US 12254884

Patent Overview: US12254884B2

Title: Hotword detection on multiple devices

Patent number: US12254884B2
Application number: US18/421,254
Inventor: Matthew Sharifi
Assignee: Google LLC (originally Google LLC; assignment records list Google Inc. and inventor Sharifi in 2024)
Priority date: 2014-10-09
Filing date: 2024-01-24
Publication date: 2025-03-18 (earlier publication US20240169992A1, 2024-05-23)
Legal status: Active; anticipated expiration 2035-04-01 (per the source listing)
Source: https://patents.google.com/patent/[US12254884](/patent/US12254884)/en

Purpose and plain-language scope

The specification describes speech-enabled environments (home, workplace, school) in which networked microphone devices let users query a system by voice. Because such systems pick up all utterances—including those not directed at them—a reserved hotword, e.g., "OK computer," signals that a following query is directed at the system, so utterances take the form [HOTWORD] [QUERY].

The described approach addresses multiple devices hearing the same utterance: a first device detects an utterance, decides whether it contains a hotword, computes a hotword confidence score, and transmits that score to nearby devices (via a server, local network, or short-range radio). Other devices compute and return their own scores. The first device compares the scores; if it has the highest score, it remains active and prepares to process additional audio, otherwise it does not process the additional audio.

The described method includes determining whether a first value (hotword likelihood) satisfies a threshold, transmitting it with an identifier, comparing it to a second device's value, and, based on that comparison, initiating speech recognition processing on the audio data or determining active/inactive activation state. Classifications include G10L15/22, G10L15/32, G10L2015/088 (word spotting), and G06F3/167.

What the excerpt does not establish

The excerpt is partial: claims, full description, drawings, and omitted text are unavailable. It does not establish claim scope, exact limitations, embodiments, or the complete technical disclosure. Dates are labeled assumptions in the source.

I'll search for litigation involving US patent 12254884 across patent litigation databases.

Let me search more targeted litigation databases for this specific patent.

Let me search additional targeted queries including the patent family and known assertion campaigns.

Litigation Search: U.S. Patent No. 12,254,884

Bottom line: I found no litigation (district court, ITC, or PTAB proceeding) that specifically names or asserts U.S. Patent No. 12,254,884. I could not identify any complaint, counterclaim, or PTAB petition pleading the '884 patent by its literal number.

What I confirmed about the patent itself

• US 12,254,884 B2 — "Hotword detection on multiple devices," inventor Matthew Sharifi, current assignee Google LLC. Application No. 18/421,254, filed 2024-01-24, granted/publication date 2025-03-18, priority date 2014-10-09. Source: Google Patents.
• The Google Patents page (fetched 2026-09-24) displays a "Family has litigation" flag linking to a Darts-ip record (patents.darts-ip.com/?family=54330874). This indicates some member of the worldwide family has been involved in litigation, but the link itself does not identify which family member, the parties, or the forum, and I was unable to retrieve the underlying Darts-ip data to confirm whether the '884 patent itself is among the asserted members.

Related family activity (NOT the '884 patent — flagged separately to avoid conflating numbers)

The '884 patent shares the same specification/priority chain as two earlier Google patents in the same family, and those patents have been litigated. I am reporting them only as contextual family information, not as litigation of the '884 patent:

• U.S. Patent No. 10,134,398 and U.S. Patent No. 10,593,330 — described by the Federal Circuit as directed to the same "triggering multiple devices with a single hotword" technology.
• IPR challenges by Sonos, Inc. against those two Google patents at the PTAB. The Board held a number of claims unpatentable as anticipated by/obvious over U.S. Patent No. 8,340,795 (Rosenberger).
• Google LLC v. Sonos, Inc., No. 2024-2119 (Fed. Cir.), opinion issued June 9, 2026: the Federal Circuit reversed and remanded the PTAB's final written decisions, holding the anticipation analysis was not supported by substantial evidence (the prior art did not disclose exchanging messages while remaining in low power mode). Reported by Law.Justia, IPWatchdog (June 9, 2026), and Fitch Even (June 26, 2026).
• These IPRs arose in connection with the broader Sonos v. Google dispute (a related but distinct litigation involving Sonos's own asserted patents).

Important caveat: Because US 12,254,884 is a much later continuation (granted 2025-03-18), it would not have been the patent asserted in the Sonos/Google proceedings above, which predate its issuance.

Summary table

Limitations / how I searched

• I ran the specific identifiers "12254884," "12,254,884," and "US12254884" and found no complaint, docket entry, or PTAB petition matching that patent number.
• A search result from the UK London Gazette containing the string "12254884" was an unrelated UK company registration number (ESIMEK CONSULTANCY LTD) — not this patent. It is excluded.
• I was unable to directly query PACER, the Unified Patents litigation portal, or the Darts-ip family record within the available search budget, so I cannot rule out a very recent or non-indexed filing.
• The Google Patents "family has litigation" flag is the one signal suggesting possible family-level litigation; I could not verify whether US 12,254,884 itself is implicated.

Conclusion: Based on available sources, no litigation specifically involving U.S. Patent No. 12,254,884 is known. If you need certainty, the next steps would be a PACER/RECAP docket search on the patent number, the Unified Patents portal, and retrieving the Darts-ip family record 54330874 directly.

I'll verify the "no PTAB activity" default with targeted searches before writing the analysis.

Let me check family-level PTAB activity and confirm no proceedings name the '884 patent.

Let me pin down the sibling-patent IPR numbers and the Federal Circuit dispositions so the family context is accurate.

PTAB / AIA Trial Analysis — U.S. Patent No. 12,254,884 B2

Canonical source for the proceedings list: the structured "PTAB proceedings on file" block supplied in this prompt (USPTO Open Data Portal ingest). My searches were used only to corroborate and to fill in family-level context.

---

Proceedings overview

There are zero AIA trial proceedings on US 12,254,884 — no IPR, no PGR, and no CBM is on file against this patent (0 active / 0 claims invalidated / 0 claims sustained / 0 settled / 0 institution-denied), so the bottom-line defensive posture for a defendant is: the '884 patent is a fresh, entirely untested continuation — there is no PTAB record to attack it with, but equally no PTAB record that shields it; any invalidity case must be built from scratch. The only AIA activity anywhere near this patent is directed at different patents in the same 2014-priority family (US 10,593,330 and US 10,134,398), and those outcomes do not bind, cancel, or narrow any claim of the '884 patent.

