035 (Apple) and IPR2023-01194 (Amazon, joined as an understudy to the Apple IPR) — which together produced a single merits outcome: all challenged claims 1–6 held unpatentable for obviousness (§ 103) on 2024-06-11, and that holding was affirmed by the Federal Circuit on 2026-09-25. Breakdown: 0 active · 2 claims-invalidated · 0 sustained · 0 settled · 0 institution-denied. The defensive posture for a defendant today is the favorable one: claims 1–6 of the '377 patent have been adjudged unpatentable and the affirmance is final on appeal — any demand letter asserting claims 1–6 is asserting dead claims.

---

IPR2023-00035 — Apple Inc. v. Zentian Limited

• Type: Inter Partes Review (35 U.S.C. §§ 311–319)
• Filed: 2022-11-15
• Status: Final Written Decision – Appealed (PTAB terminated 2024-06-11); affirmed by the Federal Circuit 2026-09-25
• Judge panel: Administrative Patent Judges Kevin F. Turner, Jeffrey S. Smith, and Christopher L. Ogden (panel confirmed in the joined IPR2023-01194 institution decision caption); the FWD was authored as a single combined decision for both proceedings.
• Petition grounds: Challenged claims 1–6, all under 35 U.S.C. § 103 (obviousness). No § 102 anticipation ground; no § 112 ground. The specific grounds (confirmed by Zentian's own Notice of Appeal):
  • Claim 1 — obvious over US 6,374,219 B1 (Jiang) + US 5,819,222 (Smyth)
  • Claims 2, 3 — Jiang + Smyth + US 6,879,954 B2 (Nguyen)
  • Claim 4 — Jiang + Smyth + Nguyen + US 6,959,376 B1 (Boike)
  • Claim 5 — Jiang + Smyth + Nguyen + US 2002/0049582 A1 (Baumgartner)
  • Claim 6 — Jiang + Smyth + Nguyen + Boike + Baumgartner
• Institution decision: Instituted on all challenged claims and all grounds, 2023-06-12. The Board declined to exercise § 314(a) Fintiv discretion in favor of denial because Apple filed a Sotera stipulation on 2023-04-17 ("Petitioner will not pursue in the parallel district court proceeding the same grounds as in the Petition or any grounds that could have reasonably been raised in the pending Petition"). That stipulation was the basis on which institution was granted notwithstanding the co-pending W.D. Tex. case.
• Final Written Decision (2024-06-11, Paper 34): Every challenged claim — claims 1, 2, 3, 4, 5, and 6 — held unpatentable as obvious. No claim was sustained. The dispositive merits issue was claim construction of "feature vector." Although neither party formally moved for construction, the Board found the parties' positions "raised implicit claim constructions" and construed the term:
  • Zentian's position (rejected): a feature vector is limited to "pure" quantities actually extracted from the digital audio stream.
  • Apple's position (adopted): a feature vector also includes quantities derived from the stream as representations of it.
  • Board's construction: "a 'feature vector,' as defined by independent claim 1, comprises a plurality of extracted qualities [sic] from the digital audio stream during the defined audio time frame, and/or derived quantities that are representative of the digital audio stream during the defined audio time frame, such as the derived values of a codeword." The Board reasoned that "[a] codeword, or representative feature vector, 'comprises a plurality of … derived,' or vector quantized, 'quantities from said digital audio stream'" — i.e., codewords count as feature vectors. On that construction, the Board found Jiang disclosed claim element 1[g] and held all claims obvious. (Note: the Board's "qualities" for "quantities" was acknowledged as a typo; the Federal Circuit held nothing turned on it.)
• Settlement / termination: None. No adverse judgment, no settlement, no dismissal. Both proceedings ran to FWD.
• Appeal: Yes — CAFC No. 24-2205, Zentian Ltd. v. Apple Inc. (notice of appeal filed 2024-08-09; docketed 2024-08-13; appellees Apple Inc., [Amazon.com Services LLC](/litigations/by-plaintiff/Amazon.com%20Services%20LLC), and Amazon Web Services, Inc.). Oral argument 2026-06-08. Nonprecedential opinion issued 2026-09-25 (panel: Dyk, Mayer, Prost) — AFFIRMED. Zentian's sole appellate issue was the "feature vector" construction; the court affirmed the Board's construction (codewords are "derived quantities … representative of the digital audio stream") and found substantial evidence supported the obviousness determination.
• Defensive value: Claim 1 is dead, and every dependent claim with it. Any infringement theory built on claims 1–6 is now untenable; pressing one invites Rule 11 exposure. A defendant can cite the affirmed FWD directly for issue preclusion / collateral estoppel effect once the § 318(b) certificate issues.

---

IPR2023-01194 — Amazon.com Services LLC & Amazon Web Services, Inc. v. Zentian Limited

• Type: Inter Partes Review, instituted and joined to IPR2023-00035 under 35 U.S.C. § 315(c)
• Filed: 2023-07-12 (with a contemporaneous Motion for Joinder)
• Status: Final Written Decision – Appealed (institution + joinder granted 2024-01-10; terminated by FWD 2024-06-11)
• Judge panel: Same panel as the Apple IPR — Kevin F. Turner, Jeffrey S. Smith, and Christopher L. Ogden (Turner authored the institution-and-joinder decision)
• Petition grounds: Identical to Apple's — claims 1–6, § 103 obviousness, on the same Jiang/Smyth/Nguyen/Boike/Baumgartner combinations, supported by a declaration adopting Apple's expert (Christopher Schmandt). Amazon's petition was "intentionally identical in substance to the Apple Petition."
• Institution decision: Instituted on all challenged claims (2024-01-10) and joinder GRANTED. The Board declined Zentian's § 314(a)/Fintiv arguments and its objection that Amazon — unlike Apple — offered no Sotera stipulation. The Board reasoned that (a) Fintiv factors 1–5 weighed heavily against discretionary denial, (b) the Board would decide patentability well before the parallel district court's scheduled 2023-09-23 trial date, and (c) a Sotera stipulation is "no such stipulation is required where the Fintiv factors as a whole do not otherwise weigh in favor of discretionary denial." Amazon committed to an "understudy" role, agreeing not to make substantive filings and to be bound by Apple's filings absent Apple's termination.
• Final Written Decision (2024-06-11, Paper 11): A Judgment / Final Written Decision was entered in the joined proceeding reaching the same result — all challenged claims 1–6 unpatentable. (Because the merits trial was consolidated with IPR2023-00035, the substantive analysis is the shared FWD cited at Apple Inc. v. Zentian Ltd., No. IPR2023-00035, 2024 WL 2946335 (P.T.A.B. June 11, 2024).)
• Settlement / termination: None.
• Appeal: Covered by the same appeal, CAFC No. 24-2205 (Zentian's Notice of Appeal was captioned to both IPR2023-00035 and IPR2023-01194). Affirmed 2026-09-25.
• Defensive value: Confirms that joining petitioners to an instituted Apple IPR is now foreclosed for anyone hoping to re-litigate the same art — and Amazon is separately bound by estoppel. For a defendant, the joined proceeding reinforces that the § 103 invalidity theory on claims 1–6 is fully adjudicated and affirmed.

