US 7558729

I'll search for information on this specific patent number.

The CAFC docket search returned nothing. Let me probe further for litigation and docket activity.

US Patent 7,558,729 — Analyst Summary

Search scope note: I searched for this exact number (US7558729 / US 7,558,729 B1). I found the granted patent, its family, and citation records. I searched separately for CAFC 2026 docket activity on this patent and found none — no appeal, IPR, or Federal Circuit case tied to 7,558,729 surfaced in my searches. I cannot confirm whether litigation exists in district court under a different caption; the assignee, Nytell Software LLC, is an active litigating entity, but I found no docket naming this patent. Treat the "no docket activity found" conclusion as an absence of evidence, not evidence of absence.

Bibliographic Data

Citations of record (6): US 5,274,705 (Tellabs); US 6,424,635 (Nortel); US 6,633,841 (Mindspeed — "Voice activity detection speech coding to accommodate music signals"); US 6,760,435 (Lucent); US 2007/0136053 A1 (Acoustic Technologies, "Music detector for echo cancellation and noise reduction"); US 7,430,506 (RealNetworks Asia Pacific). Non-patent literature: Tanrikulu et al., "A new non-linear processor (NLP) for background continuity in echo control," ICASSP 2003; Zhu et al., "Music Key Detection for Musical Audio," MMM 2005.

Abstract (verbatim)

A method of using music detection to enhance an operation of an echo canceller is provided, wherein the echo canceller includes an adaptive filter and a nonlinear processor. The method comprises receiving an input signal including an echo signal by the echo canceller from a near end device, filtering the input signal using the adaptive filter to eliminate linear components of the echo signal in the input signal and generate an error signal, analyzing the error signal using a music detector to determine existence of a music signal in the error signal, bypassing the nonlinear processor if the analyzing determines the music signal exists in the error signal, and eliminating nonlinear components of the echo signal from the error signal using the nonlinear processor if the analyzing determines the music signal does not exist in the error signal.

Core Technical Problem

Conventional voice activity detectors (VADs) distinguish speech from background noise using signal stability — speech spectra change rapidly, noise spectra are stable. Music is also stable over many frames, so conventional VADs misclassify music as background noise. The consequence in an echo canceller: the non-linear processor (NLP) attenuates the "noise" below the noise floor and replaces it with comfort noise or silence, destroying music. In a speech encoder, music gets coded at low rate as noise, yielding poor perceptual quality.

The invention's fix: detect music explicitly and bypass the destructive processing stages (NLP, noise suppressor, pitch interpolation, pitch pre-processing), and select a high bit rate when music is present.

Plain-Language Overview of the Independent Claims

There are nine independent claims: 1, 4, 5, 6, 9, 10, 11, 14, 15. They fall into three families — method claims, system claims, and computer-readable-medium claims — each with three variants keyed to different encoder subsystems (noise suppressor / pitch interpolation / pitch pre-processing).

Method Claims

Claim 1 — Echo canceller + speech encoder with a noise suppressor (the anchor claim)
A processor-executed method with these steps:

1. The echo canceller receives an input signal containing an echo, from a near-end device.
2. The adaptive filter removes the linear part of the echo and outputs an error signal.
3. A music detector analyzes that error signal to see whether music is present.
4. If music is present → bypass the NLP (so music is not erased).
5. If music is absent → the NLP eliminates the non-linear residual echo as normal.
6. If music is present → the noise suppressor's attenuation gain is gradually ramped to zero (0 dB, i.e., no attenuation) for a smooth transition.
7. If music is absent → the noise suppressor attenuates the error signal normally.

Key structural point: this claim ties the echo-canceller NLP bypass and the encoder's noise-suppressor gain ramp to one shared music detection result. It also requires the gain to be gradually reduced — a hard bypass alone is not enough for claim 1 (that is the subject of dependent claim 2).

Claim 4 — Echo canceller + speech encoder with pitch interpolation
Same first five steps (receive → adaptive-filter → detect music → bypass NLP / apply NLP). Then, when music is present:

• disable the encoder's pitch interpolation;
• transmit information to the decoder to disable the decoder-side pitch interpolation as well (this is an encoder/decoder signaling element, not just a local bypass);
• when music is absent, enable pitch interpolation.