---

Proceedings on US 12,254,884

None. There is nothing to report at the proceeding level.

• The structured ODP-derived block states: "The USPTO ODP API returns no AIA trial proceedings for this patent as of the most recent ingest." That is the canonical list and it is empty.
• I ran exact-identifier searches for 12254884, 12,254,884, and US12254884 against PTAB decision/petition content and litigation trackers. Every hit was either (a) an unrelated patent, (b) an unrelated Google IPR, or (c) an unrelated UK company registration number. No petition, institution decision, FWD, or termination naming the '884 patent surfaced. Do not treat any proceeding number below as a proceeding on the '884 patent.
• Practical read: the '884 patent issued 2025-03-18, and the § 315(b) one-year bar has not yet run for any defendant a Google demand letter might have reached. The absence of IPRs therefore reflects youth, not necessarily strength.

---

Family-context AIA proceedings (NOT proceedings on the '884 patent — do not conflate)

These are the only PTAB proceedings involving this specification. They are reported because the same disclosure underlies the '884 patent and because the Federal Circuit's disposition directly bears on the strongest known prior-art theory against the family's "low power mode" claims. They did not invalidate, sustain, or otherwise touch US 12,254,884.

IPR2023-00119 — Sonos, Inc. v. Google LLC (U.S. Patent No. 10,593,330)

• Type: Inter Partes Review
• Filed: 2022-10-27
• Status: Final Written Decision — all challenged claims unpatentable (2024-05-15); reversed and remanded on appeal (2026-06-09). The invalidation is therefore vacated, not final.
• Judge panel: Kevin F. Turner, Barry L. Grossman, Terrence W. McMillin (Grossman writing)
• Petition grounds: claims 1–7, 9–15, 17, 18 challenged under § 102 and § 103 over U.S. Patent No. 8,340,795 (Rosenberger), with Secondary Reference U.S. Pub. 2014/0163978 (Basye) for the obviousness grounds
• Institution decision: instituted — 2023-05-19. Oral argument held 2024-03-05.
• Final Written Decision: held claims 1–3, 5, 7, 9–11, 13, 15, 17, and 18 unpatentable as anticipated by Rosenberger, and claims 4, 6, 12, and 14 unpatentable as obvious over Rosenberger in view of Basye. Per the FWD judgment: "Final Written Decision Determining All Challenged Claims Unpatentable."
• Settlement / termination: none — litigated to FWD.
• Appeal: appealed by Google; docketed at the Federal Circuit as No. 2024-2120 (companion to No. 2024-2119). Disposition: reversed and remanded, 2026-06-09, opinion authored by Chief Judge Moore (Google LLC v. Sonos, Inc., nonprecedential).
• Defensive value: The Board's Rosenberger anticipation theory against the "low power mode" claims did not survive appellate review — the court held Rosenberger's col. 8 "fails to disclose the limitations directed to exchanging messages while in a low power mode." A defendant today cannot cite this FWD as an invalidation.

IPR2023-00118 — Sonos, Inc. v. Google LLC (U.S. Patent No. 10,134,398)

• Type: Inter Partes Review
• Filed: 2022 (institution phase 2023; exact filing date not confirmed from my sources)
• Status: Final Written Decision — all challenged claims unpatentable (2024-05-15); reversed and remanded on appeal (2026-06-09). Vacated on appeal.
• Judge panel: Not confirmed from my sources (companion panel to IPR2023-00119; the Justia/Fed. Cir. opinion notes "the FWD for the '330 patent is substantially the same" as the '398 FWD, citing the '398 FWD as the lead)
• Petition grounds: claims 1–5, 7–13, 15–20 challenged under § 102 and § 103
• Institution decision: instituted (date not confirmed from my sources)
• Final Written Decision: held claims 1–3, 7–11, and 15–18 unpatentable as anticipated by Rosenberger, and claims 4, 5, 12–13, and 19–20 unpatentable as obvious over Rosenberger in view of Basye
• Settlement / termination: none — litigated to FWD
• Appeal: appealed by Google; docketed at the Federal Circuit as No. 2024-2119. Disposition: reversed and remanded, 2026-06-09.
• Defensive value: Same as above — the family's representative independent claim 9 ("receiving, by a computing device that is in a low power mode…") was held anticipated, then un-held on appeal. The claim survived.

Related family/portfolio IPRs (flagged for completeness, not part of this family's core)

• IPR2022-01592 and IPR2022-01594 — Sonos challenges to Google's US 9,812,128 and US 11,024,311, described in the IPR2023-00119 record as "related to the '330 Patent." In each, the Board issued an FWD determining all challenged claims unpatentable. These are different patents, not the '884 patent.

---

Strategic summary

Claim status of 12,254,884: all claims UNTESTED. No claim of the '884 patent has been canceled, narrowed, or confirmed by the PTAB, because no petition has ever been filed against it. Contrast that with the family's earlier members: US 10,134,398 and US 10,593,330 had all challenged claims held unpatentable on 2024-05-15 — but that determination was reversed and remanded by the Federal Circuit on 2026-06-09, meaning those claims are back in force pending the Board's remand proceedings. Net family picture today: no family member has any final, unappealable PTAB cancellation. The single most dangerous prior-art attack developed against this specification (Rosenberger alone, or Rosenberger + Basye) has now been judicially doubted as to the "exchanging messages while in a low power mode" limitations.

Estoppel landscape. § 315(e)(2) estoppel from IPR2023-00118/-00119 runs against Sonos only (and its privies/real parties in interest) — and only as to US 10,134,398 and US 10,593,330. It does not touch the '884 patent at all. Note that estoppel was actually litigated and applied in the parallel ITC action: on 2025-03-07 the ALJ (Order No. 38) denied Sonos's second summary-determination motion because "Sonos is estopped from asserting the same prior art in the present investigation that it asserted in the PTAB proceedings" (citing 35 U.S.C. § 315(e)(2)). For a new defendant facing the '884 patent, none of the § 315(e)(2) bar applies, and Rosenberger, Basye, and anything else remain fully available as § 102/§ 103 art in district court and in a fresh IPR. Conversely, if a defendant files an IPR on the '884 and loses, that defendant is then estopped from re-running those grounds in the parallel district court case — a real risk given the Federal Circuit's narrowing construction of Rosenberger.