---

Strategic summary

Claim status. All six claims of the '377 patent — independent claim 1 and dependent claims 2–6 — were held unpatentable by the Board on 2024-06-11, with no claim sustained and no ground partially upheld. The Federal Circuit affirmed on 2026-09-25. There are no "surviving claims" of the '377 patent. Formally, "cancellation" occurs when the Director issues a certificate under 35 U.S.C. § 318(b) following the appellate mandate; because the affirmance is only days old (2026-09-25), the certificate may not yet have printed, but the outcome is final. There are zero untested claims — the entire claim set was challenged and invalidated. The patent is also recorded Expired – Lifetime (anticipated expiration 2025-09-14), so there is no term left to assert even absent cancellation.

Estoppel landscape. Under § 315(e)(2), Apple — and, as a joined party, Amazon (and their real parties in interest/privies) — are estopped from asserting in district court any ground raised in the IPRs or that they reasonably could have raised, as to claims 1–6. Apple's Sotera stipulation (2023-04-17) independently bars Apple from pursuing those grounds in the parallel litigation. For a newly sued defendant, the practical picture is: the claims are invalid by judgment, and the invalidity record is public and affirmed — so the cheapest path is to move for issue preclusion / summary judgment of invalidity based on the affirmed FWD rather than to run an expensive new IPR. Note that estoppel from these proceedings runs against the petitioners, not against a non-privy defendant, but the underlying adjudication of unpatentability is available to that defendant as a judgment on the merits.

Pattern signals. The same aggressor dynamic ran twice on this patent: Apple filed first (2022-11-15, instituted 2023-06-12 with a Sotera stipulation), and Amazon filed and joined as an understudy (2023-07-12, joined 2024-01-10) after being appear time-barred — meaning Amazon could only get in through § 315(c) joinder, which the Board allowed. This is the classic large-operating-company / Big-Tech co-defendant pattern responding to an NPE (Zentian Ltd., characterized as a small-company NPE) asserting a 2004-priority speech-recognition family. There is no defensive aggregator (e.g., Unified Patents) in the petition chain here — the petitioners are Apple and Amazon. The patent owner (Zoentian, via Noroozi PC) appealed aggressively across the family: the sibling appeals 24-2204 (Rule 36 affirmance, 2026-06-08), 24-2206 (dismissed), and the related 24-2207 (Apple's cross-appeal on US 10,839,789) confirm that the '377 win for petitioners is part of a broad, coordinated invalidation of the Zentian speech-recognition family — which strengthens, not weakens, the defensive value of the '377 result.

Watch-outs / corrections to the earlier sections. (1) The previously generated "Obviousness" section states claim 1 of this patent was held obvious over "Jiang, Baumgartner, and Brown, with or without Smyth." That is the ground for the sibling '277 patent (IPR2023-00034/01193), not the '377 patent. For the '377, the operative claim-1 ground is Jiang + Smyth (per Zentian's Notice of Appeal and the FWD). "Brown" is not a claim-1 reference for the '377. Flagging this cross-reference because it would be a misstatement in any filing. (2) The prior "Litigation" section correctly identified the two IPRs and the 2024-06-11 FWD, but attributed a "Final Written Decision June 11, 2024" in IPR2023-01194 to a separate Amazon petition; in fact Amazon's petition was joined and consolidated into the Apple merits trial, producing one shared FWD.

---

Recommended next steps

If you are a defendant facing assertion of the '377 patent:

1. Treat claims 1–6 as invalidated. Quote the Board's dispositive disposition and the appellate affirmance, and move for preclusion. Primary sources:

   • PTAB FWD (shared): Apple Inc. v. Zentian Ltd., No. IPR2023-00035 (P.T.A.B. June 11, 2024), 2024 WL 2946335 — via PTAB E2E: https://ptab.uspto.gov (search application number 14/788,164) and the docketed decision at Docket Alarm: https://www.docketalarm.com/cases/PTAB/IPR2023-00035/Apple_Inc._v._Zentian_Limited/
   • Joined FWD: IPR2023-01194, Paper 11 (P.T.A.B. June 11, 2024) — https://ipverse.greyb.com/ptab-web/cases/case-details/IPR2023-01194
   • Federal Circuit affirmance: Zentian Ltd. v. Apple Inc., No. 24-2205 (Fed. Cir. Sept. 25, 2026) (nonprecedential) — https://law.justia.com/cases/federal/appellate-courts/cafc/24-2205/24-2205-2026-09-25.html ; official copy at https://www.govinfo.gov/app/details/USCOURTS-ca13-24-02205 ; oral-argument audio (2026-06-08) at https://www.cafc.uscourts.gov/06-08-2026-2024-2205-zentian-ltd-v-apple-inc-audio-uploaded/ ; summary at https://fedcircuitblog.com/2026/09/25/opinions-orders-september-25-2026/
   • Verbatim disposition to quote: "We affirm." / "AFFIRMED" — affirming the Board's determination that all claims of U.S. Patent No. 10,062,377 are unpatentable for obviousness.

2. Do not file a redundant IPR. Two large-operating-company petitioners already invalidated the full claim set on the strongest art; any new IPR would face § 325(d) and discretionary-denial risk, and cannot improve on "all claims canceled." The efficient defense is estoppel/preclusion in the district court, not more PTAB work.

3. Check the certificate. Once the Federal Circuit's mandate issues and the Director issues the § 318(b) certificate canceling claims 1–6, obtained a copy from USPTO PatentCenter (https://patentcenter.uspto.gov/) and attach it to your preclusion papers. Until then, cite the affirmed FWD plus the affirmance as the operative proof of invalidity.