Claim 5 — Echo canceller + speech encoder with pitch pre-processing
Same first five steps. Then:

• when music is present, disable pitch pre-processing (so speech parameters are not modified);
• when music is absent, enable pitch pre-processing.

System Claims

Claim 6 — Enhanced speech processing system (noise-suppressor variant)
A processor-based system combining an echo canceller and a speech encoder:

• Echo canceller comprises: a receiver for the near-end input signal with echo; an adaptive filter producing an error signal; a music detector; and a non-linear processor that is bypassed when music is detected and otherwise removes residual non-linear echo.
• Encoder comprises a noise suppressor that gradually reduces its attenuation gain to zero when music is detected, and otherwise attenuates the error signal.

Claim 9 — Enhanced speech processing system (pitch-interpolation variant)
Same echo-canceller architecture. Encoder side: disable pitch interpolation when music is detected; transmit information to the decoder to disable decoder-side pitch interpolation; enable pitch interpolation when music is absent.

Claim 10 — Enhanced speech processing system (pitch pre-processor variant)
Same echo-canceller architecture. Encoder side: disable the pitch pre-processor when music is detected; enable it when music is absent. (Note: the claim body uses the phrasing "determine existing of a music signal" — an apparent typographical variant in the granted text; I am reporting it literally rather than correcting it.)

Computer-Readable-Medium Claims

Claim 11 — a CRM/software product with code performing the claim-1 method (echo canceller + noise-suppressor gain ramp to zero).
Claim 14 — a CRM/software product with code performing the claim-4 method (pitch interpolation disable + decoder signaling).
Claim 15 — a CRM/software product with code performing the claim-5 method (pitch pre-processing disable/enable).

Dependent Claims

• Claim 2 (dep. on 1): also bypass the noise suppressor outright when music is present — the hard-bypass alternative to the gradual gain ramp of claim 1.
• Claim 3 (dep. on 1): defines how the music detector decides existence, using a three-threshold scheme.
• Claim 7 (dep. on 6): system-claim analogue of claim 2 (bypass the noise suppressor on music).
• Claim 8 (dep. on 6): system-claim analogue of claim 3 — modular implementation of the threshold logic.
• Claim 12 (dep. on 11): CRM analogue of claims 2/7.
• Claim 13 (dep. on 11): CRM analogue of claims 3/8.

The threshold logic (claims 3, 8, 13), in plain terms: For a parameter P1 extracted from a frame of the error signal (e.g., pitch correlation R_p or LPC gain), three thresholds are defined — a music threshold, a background-noise threshold, and an "unsure" threshold in between.

• If P1 falls outside the [noise, music] band → classify immediately as music or noise, whichever threshold it is nearer to.
• If P1 falls inside the band (the ambiguous zone) → defer and classify by analyzing multiple P1 values across a plurality of frames (per the specification: accumulate cnt_mus / cnt_nomus counters over roughly 100–200 frames and compare them).

Enabling Disclosure Highlights (for context, not claim scope)

• FIG. 2 shows the echo canceller 200 architecture: double-talk detector 210, high-pass filter 215, adaptive filter 220, error estimator 218, NLP 230, and crucially music detector 235. Music detection may alternatively be performed outside the echo canceller and its result fed in.
• FIG. 4/5 show the pitch-correlation distribution for music vs. background noise and the T0/T1/T2 thresholding flowchart.
• The specification notes the VAD may output more than two values — e.g., 0 = background noise, 1 = voice, 2 = music — and that after music is declared, the detector holds the music indication for an extension period (~30 frames) after music actually stops, to avoid coding glitches.
• Multi-rate coders may select the maximum available bit rate on music detection (described in the specification; note rate selection 345 is discussed but is notably not recited as a required element in any independent claim).
• Background: the spec notes that voice-suppressor/NLP behavior on music would otherwise replace it with comfort noise or silence; and it incorporates by reference US 6,804,203 (double-talk detector) and US 6,507,814 (pitch determination using speech classification).