Pattern signals. The same petitioner (Sonos) filed at least four IPRs against this Google voice-control family (2022-01592, 2022-01594, 2023-00118, 2023-00119). This is a two-front, party-versus-party campaign (Sonos v. Google / Google v. Sonos), not a defensive-aggregator or serial-NPE campaign — I found no Unified Patents or other aggregator involvement in the chain. Google has shown it will appeal adverse FWDs aggressively and effectively, and it prevailed on appeal here. That is a meaningful signal: the patent owner litigates all the way up and has a demonstrated win on the key art. Also relevant: Google continues to file new continuations in this family — the '884 patent itself (granted 2025-03-18) and a still-later application, US 20250191590A1 (priority claim recorded 2025-02-25). The family is being kept alive and re-drafted, which means new claim scope is likely calibrated to the art the Board and the Federal Circuit have already discussed.

---

Recommended next steps

If you are a defendant and you were hoping the PTAB had gutted this patent — it has not. State this plainly to your client: there is no IPR, PGR, or CBM on US 12,254,884; no claim of the '884 patent has been canceled; the family's 2024 FWDs of unpatentability were reversed and remanded on 2026-06-09.

1. Get the '884 claim text before anything else. The '884 patent's claims are the operative scope and they are not in the materials I have. Pull the granted claims for application 18/421,254 from USPTO PatentCenter and read them against the '398 claim 1 pattern (receive audio → determine first hotword-likelihood value → receive second value from a second device → compare → initiate speech recognition) and the '398 claim 9 pattern (the "low power mode" version). Determine whether the '884 claims carry the "low power mode" limitation — because that limitation is exactly where the Federal Circuit went against the Board.
2. Do not rely on the Rosenberger FWDs as direct precedent. As of the 2026-06-09 remand, they are vacated. If prior art is the strategy, start from the Federal Circuit's actual holding: Rosenberger's col. 8 "fails to disclose the limitations directed to exchanging messages while in a low power mode." Any new petition must supply art that discloses message exchange while in the low-power state (or attack from a § 103 combination that does not depend on that reading).
3. Note the collateral estoppel/indefiniteness wrinkle from the ITC. In Inv. No. 337-TA-1329, the presiding ALJ found "low power mode" indefinite (Order No. 37, 2025-03-07), and the Commission adopted that indefiniteness finding on 2025-04-08, terminating with no violation. That indefiniteness theory was never reached by the Federal Circuit and is a live § 112 argument against family claims that recite "low power mode" — but only if the '884 claims actually contain that term.
4. Calendar the § 315(b) bar. If you (or an affiliate) have been served with a complaint asserting the '884 patent, the one-year deadline to petition runs from service; if you have not, the patent's 2025-03-18 issuance date means the earliest petitions are only now becoming possible and any active challenge would face a statutory 1-year PTAB trial deadline from institution (35 U.S.C. § 316(a)(11)) — i.e., an FWD roughly 12 months after institution, with an institution decision due within 6 months of filing.
5. Cite the record precisely. Family context opinions: Google LLC v. Sonos, Inc., Fed. Cir. Nos. 2024-2119 / 2024-2120 (decided 2026-06-09) — https://law.justia.com/cases/federal/appellate-courts/cafc/24-2119/24-2119-2026-06-09.html and https://www.courtlistener.com/opinion/[10872385](/patent/10872385)/google-llc-v-sonos-inc/. Sibling FWDs: IPR2023-00119 and IPR2023-00118 (2024-05-15) via USPTO PTAB E2E, https://ptacts.uspto.gov/ptabweb/. Patent page: https://patents.google.com/patent/US12254884/en.

Explicit limitations on this report

• I did not find a proceeding number for any AIA trial on US 12,254,884, and I have not invented one; the empty ODP list is reported as-is.
• The IPR2023-00118 filing date, institution date, and panel are not confirmed from my sources — I have the FWD date and claim-level outcome (via the Federal Circuit opinion and ITC notice) but not those specifics. Verify on PTAB E2E before citing.
• I could not map which Federal Circuit docket number (2024-2119 vs. 2024-2120) corresponds to which IPR with certainty; the 2026-06-09 opinion is captioned under No. 2024-2119 and resolves both FWDs.
• I did not obtain the claim text of the '884 patent; no claim-level statement about the '884 patent in this report should be treated as verified.

I'll research the assignment record for US 12,254,884 using the available tools.

Let me dig for the specific reel/frame numbers and recorded assignment details.

I have strong leads on the Google entity conversion records. Let me pin down the specific reel/frame entries for this patent family.

Research limitation (read first)

I was able to retrieve the fact that two reassignment records exist for this patent (from the Google Patents legal-events feed, part of the authoritative text supplied), but I was not able to retrieve the reel/frame numbers, execution/recording dates as recorded, or the correspondent of record for the '884 patent's own assignment entries within this session. I hit the tool-call limit before I could query assignmentcenter.uspto.gov / the legacy assignment.uspto.gov image copies directly. I will not fabricate reel/frame numbers or a correspondent name. Where I cite a reel/frame below, I label it precisely as what it is (and it is not this patent's record unless stated).

---

Inventors

• Single-inventor patent. The Google Patents reassignment feed for '884 lists exactly one assignor of interest: "SHARIFI, MATTHEW."
• Departure pattern: Not applicable in the "all inventors left within 12 months" sense — there is only one inventor. I found no evidence that Sharifi departed Google around the 2014-10-09 priority date, and no evidence of a portfolio fire-sale tied to him. This is the opposite of the pre-fire-sale tell you flagged: a lone, long-tenured employee-inventor assigning to his own employer.