If you are monitoring the family (not just the '377): the sibling patents have their own, separate PTAB records — IPR2023-00033/01192 ('319, appeal 24-2204 affirmed by Rule 36 on 2026-06-08), IPR2023-00034/01193 ('277, appeal 24-2206 dismissed), IPR2023-00036/01195 ('789, appeal 24-2207), and IPR2023-01197 ('140). Do not conflate those outcomes with the '377 result — the 2026-09-25 affirmance is specific to the '377 patent in appeal 24-2205. (There is no PGR or CBM proceeding on the '377 patent in the record, and the structured ODP "no proceedings" block is unreliable as noted above.)

Confidence and caveats. High confidence on: the existence, numbers, petitioners, filing/institution/termination dates, panel (Turner, Smith, Ogden), grounds, and the all-claims-unpatentable outcome for IPR2023-00035 and IPR2023-01194; and on the 24-2205 affirmance (2026-09-25, nonprecedential, Dyk/Mayer/Prost). Lower confidence / unverified: (a) the precise paper numbers of the joined IPR2023-01194 FWD (reported as Paper 11; the Apple FWD is Paper 34, confirmed); (b) whether any petition for rehearing en banc has been filed in 24-2205 — my search budget ended before I could confirm; the opinion is only days old, so reheaaring activity, if any, may not yet be docketed; and (c) the exact date the § 318(b) certificate is expected to issue. Verify (b) and (c) against the CAFC docket (https://www.cafc.uscourts.gov/) and PatentCenter before relying on them.

I'll research the ownership chain for US 10,062,377. Let me start with the assignment records and the original assignee.

Let me dig into the assignment records specifically and Zentian's corporate status.

I hit my search-step limit, so I'm reporting what the retrieved records support and flagging precisely what I could not verify. Per the operating rules, I will not invent reel/frame numbers.

Assignment & Ownership Reconstruction — US 10,062,377 B2

Critical source limitation (read first). I was not able to retrieve the actual USPTO Patent Assignment Search reel/frame entries for this patent within my search budget. The authoritative Google Patents record supplied for this patent lists Original Assignee: Zentian Ltd and Current Assignee: Zentian Ltd, and its legal-events timeline contains only filing/priority/publication/grant events — no assignment records. That is consistent with ownership never having left the originating company, but I could not open assignmentcenter.uspto.gov directly to enumerate reel/frame entries, so the reel/frame column below is unpopulated rather than fabricated. Treat the assignment-timeline section as incomplete pending direct Verification at https://assignmentcenter.uspto.gov/ (search "10062377").

---

Inventors

Pattern note. This is not a case of all inventors departing a large operating assignee within 12 months of filing. The inventors are the company: Zentian was a founder-led startup whose engineers "worked in exchange for share options … investing their own sweat equity" (per the 2022 complaint, ¶34). The only detectable departure signal is Guy Larri later appearing with NVIDIA. All four inventors were, at filing, contributors to a single small UK company, not employees of a public operating company.

---

Original assignee

• Name as issued: Zentian Limited, Cambridge (GB). Google Patents renders the current/original assignee as "Zentian Ltd" — the same UK entity (Ltd = Limited); this is a naming variant, not an assignment.
• Primary line of business: speech-recognition silicon IP — a hardware accelerator for the acoustic (distance/probability) stage of speech recognition to run alongside a software search stage in mobile/embedded devices.
• Did it ship a product embodying the claims? No. Per Zentian's own 2022 complaint: most of Zentian's work was unpaid, funds were hard to raise after the dot-com crash, and Zentian achieved only a partial prototype on a programmable hardware-emulation platform plus a demonstrator pairing its acoustic accelerator with the open-source Sphinx recognizer (reported 2× speedup). It pitched Nuance, IBM, Sensory, Samsung, Qualcomm, Hutchison 3G and (in 2006) Apple's Silicon Engineering Group — no commercial product resulted, and Apple "ceased communication" after the pitch (¶40).
• Financing: ~$150K total raised (investor: BIS), last round ~2003 (CB Insights).
• Current status: as of the 2022 lawsuits it was a private UK corporation, registered address 27 Queens Close, St Ives, Cambridgeshire, PE37 5QD, HQ at St John's Innovation Centre, Cowley Road, Cambridge. I found no evidence of acquisition, dissolution, or insolvency. The asserted patent's term has since expired (anticipated 2025-09-14).

Key point: Zentian is both the originator and (per the records retrieved) the current owner — there was no operating-company → NPE transfer because Zentian never operated a product line to begin with.

---

Assignment timeline

No recorded post-issuance assignment was retrievable. The authoritative Google Patents legal-events record for this patent contains no assignment events, and both the original and current assignee fields read Zentian Ltd. This means, on the retrieved record, the chain is a single link: inventors → Zentian Limited (the original filing/company assignment), with no subsequent transfer.

• Unverified (not fabricated): an inventors → Zentian Limited founding/employment assignment almost certainly exists (the patent issued to Zentian Limited), but I could not confirm its reel/frame, execution date, recording date, or correspondent. Do not cite a reel/frame for this patent without pulling it from Assignment Center.

Because I cannot produce the required reel/frame rows, I am not going to invent them. The honest finding is: appears to be a single-owner chain with no post-issuance transfers, pending direct Assignment Center verification.

---

Timeline diagram

timeline
    title Ownership of US 10062377
    2000 : Zentian founded by Mark Catchpole
    2004 : GB priority application filed
    2005 : PCT application filed
    2006 : Zentian pitches Apple on silicon design
    2018 : Patent issued to Zentian Limited
    2022 : Zentian sues Apple and Amazon
         : Apple files IPR2023-00035
    2023 : Amazon IPRs joined to Apple proceeding
    2024 : PTAB invalidates all claims 1 to 6
    2025 : Anticipated patent term expiry
    2026 : Federal Circuit affirms invalidity

(Event text kept short and free of parentheses, slashes, and commas for parser safety.)

---

NPE / troll-pattern signals

---

Verdict

NPE — moderate confidence.