Uncertainty Flags

1. Adjusted expiration conflict: 2027-06-18 (Google Patents) vs. 2027-10-29 (patents-review.com). Unresolved.
2. Assignee chain: Google Patents lists Nytell Software LLC as current assignee, reached via assignment to O'Hearn Audio LLC (effective 2012-10-30) and a subsequent merger (effective 2015-08-26). Some citation records (e.g., other Google Patents citation tables) still show "Mindspeed Technologies, Inc." as the listed assignee for this number. Nytell is the more current record.
3. No CAFC 2026 docket found. I could not locate any Federal Circuit appeal, PTAB proceeding, or district court complaint specifically naming 7,558,729. Do not infer either the existence or the nonexistence of litigation from this — my search coverage of docket databases via web search is incomplete.
4. MPEP/terminal-disclaimer and fee-status details (e.g., the 2021 late-payment surcharge and 12th-year fee) come from Google Patents' legal-events table and were not verified against USPTO Patent Center directly.

I'll search litigation databases for this specific patent number, as instructed.

First-pass searches returned noise. Let me query litigation-specific databases directly.

The Nytell searches are matching an unrelated 1989 "NYTel" bankruptcy case. Let me target patent-litigation databases specifically.

Litigation Search — US Patent 7,558,729

Bottom line

I found no litigation naming US Patent 7,558,729. Despite targeted searching of litigation-oriented sources (Unified Patents litigation content, PACER/CAFC-adjacent material surfaced via web search, and dockets naming the current assignee Nytell Software LLC), no complaint, ITC investigation, PTAB proceeding, or Federal Circuit appeal specifically asserting or challenging 7,558,729 surfaced.

Per your instruction, I did not report results for similar numbers. The searches that returned hits involved different patents and different parties, which I flag below as non-matches rather than results.

What I searched, and what came back

I hit a tool-call/step limit before completing the last three queries, so this is a partial search, not an exhaustive one.

Critical caveats (read before relying on this)

1. Absence of evidence ≠ evidence of absence. My coverage of PACER, Unified Patents' full caselist, and Docket Navigator is incomplete via general web search. A negative finding here should not be represented as confirmation that no suit exists.
2. The current assignee is an active litigating entity. Nytell Software LLC (per Google Patents, reached via assignment to O'Hearn Audio LLC, then merger) is the type of entity that asserts patents. That raises, not lowers, the probability that campaign litigation exists somewhere — my search simply did not surface it for this number.
3. This patent is the CIP parent/child of a family that may be litigated together. US 7,558,729 is a continuation-in-part of Ser. No. 10/981,022 (US 7,120,576) and the parent of Ser. No. 11/156,874 (US 7,130,795). A campaign asserting the family could plausibly name the sibling patents rather than, or in addition to, 7,558,729. I did not surface a docket for any of the three.
4. Name-matching hazard. Because "Nytell" fuzzily matches "NYTel," generic searches produce false positives. Any future search should use "Nytell Software LLC" in quotes and combine with the patent number.

Recommended next steps to resolve definitively

• Unified Patents litigation portal (https://portal.unifiedpatents.com/litigation/caselist) — query by patent number 7558729, not by party name.
• PACER — search the full-text "Document" field for 7,558,729, and separately run a party search for Nytell Software LLC to enumerate all of its cases, then check each complaint's asserted-patent list.
• PTAB/PTA CTS — search IPR/PGR petitions by patent number 7,558,729.
• CAFC — search the docket for appeals originating from any Nytell district court case, since the '729 patent could be at issue on appeal even if not named in the original caption.

If you can confirm whether Nytell has a known assertion campaign and the courts involved, I can narrow the search to those dockets.

Status of this section: No litigation identified for US 7,558,729 as of the search date. This finding is provisional and incomplete, and I am explicitly flagging that I could not complete all planned queries before the tool limit was reached.

I'll verify the canonical "no proceedings" signal against external sources before finalizing.

PTAB / AIA Trial Landscape — US 7,558,729

Search and verification note

The structured "PTAB proceedings on file" block supplied in this prompt is authoritative and states: the USPTO ODP API returns no AIA trial proceedings for this patent as of the most recent ingest.