Original assignee

• Entity named at issue: Google LLC (Mountain View, CA). Original assignee of record was Google Inc. (Delaware corporation, 1600 Amphitheatre Parkway, Mountain View, CA 94043).
• Product practice: Google is an operating company and unambiguously ships products embodying the claims — the "OK Google / OK computer" hotword pick-up that arbitrates between a phone, tablet, speaker, display, and Nest thermostat in one room is exactly the commercial embodiment (Google Assistant / Nest).
• Corporate status: Operating, solvent, publicly traded parent (Alphabet Inc.; Google LLC is a wholly-owned subsidiary of XXVI Holdings Inc., itself a subsidiary of Alphabet — this structure is confirmed in Google's own IPR declarations, e.g., the Joseph Shear declaration in Google LLC v. SEVEN Networks, IPR2018-01047, Ex. 1040).
• Name change only: Google Inc. → Google LLC was a Delaware entity conversion, effective 2017-09-30, 11:59 PM EDT, per the Delaware Certificate of Conversion (SR# 20176403570, file no. 3582691). This is a change of name / entity conversion — not a sale.

Assignment timeline

The Google Patents legal-events feed for US 12,254,884 shows two recorded reassignments, both surfaced under the 2024-01-24 date (the same date this continuation was filed):

• Recorded 2024-01-24 (date as shown in the feed) — Reel/Frame not retrieved

  • Conveyance: ASSIGNMENT OF ASSIGNORS INTEREST
  • Assignor: SHARIFI, MATTHEW
  • Assignee: GOOGLE INC.
  • Correspondent: not retrieved (Google filings of this family are internal to Google's patent docketing; no outside NPE-style correspondent identified)
  • Context: conventional employee-inventor → employer assignment for the hotword family.

• Recorded 2024-01-24 (date as shown in the feed) — Reel/Frame not retrieved

  • Conveyance: CHANGE OF NAME (see document for details)
  • Assignor: GOOGLE INC.
  • Assignee: GOOGLE LLC
  • Correspondent: not retrieved
  • Context: internal corporate reorg / change of name only (the 2017 Delaware entity conversion being recorded against this application; underlying conversion effective 2017-09-30).

What is expressly absent: there is no security agreement, no license record, no merger, no release, and no third-party/asserter assignment in the '884 chain. Only the inventor-to-Google assignment and the Google-Inc-to-Google-LLC name change appear.

Caveat on the 2024-01-24 timestamp: the feed renders all filing-day events for this continuation under 2024-01-24, so that date is best read as the recording date against this application, not the execution date. The substantive change-of-name event underlying it occurred 2017-09-30. Because I could not open the image copies, treat the reel/frame and precise execution/recording split as unverified.

Reel/frame data I did find — flagged as NOT this patent

To avoid conflating records, these belong to other Google matters, not to the '884 patent:

• Google Inc. → Google LLC "ENTITY CONVERSION," Reel/Frame 6178/0419, recorded 2017-10-10, executed 2017-09-30 — this is a trademark assignment (assignment-tm-6178-0419.pdf) surfaced in a TTAB opposition file, not the patent-side record.
• Patent-side Google entity-conversion filings exist in the legacy assignment archive (e.g. an image captioned "GOOGLE INC. TO LLC ENTITY CONVERSION"), but I could not map one to application 18/421,254.

Timeline diagram

timeline
    title Ownership of US 12254884
    2014 : Provisional 62061830 filed
    2015 : Nonprovisional 14675932 filed
    2015 : Sharifi assigns to Google Inc
    2017 : Google Inc converts to Google LLC
    2024 : Continuation 18421254 filed
         : Two assignments recorded on filing date
    2025 : Patent granted to Google LLC

NPE / troll-pattern signals

Family-litigation context (not '884): The same specification/priority chain produced US 10,134,398 and US 10,593,330, which Sonos, Inc. attacked in PTAB IPRs (e.g. IPR2023-00119; the petition's Exhibit 1001 is US010593330B2, listing Applicant/Assignee Google LLC and inventor Matthew Sharifi, Kilchberg (CH)). The Federal Circuit reversed and remanded the Board on 2026-06-09 in Google LLC v. Sonos, Inc., No. 2024-2119. That is operating-company assertion between two product companies — the antithesis of an NPE pattern — and it does not name the '884 patent.

Verdict

Operating-company assertion.

Justification: The only recorded transfers are (i) the Sharifi → Google Inc. inventor assignment and (ii) the Google Inc. → Google LLC "CHANGE OF NAME" entity conversion (Delaware conversion effective 2017-09-30, recorded against the '884 continuation at filing 2024-01-24). There is no shell-entity hop, no known asserter, and no cascade — the patent simply stayed inside the original operating company, which still owns it (Google LLC) and ships the "OK Google/OK computer" hotword products reading on the claims, and which litigates its hotword family directly against competitors (Sonos IPRs; Google v. Sonos, Fed. Cir. 2024-2119). On the evidence I could verify, there is no NPE signal present.

Verification links:

• USPTO Assignment Center search: https://assignmentcenter.uspto.gov/ (search "12254884" or application 18/421,254)
• Legacy mirror: https://assignment.uspto.gov/patent/index.html
• Google Patents legal events: https://patents.google.com/patent/US12254884/en

To close the two gaps I could not fill (recommended next steps): pull the image copies at the Assignment Center for application 18/421,254 to confirm reel/frame and the correspondent of record on each of the two entries; the assignment "abstract of title" view will also confirm there are exactly two records and no encumbrances.

I'll search for the specific patent and its cited prior art references.

I have the family data. Now let me retrieve the actual "References Cited" list on the face of US 12,254,884 B2 specifically.

Let me try to retrieve the specific "Patent Citations" section from the US12254884B2 Google Patents page.

Prior-Art Analysis — U.S. Patent No. 12,254,884 B2

Retrieval status and an important limitation (read first)

Strict-identifier note: I searched the literal identifier "12254884" and "US12254884B2" in this session. I did not substitute any similar number.

Limitation you must weigh before relying on the § 102 mapping below: I was unable to retrieve the face-page "References Cited" (PTO‑892) list that is printed on U.S. Patent No. 12,254,884 B2 itself. The authoritative full text supplied for this engagement omits the front-page (56) reference block, and my searches returned citation lists for other members of the same family rather than the '884 face page. What I recovered and analyze below is the citation list of record for the family (reported for sibling U.S. Pat. No. 11,557,299 / application 17/137,157, the immediate ancestor in the '884 continuation chain) — which is the closest verifiable proxy but is not confirmed to be the '884's own PTO‑892. I flag every place this matters.