Zentian asserts the patent as a small non-practicing entity — it developed silicon IP but never shipped a commercial product embodying the claims (only an FPGA/emulation prototype and a Sphinx-based demonstrator), and it monetizes via litigation against Apple and Amazon. The single strong signal is external classification: Unified Patents labels the owner "NPE (Small Company)" and the Stanford NPE Litigation Database records the Apple suit for this patent. Confidence is only moderate, not high, because the classic structural tells are absent on the retrieved record: there is no shell-entity transfer, no cascading LLC hops, no pre-litigation assignment, and no bankruptcy sale — the patent appears to have remained with its original owner (Zentian) from issue through assertion. I could not run the repeat-recording-correspondent test (signal #3) because no reel/frame data was retrievable, which is the principal gap in this analysis.

Verification link (Assignment Center): https://assignmentcenter.uspto.gov/ — search patent number 10062377 (mirror: https://assignment.uspto.gov/patent/index.html) to pull the reel/frame, recording dates, and correspondent that this analysis could not populate.

Primary sources used: https://patents.google.com/patent/US10062377/en ; https://portal.unifiedpatents.com/ptab/caselist?lawfirms.patent_owner=Noroozi+PC ; https://npe.law.stanford.edu/patent/10062377 ; https://www.cbinsights.com/company/zentian ; Zentian complaint (N.D. Cal. 3:23-cv-02921, ECF 1, filed 2022-02-02, via https://ai-lab.exparte.com/documents/dct/[4467892](/patent/4467892)/3-23-cv-02921_cand_complaint_exparte.pdf) ; https://ipverse.greyb.com/ptab-web/cases/case-details/IPR2023-00035 .

Caveats: (1) The assignment reel/frame/correspondent data was not retrievable and is left blank rather than guessed. (2) Guy Larri's current NVIDIA affiliation comes from a third-party patent-aggregator page and should be confirmed before being relied on. (3) Per the earlier sections, note the family-wide 2024 PTAB invalidation (affirmed 2026-09-25) and the 2025-09-14 anticipated expiry — the patent is expired and its claims have been adjudicated unpatentable, which materially limits any residual assertion value regardless of the ownership chain.

I'll research the specific prior-art citations for US 10,062,377. Let me search for the patent's citation records.

Let me get the front-page "References Cited" list specifically for the '377 patent.

Let me retrieve the full front-page "References Cited" list and the Google Patents citation set for this exact patent.

Prior-Art Analysis — US 10,062,377 B2

Scope discipline. I searched the exact identifier 10062377 (US 10,062,377 B2, Distributed pipelined parallel speech recognition system, Zentian Limited, application 14/788,164). I intentionally did not return or conflate the similarly-numbered items that surfaced in the same searches — notably DE 10062377 A1 (a German laid-open document dated 27-06-2002) and the sibling US patents 7,979,277 / 10,839,789 / 10,971,140. Only US 10,062,377 is addressed below.

A structural caveat first. The authoritative Google Patents text supplied in this prompt omits the "(56) References Cited" / patent-citation block — it jumps from the classification tree straight into the "Definitions" narrative. So the face-of-patent citation list had to be recovered from the patent's own front page as reproduced in the IPR exhibits. I recovered a partial OCR of that front page and am reporting it as found, with the gaps flagged rather than reconstructed. I was not able to open patentimages or USPTO Patent Center directly to enumerate the complete list.

---

1. Front-page "References Cited" — the citations actually printed on the '377 patent

From the front page of US 10,062,377 B2 as reproduced in Exhibit 1001 of the IPRs (Docket Alarm, IPR2023-01194 … Exhibit_1001_US_Patent_No_10,062,377_.pdf; corroborated by the front page of the sibling '789 patent in IPR2023-00036, Ex. 1001):

U.S. Patent Documents

Foreign Patent Documents

What this list tells us. A striking share of the face citations are the patent's own family (Larri/Catchpole/Zentian: '277, '262, '441, and the Catchpole '890) — i.e. the examiner cited the applicant's own chain, not independent third-party art. That is consistent with the litigation record: the face of the '377 is a thin prior-art record, and the art that ultimately killed the patent was not on the face — it was located by the IPR petitioners.

Honest limitation: I could not confirm the titles/subject matter of most of these face references (Dulong, Kao, Wells, Kato, Zhao, Wang, and the two foreign documents) from the retrieved evidence. I am not going to invent descriptions for them.

---

2. The references that actually matter — the IPR grounds

The art that was adjudicated is not the face-citation list. It comes from IPR2023-00035 (Apple) and the joined IPR2023-01194 (Amazon), whose Final Written Decision (2024-06-11, Apple Inc. v. Zentian Ltd., Paper 34) held claims 1–6 unpatentable, affirmed by the Federal Circuit in No. 24-2205 (Sept. 25, 2026). The grounds (confirmed by the Schmandt Declaration table of contents and Zentian's Notice of Appeal) are:

Detailed citations

① Jiang — the primary reference

• Full citation: U.S. Patent No. 6,374,219 B1 (Jiang) — "Ex. 1004"
• Date: granted April 2002 (pre-priority by >1 yr → § 102(b))
• Description (from the Board/petition record): A speech recognizer in which "feature extraction module 66" divides the digital signal into ~10 ms frames and "encodes each frame into a feature vector reflecting the spectral characteristics for a plurality of frequency bands"; "in the case of discrete and semi-continuous hidden Markov modeling," it "encodes the feature vectors into one or more codewords using vector quantization techniques and a codebook derived from training data" (col. 7:1–5). A "tree search engine 74" traverses a prefix tree / lexicon 70 with pruning, referencing an HMM model memory and a language model. Jiang expressly contemplates splitting the work across processors: feature extraction on a "conventional array processor," while tree search is "implemented in [a CPU] … or … a dedicated speech recognition processor."
• § 102 significance: Jiang is the closest single-reference prior art and the anchor of every ground. On the Board's (and Federal Circuit's) construction of "feature vector" — encompassing derived quantities such as codewords — Jiang disclosed the claim-1 element 1[g]/1[b] the patentee argued was missing. Jiang maps to claim 1(a) (feature vector from audio), 1(b)/1(c) (distance calculation vs. acoustic states), and 1(d) (lexical-tree search).

② Smyth — the secondary reference

• Full citation: U.S. Patent No. 5,819,222 (Smyth et al.) — "Ex. 1005"
• Date: granted 1998 (→ § 102(b))
• Description: A "speech recognition system [that] recognizes connected speech using a plurality of vocabulary nodes." Recognition processor 3 includes a frame generator 32 (frames from digital speech in) and a feature extractor 33. The classifier processor 341 may be a DSP, such as the Motorola DSP56000 — i.e., the reference was relied on for putting the calculating apparatus and memory on a single integrated circuit.
• § 102 significance: Smyth supplies a second programmable device that calculates the state/class distances and the single-chip DSP teaching. It is essential to the combination because Jiang alone did not teach the separate distance-calculating device as claimed.

③ Nguyen

• Full citation: U.S. Patent No. 6,879,954 B2 (Nguyen) — mapped to claims 2 and 3 (pending/overlapped processing; physically distinct circuits).

④ Boike

• Full citation: U.S. Patent No. 6,959,376 B1 (Boike et al.) — "Ex. 1006"
• Description: teaches "a system-on-a-chip (SoC) integrated circuit 10 containing multiple digital signal processors (DSPs)" (2:52–54).
• § 102 significance: used for claim 4 / claim 6 (single-integrated-circuit / SoC limitations).

⑤ Baumgartner

• Full citation: U.S. Patent Application Publication No. 2002/0049582 A1 (Baumgartner et al.) — "Ex. 1007"
• Date: published April 2002 (→ § 102(b))
• Description: A speech recognition system with "Speech Label Accelerators (SLAs)," distinct processors that "speed the labeling phase of the speech recognition computation" and "reduce the computation load of a processor."
• § 102 significance: maps to claim 5 / claim 6 (accelerator/distinct-device limitations).

Supporting/background references cited in the Schmandt Declaration exhibit table (used for POSITA background, not as standalone grounds): US 6,499,012 (Peters); US 7,120,582 (Young); US 4,926,488 (Nadas); US 6,182,037 (Maes); US 2003/0220796 (Aoyama); US 6,182,203 (Simar, Jr. et al., Jan. 30, 2001 — confirmed from Apple Ex. 1039); and non-patent literature (Newton; Kloker, Motorola DSP56000, IEEE Micro 1986; Mariani 1989; Rabiner 1993/1989; Klevans 1997; Holmes 2001; Jelinek 1997; Katagiri; Ravishankar thesis).

---

3. § 102 anticipation — direct answer

No cited reference, alone, anticipates any claim of the '377 patent. Every adjudicated ground was § 103 obviousness, not § 102. Confirming this is important because the task asks per-claim § 102 mapping:

Bottom line for the § 102 question: there is no anticipation case. The invalidation theory is obviousness over Jiang + Smyth (independent claim 1), escalating to Jiang + Smyth + Nguyen/Boike/Baumgartner for dependent claims 2–6.

---

4. Corrections to earlier sections

1. Ground correction (reconfirming the PTAB section's flag). Any statement that this patent's claim 1 was held obvious over "Jiang, Baumgartner, and Brown, with or without Smyth" is the ground for the sibling '277 patent (IPR2023-00034/01193), not the '377. For the '377, claim 1 = Jiang + Smyth. "Brown," "Vensko," and "Kazeroonian" are '277 references and should not be mapped onto the '377 claim set.
2. The earlier "Obviousness" section analyzed the specification's own Background art (Davis et al., Sukun Kim et al., Ney et al.). None of those three was the art used to invalidate the '377. They remain useful as of the priority date for a § 103 narrative, but the operative adjudicated art is the IPR set above.
3. Note the recurring 6,374,219 (Jiang) vs. 6,374,220 (Kao) adjacent-number trap — the '377 front page cites Kao 6,374,220; the IPR uses Jiang 6,374,219. These are different references; keep them separate in any filing.

---

5. Caveats and confidence

• High confidence: the IPR grounds and reference combinations (Jiang / Smyth / Nguyen / Boike / Baumgartner) and their claim mapping, taken from the Schmandt Declaration TOC and the FWD; the Baumbgartner and Boike bibliographic details quoted from the record; Simar = US 6,182,203 (Jan. 30, 2001).
• Medium confidence: the exact grant/publication days for Jiang (6,374,219), Smyth (5,819,222), Nguyen (6,879,954), and Boike (6,959,376) — I have month/year but did not verify the precise dates; confirm at Patent Center before citing.
• Low confidence / unverified: the subject matter of every non-family front-page citation (Dulong, Kao, Wells, Kato, Zhao, Wang, EP 1 178 466, GB 2 333 172). The 8,026,890 vs. 8,036,890 OCR ambiguity for the Catchpole document.
• Incompleteness: the retrieved front page was a partial OCR; the '377 may cite additional U.S. patents/applications I did not capture. The complete "(56) References Cited" and the "Cited By" list should be pulled directly from the patent PDF at patentimages.storage.googleapis.com or from Patent Center (patentcenter.uspto.gov, application 14/788,164) to close this gap. I flagged rather than fabricated.
• Status context (from prior sections): claims 1–6 were held unpatentable 2024-06-11 and affirmed 2026-09-25 (CAFC No. 24-2205); the patent is recorded Expired – Lifetime (anticipated 2025-09-14). Prior-art analysis of this patent is therefore of historical/defensive value — there are no surviving claims to invalidate.

Primary sources used: the '377 patent front page as reproduced in Apple/Amazon IPR Exhibit 1001 (Docket Alarm, IPR2023-01194); Schmandt Declaration (Apple Ex. 1003) TOC and claim-element map (Docket Alarm, IPR2023-00035); PTAB FWD (2024-06-11) and Federal Circuit No. 24-2205 opinion, https://law.justia.com/cases/federal/appellate-courts/cafc/24-2205/24-2205-2026-09-25.html ; Docket Alarm exhibit list for IPR2023-00035.

I'll build on the prior-art section, but first let me try to pin down the exact claim language for claims 1–6, since the earlier sections flagged that gap and the obviousness mapping depends on it.

Excellent — I retrieved the actual claim-listing appendix from Apple's petition. Let me pull claims 3–6 and the petition's motivation-to-combine reasoning.

I now have the verbatim claim set (from Apple's petition claim-listing appendix and the Schmandt Ex. 1003 claim listing) plus the Board's claim-by-claim reasoning. That closes the gap the earlier sections flagged and lets me map the art claim-element by claim-element rather than by inference. Below I build on the Prior-Art section rather than restating it.