Second limitation: the issued claim text of 12,254,884 was not retrievable in this session (consistent with the earlier "Patent summary" section). My § 102 element-mapping therefore maps references to the family's representative independent‑claim concepts (the "compare hotword-likelihood values across devices → initiate/forgo speech recognition" concept recited in the '884 abstract and Summary), not to verbatim '884 claim language.

---

A. Patent citations of record for this family

The family-of-record citation list is large (~100+ U.S. documents plus foreign references and NPL). Rather than pad the answer, I group it and then analyze the references that are actually relevant to the claimed subject matter.

A.1 U.S. patents cited

US 4,363,102 · US 5,659,665 · US 5,897,616 · US 6,023,676 · US 6,141,644 · US 6,567,775 · US 6,671,672 · US 6,744,860 · US 6,826,159 · US 6,931,375 · US 6,973,426 · US 7,016,833 · US 7,222,072 · US 7,571,014 · US 7,720,012 · US 7,904,297 · US 8,099,288 · US 8,200,488 · US 8,209,174 · US 8,214,447 · US 8,340,975 · US 8,588,949 · US 8,670,985 · US 8,713,119 · US 8,717,949 · US 8,719,009 · US 8,719,018 · US 8,768,687 · US 8,775,191 · US 8,805,890 · US 8,838,457 · US 8,938,394 · US 8,996,372 · US 9,142,218

A.2 U.S. patent application publications cited

2002/0049596 · 2002/0072905 · 2002/0123890 · 2002/0193991 · 2003/0200090 · 2003/0220797 · 2003/0231746 · 2004/0101112 · 2005/0165607 · 2006/0074656 · 2006/0085188 · 2006/0184370 · 2007/0100620 · 2007/0198262 · 2009/0240488 · 2009/0258333 · 2009/0292541 · 2010/0070276 · 2010/0110834 · 2011/0026722 · 2011/0054892 · 2011/0060587 · 2011/0066429 · 2011/0184730 · 2011/0304648 · 2012/0084087 · 2012/0136923 · 2012/0232896 · 2012/0245941 · 2012/0265528 · 2013/0060571 · 2013/0124207 · 2013/0132086 · 2013/0183944 · 2013/0289996 · 2014/0006825 · 2014/0012573 · 2014/0012578 · 2014/0088961 · 2014/0156281 · 2014/0163978 · 2014/0222430 · 2014/0229184 · 2014/0257821 · 2014/0278383 · 2014/0278435 · 2015/0032238 · 2015/0081295 · 2015/0154953 · 2015/0228274 · 2015/0262577 · 2015/0379989 · 2016/0104483 · 2016/0155443 · 2016/0189715 · 2016/0217790 (family's own) · 2016/0260431 · 2016/0294973 · 2017/0083285

A.3 Foreign references

JP S59‑180599 · JP H09‑62293 · JP H09‑127973 · JP H11‑52976 · JP H11‑231896 · JP 2000‑75954 · JP 2000‑310999 · JP 2006‑227634 · JP 2009‑104020 · JP 2014‑92777 · JP 2015‑520409 · JP 2019‑21428 · KR 10‑2014‑0031391 · WO 1998/040875 · WO 2013/006385 · WO 2013/163113 · WO 2014/008194 · WO 2014/130463 · WO 2015/025330

A.4 Non-patent literature (family record)

Int'l Search Report & Written Opinion, PCT/US2015/052860 (Dec. 8, 2015); PCT/US2017/045123 (Nov. 8, 2017); PCT/US2015/030569; Korean/Japanese/European Office Actions (2016–2020); Jae-Seung Choi, "Text-dependent Speaker Recognition…," J. Korean Inst. Info. Tech. 10(5) (2012); Auckenthaler et al., "Score Normalization for Text-independent Speaker Verification System," Digital Signal Processing, vol. 10 (2000).

---

B. The references that actually matter for the '884 claims (§ 102 focus)

The '884's claimed subject matter (per the abstract/Summary) is multi-device hotword arbitration: receive audio → compute a first hotword-likelihood value → receive a second such value from another device → compare → initiate (or forgo) speech recognition. Only a handful of the cited references even touch on this. Under § 102, a single reference must disclose all elements; as shown below, the family record does not contain a clean anticipatory reference for the full multi-device-exchange concept, so the strongest § 102 candidate is partial and the rest are § 103-type art.

B.1 US 8,340,975 B1 — Rosenberger (★ strongest § 102 candidate in the record)

• Full citation: U.S. Pat. No. 8,340,975 B1, "Interactive speech recognition device and system for hands-free building control," Theodore A. Rosenberger.
• Dates: priority 2011‑10‑03/04; issued Dec. 25, 2012 (pre-dates the 2014‑10‑09 priority).
• Description: Wall/hands-free speech-recognition control unit for building automation; a device detects a spoken command locally and coordinates with other units/system.
• § 102 mapping: In the Sonos IPR2023‑00119 record (family context — see earlier "Litigation summary" section), Rosenberger was the sole primary anticipation reference and was mapped element-by-element to the family's independent claims, including: "receiving audio data that includes an utterance of a particular, predefined hotword" → the disclosed local hotword/command detection; "transmitting… a message" and "receiving… an additional message" → the network messaging between units; and "determining… to exit the low power mode" → waking the unit in response to the coordinated message. Those concepts correspond to the '884 family's "receiving a second value… from a second computing device," "comparing," and "determining an activation state / initiating speech recognition."
• Caveat: The Federal Circuit (Google LLC v. Sonos, Inc., No. 2024‑2119, June 9, 2026) reversed the PTAB anticipation holdings, finding the "exchanging messages while remaining in low power mode" limitation was not supported by substantial evidence. So Rosenberger is the best § 102 candidate on paper, but its anticipatory force against the family claims has been judicially rejected.