---

Obviousness Analysis — U.S. Patent No. 10,062,377 B2 under 35 U.S.C. § 103

Scope discipline. I searched and analyzed the exact identifier 10062377 (US 10,062,377 B2, Distributed pipelined parallel speech recognition system, Zentian Limited). I did not conflate DE 10062377 A1 or the sibling family members ('277 / '319 / '789 / '140).

Governing statute. The '377 claims priority to GB 0420464.0, filed 2004-09-14 (confirmed on the face of the patent). It is therefore a pre-AIA case, and the Board analyzed it under pre-AIA § 103(a). Apple Inc. v. Zentian Ltd., No. IPR2023-00035, Paper 34, 2024 WL 2946335 (P.T.A.B. June 11, 2024). I note where a reference's own status as prior art is date-sensitive.

Date conflict (flagged once more, for the record). Your task states the current date is 2026-04-26, but the search results returned an operative Federal Circuit decision dated 2026-09-25 (Zentian Ltd. v. Apple Inc., No. 24-2205). I continue to report the searched events as ground truth per the operating rules.

---

1. The legal framework applied

Obviousness is resolved on the Graham factors: (1) scope and content of the prior art; (2) differences between the claim and the art; (3) level of ordinary skill; and (4) objective indicia. Graham v. John Deere Co., 383 U.S. 1, 17–18 (1966). Under KSR Int'l Co. v. Teleflex Inc., 550 U.S. 398 (2007), the analysis "need not seek out precise teachings directed to the specific subject matter of the challenged claim, for a court can take account of the inferences and creative steps that a person of ordinary skill in the art would employ." Id. at 418. The Board applied exactly this formulation and the In re Magnum Oil Tools evidentiary standard (petitioner must articulate specific reasoning, not conclusory statements). (The Board's statement of law is reproduced verbatim in the sibling '277 FWD retrieved via the PTACTS link.)

Critically for this patent, the claim is a functional-decomposition / architecture claim: it recites the three classical speech-recognition stages (feature extraction → acoustic distance scoring → lexical search) and assigns each to a nominated hardware resource. That is the classic KSR fact pattern — "the combination of familiar elements according to known methods … when it does no more than yield predictable results." KSR, 550 U.S. at 416.

---

2. The claims under analysis (verbatim, as challenged)

Recovered from Apple's Petition claim-listing appendix (IPR2023-00035, Paper 1) and Ex. 1003 (Schmandt Decl.) claim listing. This is the actual issued claim text — it materially differs from the claim characterization in the earlier "Patent summary" section (see §14).

Claim 1 — A speech recognition circuit, comprising:

• 1(a) a first programmable device programmed to calculate a feature vector from a digital audio stream, wherein the feature vector comprises a plurality of extracted and/or derived quantities from said digital audio stream during a defined audio time frame;
• 1(b) a second programmable device programmed to calculate distances indicating the similarity between a feature vector and a plurality of acoustic states of an acoustic model, wherein said feature vector is received by the second programmable device after it is calculated by the first programmable device; and
• 1(c) a third programmable device programmed to identify spoken words in said digital audio stream using Hidden Markov Models and/or Neural Networks, wherein said word identification uses one or more distances that were calculated by the second programmable device, wherein said identification of spoken words uses one or more distances calculated from a first feature vector; and
• 1(d) a search stage for using the calculated distances to identify words within a lexical tree, the lexical tree comprising a model of words.

(Element lettering follows Ex. 1003's consolidated a–d format; Apple's appendix prints the same text as eight sub-elements 1(a)–1(h). Substance identical.)

Claim 2 — … wherein the time period taken for said identification of spoken words overlaps the time period taken for said second programmable device to calculate one or more distances from a second feature vector.
Claim 3 — … wherein the first, second and third programmable devices are physically distinct processing circuits.
Claim 4 — … wherein said physically distinct processing circuits are all formed on a single integrated circuit.
Claim 5 — … wherein the first and third programmable devices comprise one physical uni-processor or multi-processor, and the second programmable device is a physically distinct processing circuit.
Claim 6 — … wherein said one physical uni-processor or multi-processor and said second programmable device are all formed on a single integrated circuit.

Structural read: claim 1 = the 3-stage pipeline split across three "programmable devices." Claim 2 = pipelining/overlap (the actual "distributed pipelined" feature). Claims 3–4 = three physically distinct circuits / all on one IC. Claims 5–6 = the hybrid variant (front-end + search share a processor; the distance engine is separate; all on one IC). Claims 2 and 3 are the ancestors of the two alternative architecture sub-trees (3→4 and 5→6).

---

3. Level of ordinary skill (POSITA)

The Board adopted, essentially unmodified: a master's degree in computer engineering, computer science, electrical engineering or a related field with ≥2 years of speech-recognition experience, or a bachelor's degree in the same fields with ≥4 years of experience, with additional education/experience substitutable. (Reproduced in Apple's Petition § II.D and in the Ex. 1003 declaration.)

This is a software/systems-level skill — not a semiconductor-fabrication skill. That framing matters: it is why the Board found the hardware-architecture limitations (separate devices, single IC) obvious when a POSITA is treated as a consumer of integrated circuits fabricated by others (see §12 and the sibling-appeal discussion).

---

4. The prior-art corpus for the § 103 analysis

Building on the Prior-Art section rather than repeating it, the operative references and their claimed-element function are:

Admitted background art (from the '377's own specification): Davis et al., A low-power accelerator for the SPHINX 3 speech recognition system (Nov. 2003); Sukun Kim et al., Hardware Speech Recognition in Low Cost, Low Power Devices (Spring 2003); Ney et al. (IEEE SP Magazine, Sept. 1999). See §11.

---

5. Claim 1 — obvious over Jiang + Smyth

(a) What each reference discloses against the elements

Jiang alone maps virtually every element except the three-distinct-device requirement. Jiang teaches two distinct devices (array processor for extraction; "dedicated speech recognition processor" for tree search — 6:38–42). Smyth supplies the third distinct device and, more importantly, expressly locates the distance calculation on a physically distinct unit (the classifier with its own processor 341).