B.2 US 2014/0163978 A1 — Basye et al. (secondary reference)

• Full citation: U.S. Pub. 2014/0163978 A1, "Speech recognition power management" (Google). Pub. June 12, 2014 (pre-dates the priority).
• Description: Manages power states for always-listening speech recognition (e.g., low-power hotword detection vs. higher-power recognition).
• § 102 mapping: Directly addresses the family's "low power mode," "maintaining the first computing device in the low power mode," and activation-state limitations (see the '299 claim 5 / '884 family concepts). Used in the Sonos IPR as the obviousness secondary reference rather than an anticipatory reference. On its own under § 102 it does not disclose the multi-device value exchange.

B.3 US 2014/0278435 A1 — Ganong, III et al. (Nuance)

• Full citation: U.S. Pub. 2014/0278435 A1, "Methods and apparatus for detecting a voice command," pub. Sept. 18, 2014.
• Description: Two-stage voice-command detection (lower-power trigger + higher-power confirmation), i.e., threshold-based hotword detection.
• § 102 mapping: Maps to the '884 dependent concept "determining that the first value satisfies a hotword score threshold" and "without performing automated speech recognition on the audio data." Not anticipatory of the exchange/compare features.

B.4 US 8,938,394 B1 — Faaborg et al.

• Full citation: U.S. Pat. No. 8,938,394 B1, "Audio triggers based on context," issued Jan. 20, 2015 (priority 2014‑01‑08).
• Description: Selects device behavior in response to audio triggers based on context (e.g., which device/state is active).
• § 102 mapping: Maps to the family concepts of "device in use / active display" determining which device responds (the '299 claim 8 / '884 family "second computing device was in use" concept). Context-based selection is suggestive of the arbitration concept but not a full anticipatory disclosure.

B.5 US 8,214,447 B2 — Deslippe et al. (Bose)

• Full citation: U.S. Pat. No. 8,214,447 B2, "Managing an audio network," issued July 3, 2012 (filed 2004‑06‑07).
• Description: Coordinated control/arbitration among multiple networked audio devices.
• § 102 mapping: Maps to the "device group" / identifying other devices configured to respond concept (the '884 family's networked-device discovery and messaging). Lacks any hotword-confidence-score comparison.

B.6 US 8,775,191 B1 — Sharifi et al. (same inventor)

• Full citation: U.S. Pat. No. 8,775,191 B1, "Multi-stage hotword detection," issued July 8, 2014.
• Description: Two-stage hotword detection (triggering then confirming a hotword).
• § 102 mapping: Maps to "determining a first value corresponding to a likelihood that the utterance includes a hotword" and threshold-satisfaction concepts. Same-inventor art; not anticipatory of the multi-device comparison.

B.7 Networked-audio / multi-zone references

• US 7,571,014 B1 (Lambourne et al., "Method and apparatus for controlling multimedia players…," issued Aug. 4, 2009) and US 8,588,949 (Lambourne).
• § 102 mapping: Relevant to multiple co-located devices coordinating which one responds (Sonos-type multi-zone arbitration), but not to hotword confidence scoring. Context only.

B.8 Foreign references touching the concept

• WO 2013/163113 A1 (Nuance, "Embedded system for construction of small footprint speech recognition with user-definable constraints," pub. Oct. 31, 2013) and WO 2014/130463 A2 (Sony, "Hybrid performance scaling for speech recognition," pub. Aug. 28, 2014) — relevant to on-device hotword/recognition power management, i.e., the "low power / initiate recognition" concepts. Not anticipatory of the exchange-and-compare.
• JP 2014‑92777 A (Magic Hand KK, "Activation of mobile communication device via voice," pub. May 19, 2014) and JP H09‑62293 / JP H11‑52976 / JP 2000‑75954 — voice activation / dialogue devices; general hotword-enablement context only.

---

C. Bottom line on § 102

1. No cited reference in the family record discloses all elements of the '884's core claim concept (receiving a second hotword-likelihood value from a second device → comparing → initiating/forgoing speech recognition). The bulk of the cited art is speaker-verification, keyword-spotting, and single-device voice-activation art that is cumulative background.
2. US 8,340,975 (Rosenberger) is the single best § 102 candidate in the record — it was the PTAB's chosen anticipation reference against this family — but the Federal Circuit (No. 2024‑2119, June 9, 2026) rejected that anticipation holding for lack of substantial evidence on the message-exchange element. Any honest § 102 opinion must account for that.
3. US 2014/0163978 (Basye), US 2014/0278435 (Ganong), US 8,938,394 (Faaborg), and US 8,214,447 (Deslippe) are better characterized as § 103 obviousness candidates that individually supply the "low-power mode," "hotword-score threshold," "device-in-use," and "networked device coordination" elements.

---

D. What I could NOT verify (explicit)

• The exact PTO‑892 list printed on U.S. Pat. No. 12,254,884 B2 — I used the family citation record (reported for US 11,557,299 / app. 17/137,157) as a proxy. Do not treat Section A as the '884 face page.
• The verbatim issued claims of 12,254,884 — element mapping in Section B is to the family's representative independent-claim concepts, not to '884 claim language.
• I did not independently re-verify each of the ~120 citations' titles/dates line-by-line; descriptions of the lower-relevance references (Section A.1/A.2 bulk) are given at a summary level, and individual publication dates for a few of the older references should be confirmed against the documents themselves.

Recommended verification steps: pull the '884 front page from USPTO PatentCenter (application 18/421,254) for the authoritative (56) reference list, and pull the issued claims from the same source to finalize the element-by-element § 102 mapping.

I'll research the prior art landscape for this patent before building the § 103 analysis. Note: the task says to "use the results from the Prior Art section of this page," but no dedicated Prior Art section was included in this page's content or in the previously-generated sections — only the Google Patents "prior art keywords" field (computing device, hotword, utterance, voice command, audio data) and the family-litigation references (US 8,340,795). I'll flag this gap and work from what I can verify.

I have strong prior-art leads now. Let me try to retrieve the actual claim text of the '884 patent (or its pre-grant publication) and understand the Basye reference used in the family IPR.