(b) Motivation to combine — articulated with rational underpinning

A POSITA at the 2004 priority date would have been motivated to arrange Jiang's stages onto separate programmable devices, including a dedicated distance-calculating device, for at least these reasons:

1. Jiang itself points the way. Jiang expressly discloses multi-processor and dedicated-processor implementations ("tree search engine 74 is preferably implemented in [a CPU] … or … a dedicated speech recognition processor"; feature extraction on a separate array processor). Combining a reference with itself-plus-a-known-alternative is the weakest form of "combination."
2. Smyth expressly motivates the architecture. Smyth teaches the recognizer's stages on distinct programmable devices, including a classifier processor realized as a commercially available DSP (Motorola DSP56000) — i.e., an off-the-shelf, physically distinct, programmable distance engine.
3. The stated problem is the same in both references: the acoustic-scoring stage is computationally intensive, and offloading it to a dedicated/parallel device reduces load on the general processor. That is the classic KSR "known need / design incentive" rationale.
4. Reasonable expectation of success. Both references implement the same well-understood HMM acoustic-scoring operation; substituting a dedicated DSP for the scoring stage was, on the record, an ordinary architectural choice with predictable results. The Board's claim-1 conclusion expressly rests on a reasonable-expectation finding.

Net: on the Board's construction of "feature vector" (which the Federal Circuit affirmed — see §13), Jiang + Smyth discloses every limitation of claim 1, with articulated motivations. Claim 1 is obvious.

---

6. Claim 2 — obvious over Jiang + Smyth + Nguyen

Claim 2's added limitation is, in substance, pipelined/concurrent operation: the search-stage identification for frame n overlaps in time with the second device computing distances for the next feature vector (frame n+1).

Nguyen teaches exactly this contemporaneity: "performing a similarity measure (i.e., distance calculation) while contemporaneously performing the search step" (as quoted in Zentian's POR §V and the Board's opinion).

Motivation: in a multi-device pipeline, overlapping stages is the canonical way to raise throughput and avoid idle processing elements — the very benefit that motivates splitting the stages in claim 1. Once you have physically separate devices, running them concurrently costs nothing and buys throughput. This is KSR "predictable variation."

Patentee's counterargument (rejected): Zentian argued that Jiang's feedback-based pruning (search results fed back to the scoring stage) is technically incompatible with Nguyen's contemporaneous processing, so there was no motivation / no reasonable expectation of success. The Board rejected this on three independent grounds:

1. Pruning can be performed at the search stage only, without feeding back to the distance stage (Schmandt Supp. Dec.);
2. Even if feedback is retained, it can be applied with a one-frame delay, which "is not a delay in the speech recognition itself";
3. Forgoing feedback-based pruning in favor of pipelining is itself an obvious engineering trade-off, supported by art (Mathew & Ravishankar forgo feedback in favor of pipelining).
   Claim 2 is obvious.

---

7. Claim 3 — obvious over Jiang + Smyth + Nguyen

Claim 3 = the three devices are physically distinct processing circuits. This is subsumed by the Jiang + Smyth combination already used for claim 1: Smyth expressly discloses the three recognizer stages on three distinct programmable devices, and each is a distinct physical circuit. Nguyen (already combined for claim 2, from which claim 3 depends) adds nothing adverse. Claim 3 is obvious. No independent motivation analysis is needed beyond claim 1 — it is an inherent structural consequence of the combination.

---

8. Claim 4 — obvious over Jiang + Smyth + Nguyen + Boike

Claim 4 = the physically distinct circuits are all formed on a single integrated circuit.

Boike teaches "a system-on-a-chip (SoC) integrated circuit 10 containing multiple digital signal processors (DSPs)" (2:52–54), with DSP subsystems 18–21 on the single chip.

Motivation (articulated by Boike itself, per the Board): placing multiple DSPs on a single IC reduces space and cost and conserves power — precisely the design incentives that drive embedded speech recognition (the '377's own stated field). This is the textbook KSR "market forces / design incentive" rationale, and Boike supplies it expressly.

Patentee waived the point. Zentian "did not separately address this ground … other than to argue that claim 4 depends from claim 2 and should be patentable based on its dependence." Dependent-claim patentability cannot be sustained on the strength of an invalidated parent. Claim 4 is obvious.

---

9. Claim 5 — obvious over Jiang + Smyth + Nguyen + Baumgartner

Claim 5 is the hybrid architecture: front-end and word-ID devices share one physical uni-processor or multi-processor, while the distance device is physically distinct.

Baumgartner is directly on point: it teaches that, to reduce computational load, the distance calculation is performed on a physically distinct device, while feature-vector calculation and word identification are performed on a first processor core 1910 (Ex. 1007 ¶¶ 31–33, 156–158, Fig. 19). That is claim 5's exact topology.

Motivation: Baumgartner expressly states that this configuration "reduc[es] the computational load on the system processors." Jiang already discloses a multi-processor CPU (Fig. 2), so placing the computationally intensive distance scoring on a distinct circuit while leaving extraction and search on the (multi-)processor is a direct application of Baumgartner's teaching to Jiang's system, with predictable results. Claim 5 is obvious. (Again, Zentian offered no separate argument for claim 5 beyond its reliance on claim 2.)

---

10. Claim 6 — obvious over Jiang + Smyth + Nguyen + Boike + Baumgartner

Claim 6 = claim 5's topology all on a single IC. This is the union of two already-motivated teachings: Baumgartner supplies the why for the distinct distance device (§9); Boike supplies the why for the single-IC integration (§8 — space, cost, power). Combining two references that each independently motivate their own contribution, in the same field, with predictable results, is the paradigm KSR case. Claim 6 is obvious.

---

11. Additional / alternative § 103 grounds not asserted in the IPR

These strengthen the record but should be labeled as meta-record hypotheses, not adjudicated holdings:

• Brown (Ex. 1036) for the pipelining limitations (claim 2 and the "pipelined" narrative). Brown expressly teaches that "[t]he functionality of each block of the recognition system may be performed by a respective different processor," and that "the stages are pipelined … so that the operations are synchronized with the rate of frame generation." Brown was used successfully against the sibling '277 patent for the "data is pipelined" limitation. It would supply an independent, alternative motivation for claim 2's overlap limitation and could substitute for (or supplement) Nguyen — useful if a future tribunal were to question Nguyen's § 102/§ 103 status. Flagged as inference.
• The '377's own admitted background art. The specification admits Davis et al. (SPHINX-3 special-purpose co-processor for the computationally expensive Gaussian output probabilities, Nov. 2003), Sukun Kim et al. (special-purpose processing elements per phone/state, Spring 2003), and Ney et al. (language-model lookahead, Sept. 1999). Admissions in the specification are prior art against the applicant. Davis in particular is a third-party admission that a dedicated distance-scoring co-processor was known and desirable — corroborating the core claim-1/claim-5 motivation. These were not the grounds the Board used, but they independently undercut any argument that the "dedicated distance engine" concept was inventive.
• The face-of-patent citations (Dulong, Kao US 6,374,220, Wells, Kato, Zhao US 2003/0061046, Wang US 2004/0111264, EP 1 178 466, GB 2 333 172) were never asserted as grounds and their subject matter remains unverified per the Prior-Art section. I do not rely on them. Note the adjacent-number trap: the '377 front page cites Kao US 6,374,220; the IPR relies on Jiang US 6,374,219 — different documents.