Obviousness Analysis — U.S. Patent No. 12,254,884 B2 Under 35 U.S.C. § 103

Important threshold caveats (read first)

Two gaps materially affect the confidence of any § 103 opinion here, and I will not paper over them:

1. No "Prior Art section" was actually supplied. The task says to "use the results from the Prior Art section of this page," but the page content provided (and the earlier-generated sections) contains no prior-art listing beyond the five "prior art keywords" (computing device, hotword, utterance, voice command, audio data) and the family-litigation note. I therefore base this analysis on (a) the references cited on the face of the family patents (e.g., US 10,593,330 / 10,134,398 "References Cited"), (b) the references actually relied on in the Sonos IPRs against this family, and (c) the keywords above. If you have a specific prior-art list you intended, supply it and I will re-run against it.

2. The issued claims of 12,254,884 were not in the supplied text. However, I located the pre-grant publication of the same application — US 18/421,254 published as US 2024/0169992 A1 — whose claim set is the best available proxy for the issued claims. I treat those as the operative claim set subject to the caveat that the granted claims may have been amended during prosecution (the '884 patent issued 2025-03-18, well after publication). This should be verified against USPTO PatentCenter for application 18/421,254.

The operative claim 1 (from US 2024/0169992 A1, application 18/421,254)

"1. A computer-implemented method executed on data processing hardware of a first computing device that causes the data processing hardware to perform operations comprising:
[1.1] receiving audio data that corresponds to an utterance of a voice command and a hotword preceding the voice command, the utterance of the voice command and the hotword preceding the voice command captured by the first computing device and a second computing device, the first computing device and the second computing device each configured to respond to voice commands that are preceded by the hotword;
[1.2] transmitting, from the first computing device to the second computing device after the second computing device captured the utterance of the hotword preceding the voice command, a message, the message received by the second computing device causing the second computing device to not respond to the voice command despite the second computing device capturing the utterance of the voice command and the hotword preceding the voice command; and
[1.3] based on the message, causing the first computing device to respond to the voice command."

Dependent claims (as published) add: receiving an additional message from the second computing device, and determining a hotword score that reflects a likelihood the audio includes the hotword and determining the hotword score satisfies a threshold.

Source: https://patents.google.com/patent/US20240169992A1/en.

Critical observation for § 103: Unlike the earlier family members US 10,134,398 and US 10,593,330, whose independent claims recite receiving audio "by a computing device that is in a low power mode," the '884/'992 independent claim does not recite a "low power mode." That limitation was the sole basis on which the Federal Circuit reversed the PTAB's invalidity findings in Google LLC v. Sonos, Inc., No. 2024-2119 (Fed. Cir. June 9, 2026) — the court held Rosenberger "fails to disclose the limitations directed to exchanging messages while in a low power mode." Because the '884's apparent claims drop the low-power-mode limitation, the strongest defense that saved the '398/'330 claims on appeal is absent here, and the prior art lands materially closer. This makes the '884 claims more exposed to § 103 than its siblings.

---

1. Person of Ordinary Skill in the Art (POSITA)

A POSITA at the Oct. 9, 2014 priority date would have a bachelor's degree in electrical engineering, computer science, or a comparable field, plus ~2–4 years of experience (or a master's degree plus ~1–2 years) in speech/audio processing and networked consumer electronic devices, including hotword/keyword spotting, distributed device coordination, and short-range wireless communication (Bluetooth, Wi-Fi, UPnP). This is consistent with the level the patent itself assumes (a "hotworder" using MFCCs and SVM/neural classifiers — see the specification) and with the sophistication assumed by Rosenberger and the Sharifi family references.

---

2. The prior-art landscape (grounded, with sources)

Discrepancy flag (do not auto-correct): the family patents' "References Cited" list US 8,340,975 B1 (Rosenberger), whereas the CAFC opinion and the Sonos IPR materials render it US 8,340,795. These are two different numbers. I preserve both as they appear in the sources and note that one is likely an OCR/reporting error; the reference is the same Rosenberger "interactive speech recognition control device" patent. Sources: https://patentimages.storage.googleapis.com/66/b6/bd/6fb1ff059b09fd/US10909987.pdf; https://law.justia.com/cases/federal/appellate-courts/cafc/24-2119/24-2119-2026-06-09.html; https://www.docketalarm.com/cases/PTAB/IPR2023-00119/Sonos_Inc/docs/10-27-2022-Petitioner/Exhibit-1001-US10593330.pdf.

---

3. Obviousness grounds

Ground 1 — Rosenberger alone (single-reference § 103, or § 102-style)

Rosenberger expressly frames and solves the same problem the '884 patent frames ("selecting the correct device for reacting to a hotword, and suppressing reaction to the hotword on other devices"). Element mapping:

• [1.1] Audio of a hotword + command captured by two devices: Rosenberger's system is a network of "functionally identical" control devices, each with a microphone, "constantly listening for… 'trigger phrases,'" that recognize a trigger phrase followed by a command. Multiple devices in the space recognize the same speech command (the "duplicate response problem").
• [1.2] First device transmits a message causing the second device not to respond: Rosenberger's devices transmit a weighted signal (WS) and/or a "coordination signal derived from said weighted signals"; coordination ensures "only one device will respond," i.e., the non-winning devices are made not to respond. That is precisely a message-causing-suppression architecture.
• [1.3] The first device responds based on the message: "[O]nly the single device with the highest weighted signal proceeds to interact with the user and/or process the speech command."

The dependent-claim elements (hotword score + threshold; additional message from the second device) map onto Rosenberger's weighted signal and its exchange of weighted signals/coordination signals between devices.

Why this ground is stronger here than in the Sonos IPRs: the PTAB's Rosenberger anticipation finding for the '398/'330 claims was reversed only because those claims required the exchange to occur "while… in the low power mode." The CAFC found Rosenberger devices "exit the low power 'listening' mode to calculate and exchange weighted signals." The '884's apparent claims contain no such low-power-mode requirement, so that dispositive gap does not exist.

Weakness to note candidly: If the granted '884 claims do in fact include a "low power mode" or "sleep mode" limitation, then Ground 1 collapses to the same fate as the Sonos IPRs — apply the CAFC's reasoning directly.

---

Ground 2 — Rosenberger in view of Sharifi '191 (US 8,775,191 B1)

Combination: Rosenberger's multi-device coordination + '191's hotword invocation model.