---

12. Objective indicia (secondary considerations)

None in evidence. The IPR record shows Zentian's defense was built entirely on technical-infradigibility and motivation/expectation-of-success arguments (codeword ≠ feature vector; feedback-based pruning incompatible with pipelining) — i.e., attacking the prima facie case, not proffering objective evidence of non-obviousness. There is no evidence of record of:

• Unexpected results (the combination produced exactly the expected throughput/efficiency gains);
• Long-felt but unmet need satisfied by the claims;
• Industry praise, licensing, copying, or failure of others;
• Commercial success with nexus — Zentian shipped no product embodying the claims (only a Sphinx-based demonstrator reporting ~2× speedup and an FPGA/emulation prototype, per its own complaint). Without a product, no presumption of nexus attaches, and there is no commercial-success argument to make.

The absence of secondary considerations leaves the Graham factor (4) empty, so the claim-by-claim KSR analysis controls.

---

13. The patentee's principal rebuttals, and why they fail

The Federal Circuit's affirmance means these rebuttals are now finally resolved against the patentee for the '377 claim set. (Citation: Zentian Ltd. v. Apple Inc., No. 24-2205 (Fed. Cir. Sept. 25, 2026) (nonprecedential, Dyk/Mayer/Prost) — https://law.justia.com/cases/federal/appellate-courts/cafc/24-2205/24-2205-2026-09-25.html.)

---

14. Conclusion — § 103 disposition

Every claim of the '377 patent is obvious under 35 U.S.C. § 103(a):

No ground was § 102; no single reference anticipates (confirming the Prior-Art section's conclusion). The invalidation is pure § 103.

---

15. Corrections, contradictions, and caveats

Corrections to earlier sections (material):

1. Claim 1 does NOT recite "an audio front end," a "calculating circuit," or a "search stage" as its three operative structures. It recites a first / second / third programmable device plus a "search stage." The earlier "Patent summary" section's element list ("1[audio front end] … 1[calculating circuit] … 1[search stage] …") tracks the '277 patent's claim 14 and the '377 specification's "further aspect" language — not the '377's issued claim 1. Any § 103 mapping that keys on "audio front end"/"calculating circuit" is mapping the wrong claim text. Verify against the issued patent before filing.
2. Preamble variance flagged: the petition appendix prints the preamble as "A speech recognition circuit, comprising:" while Ex. 1003 prints "A speech recognition system comprising:" (and claim 2's body says "The speech recognition system of claim 1"). This is a real discrepancy in the record renderings; the issued patent should be pulled to fix the literal preamble. It does not affect the § 103 analysis (preamble is likely non-limiting here), but do not quote a preamble without confirming it.
3. The Prior-Art section's ground table and the PTAB section's ground table are correct and now corroborated (Jiang + Smyth for claim 1; + Nguyen for 2–3; + Boike for 4; + Baumgartner for 5; + Boike/Baumgartner for 6) — confirmed against Zentian's Notice of Appeal issue list and Apple's Petition table of contents.
4. The recurring "Jiang + Baumgartner + Brown (with or without Smyth)" ground belongs to the sibling '277 patent (IPR2023-00034/01193). Do not transpose it onto the '377.

Caveats requiring verification before reliance:

• Reference dates I could not fully verify: the grant/publication days (and filing dates) for Nguyen US 6,879,954, Boike US 6,959,376, and (for § 102(e) purposes) whether Boike's filing date makes it § 102(e) rather than § 102(b) art as of a Sept. 14, 2004 priority. The Board treated all five as prior art and the parties appear not to have contested it, but confirm at Patent Center before citing.
• Boost from unasserted art (§11) is my inference, not an adjudicated ground. Brown's exact citation (Ex. 1036) and content come from the sibling '277 proceeding.
• Status is now final but not yet certificated. The affirmed FWD disposes of claims 1–6, but formal cancellation occurs on the Director's § 318(b) certificate after the appellate mandate. As of the retrieved record the certificate may not have printed. The patent is also recorded Expired – Lifetime (anticipated 2025-09-14), so there is no unexpired term regardless.
• Date conflict repeated: results through 2026-09-25 post-date the task's stated 2026-04-26.

Primary sources: Apple Petition, IPR2023-00035 Paper 1 (claim listing appendix + Grounds 1–5) — https://www.docketalarm.com/cases/PTAB/IPR2023-00035/Apple_Inc/docs/11-15-2022-Petitioner/Petition_as_filed-1-01__IPR2023_00035_Petition_for_Inter_Partes_Review_US10062377_Patent_FINAL.pdf ; Ex. 1003 (Schmandt Decl.) claim listing; Zentian POR (Paper 19) and corrected POR; Amazon Petition, IPR2023-01194 (identical grounds + claim appendix) — https://www.docketalarm.com/cases/PTAB/IPR2023-01194/Amazon.com_Services_LLC/docs/07-12-2023-Petitioner/Petition_as_filed-2-Petition_for_Inter_Partes_Review_of_US_Patent_No_10,062,377.pdf ; Zentian Notice of Appeal (issue list) — https://www.docketalarm.com/cases/PTAB/IPR2023-00035/Apple_Inc._v._Zentian_Limited/docs/08-09-2024-Patent_Owner/Notice__notice_of_appeal-35-Patent_Owners_Notice_of_Appeal.pdf ; FWD quoted language via the Federal Circuit opinion — https://law.justia.com/cases/federal/appellate-courts/cafc/24-2205/24-2205-2026-09-25.html ; patent text — https://patents.google.com/patent/US10062377/en.