Motivation to combine (KSR factors):

1. Same field / same problem: both address voice-activated systems; '191 supplies the hotword-trigger paradigm ([HOTWORD][QUERY]) that Rosenberger's "trigger phrase" presumes.
2. Predictable, known result: substituting keyword/hotword spotting for Rosenberger's generic trigger-phrase recognition is the application of a known technique to a known structure, yielding no more than the expected result (a device that wakes on "OK computer").
3. Articulated design incentive: cold-start/always-listening assistants (Google's "OK computer") required local hotword detection before cloud ASR; a POSITA would look to '191 for that.

This ground supplies the [1.1] "hotword preceding the voice command" and the dependent "hotword score/threshold" elements cleanly.

---

Ground 3 — Rosenberger in view of Basye (US 2014/0163978 A1)

Combination: Rosenberger's WS-based arbitration + Basye's device-selection/arbitration teachings (Basye was the exact secondary reference Sonos used in Ground II — obviousness — against this family's dependent claims).

Motivation: Both are directed to deciding which of several devices should handle a user interaction; combining them to select the responding device and suppress the others is a predictable use of known references addressing the same "which device responds" problem. Basye's incremental value is at the margins (e.g., the "additional message"/bilateral-exchange and "device in use" concepts), so this ground is best deployed for the dependent claims rather than independent claim 1 where Rosenberger alone is strong.

---

Ground 4 — Rosenberger + Sharifi '191 + Basye

The full three-way combination addresses every element with an express disclosure: Rosenberger (multi-device coordination + suppression + WS exchange), '191 (hotword preceding command, hotword scoring/threshold), Basye (device arbitration / messaging). KSR's "combination of familiar elements according to known methods" rationale applies with particular force: each reference is in the same field, each solves a sub-problem of the same overall need, and the combination yields only the expected result — one device responds to "OK computer," the others do not.

---

Ground 5 — Kemp US 8,200,488 ("hotword suppression") as corroborating art

Kemp is independent evidence that suppressing responses to a hotword across multiple microphones/devices was a known and desirable objective by the priority date, strengthening the "motivation" prong for all grounds above.

---

4. Element-by-element summary table

---

5. Anticipated defenses / counter-arguments a POSITA/examiner should weigh

1. The "low power mode" limitation is the entire battleground — and it may be absent.

   • If the granted '884 claims recite no low-power/sleep-mode limitation, Google's Sonos victory is inapposite and the Rosenberger-based § 103 grounds are strong.
   • If the granted claims do recite it, the CAFC's holding controls: Rosenberger discloses devices exchanging weighted signals after exiting the low-power listening mode, so a bare Rosenberger ground fails; the petitioner would need a secondary reference (e.g., Basye) teaching message exchange while remaining in the low-power state.

2. The "message causing the second device to not respond" recitation is narrower than Rosenberger's "highest WS proceeds" mechanism. A patent owner will argue Rosenberger's coordination is implicit (each device self-selects by comparing WS) rather than an explicit suppression message causing non-response. Rebuttal: Rosenberger's claim 1 expressly recites a receiver for a "coordination signal derived from said weighted signals," and the specification states coordination ensures "only one device will respond" — the functional result is a message that suppresses the others. This is a genuine claim-construction dispute (does "causing" require a dedicated suppression command, or does broadcast-and-self-defer satisfy it?).

3. Teaching away / "frustration of purpose." No strong teaching-away appears: Rosenberger wants a single responder, aligning with the '884 objective.

4. Objective indicia (secondary considerations). Google would likely point to commercial success of multi-device hotword suppression and licensing; however, the nexus to the specific claimed combination (rather than to always-on assistants generally) would be difficult to establish, and such evidence usually does not overcome a strong KSR combination.

5. § 101 overlay (not asked, but worth flagging): these claims are tied to concrete device coordination and would not face the Alice exposure of pure speech-recognition abstractions — relevant only if the matter migrates to eligibility.

---

6. Bottom line

• Most likely to succeed: a § 103 ground built on Rosenberger (US 8,340,975 / 8,340,795) as the primary reference, optionally in view of Sharifi '191 (US 8,775,191) for the hotword-scoring elements and Basye (US 2014/0163978) for the messaging/arbitration dependent claims.
• The decisive variable is whether the granted '884 claims carry the "low power mode" limitation. The published claims of the same application (US 2024/0169992 A1) do not. If the granted claims track the publication, the Federal Circuit's Google v. Sonos reversal (which turned entirely on the low-power-mode limitation) does not shield the '884 patent, and the Rosenberger-based obviousness attack is materially stronger than the attack that failed against the '398/'330 patents.
• Confidence: Bibliographic and family data — high. The prior-art disclosures of Rosenberger and the family references — moderate-to-high (grounded in the CAFC opinion and the patents' own "References Cited"). The exact issued claim text of 12,254,884 — low-to-moderate; I used the same application's pre-grant publication as a proxy. Verify against USPTO PatentCenter for 18/421,254 before relying on this opinion, and reconcile the 8,340,975 vs. 8,340,795 numbering discrepancy.

Sources: https://patents.google.com/patent/US20240169992A1/en (same-application claims); https://patents.google.com/patent/US11557299 (sibling claims); https://patentimages.storage.googleapis.com/66/b6/bd/6fb1ff059b09fd/US10909987.pdf and .../US11276406.pdf (family "References Cited"); https://www.docketalarm.com/cases/PTAB/IPR2023-00119/Sonos_Inc/docs/10-27-2022-Petitioner/Exhibit-1001-US10593330.pdf; https://www.docketalarm.com/cases/PTAB/IPR2023-00119/Sonos_Inc._v._Google_LLC/docs/02-29-2024-Petitioner/Exhibit-1034-Petitioners_Demonstratives.pdf (Ground I Rosenberger; Ground II Rosenberger + Basye); https://law.justia.com/cases/federal/appellate-courts/cafc/24-2119/24-2119-2026-06-09.html and https://ipwatchdog.com/2026/06/09/federal-circuit-reverses-ptab-findings-of-unpatentability-for-googles-hotword-patents/ (CAFC reversal on the low-power-mode limitation).