I ran independent web searches to try to falsify that signal and found nothing to contradict it:

Bottom line: I found no IPR, PGR, or CBM proceeding naming US 7,558,729, and no Federal Circuit appeal of any PTAB decision on it. I am reporting a genuine absence of proceedings, not a failure to look. See the caveats in the Strategic Summary — docket coverage via web search is imperfect, and this is an absence of evidence.

Proceedings overview

Total AIA trial proceedings on US 7,558,729: zero. Breakdown by status: 0 active, 0 with claims invalidated, 0 with claims sustained, 0 settled, 0 institution-denied.

Defensive posture this gives a defendant: tabula rasa. None of claims 1–15 has ever been tested at the PTAB, so nothing is canceled, nothing is statutorily narrowed, and no petitioner estoppel has attached to anyone. Conversely, a defendant here cannot point to an earlier tribunal's work product, claim-construction findings, or an invalidity record to leverage. The patent arrives un-scarred — but also un-vetted, which cuts both ways: there is no adverse PTAB precedent hardening the owner's position, and no prior petitioner to free-ride on.

Proceedings on file

None. There are no proceeding numbers to report, and I will not invent any. For completeness, the per-proceeding template collapses as follows:

• Type / Filed / Status: N/A — no petition has been filed.
• Judge panel: N/A — no panel has been assigned.
• Petition grounds: N/A — no grounds of record.
• Institution decision: N/A — no § 314(a) or § 324(a) determination exists.
• Final Written Decision: N/A — no FWD exists. No claim of this patent has ever been canceled or confirmed by the Board.
• Settlement / termination: N/A.
• Appeal: N/A — there is no Board decision to appeal under § 319 / § 329.
• Defensive value: The only defensive value on the PTAB side is the clean slate itself (see below).

Strategic summary

Claim status: claims 1–15 are all UNTESTED. Independent claims 1, 4, 5 (methods), 6, 9, 10 (systems), and 11, 14, 15 (computer-readable media) — together with dependents 2, 3, 7, 8, 12, 13 — have never been construed by the Board, never been the subject of an instituted trial, and never been canceled. The patent is active per the legal-events table, with maintenance fees paid through the 12th year (including a 2021-02-11 late-payment surcharge), and an adjusted expiration reported as 2027-06-18. That is a meaningful strategic fact: less than a year of term remains as of the current date (2026-09-29), which materially suppresses the expected value of any IPR campaign. Spending $300k–$500k+ on an IPR to kill a patent with roughly nine months of enforceable life is usually irrational — which may itself explain the empty docket. (Calendar note: the previously generated summary flags an unresolved conflict between the 2027-06-18 date on Google Patents and a 2027-10-29 date reported by a third-party database; even the later of the two leaves under 26 months of term.)

Estoppel landscape: no estoppel exists. Because no IPR/PGR/CBM was ever instituted, § 315(e)(2) estoppel has never attached to any party. There is no petitioner, no real party in interest, and no privy who is now barred from raising § 102/§ 103 grounds they raised or reasonably could have raised. A defendant today faces a completely open prior-art field: any reference — including the six citations of record (US 5,274,705 Tellabs; US 6,424,635 Nortel; US 6,633,841 Mindspeed; US 6,760,435 Lucent; US 2007/0136053 Acoustic Technologies; US 7,430,506 RealNetworks) and the two non-patent citations (Tanrikulu, ICASSP 2003; Zhu, MMM 2005) — plus anything else, is fair game. Note the citation table itself is a roadmap: US 6,633,841 is a Mindspeed-owned reference on "Voice activity detection speech coding to accommodate music signals," and US 2007/0136053 (Acoustic Technologies) is titled "Music detector for echo cancellation and noise reduction" — i.e., the examiner considered art squarely on the music-detection-for-NLP-control concept.

Pattern signals — and the honest limits of what I can infer.

• Same petitioner, multiple IPRs: not applicable — there is no petitioner at all.
• Patent owner as aggressive PTAB appellant: inapplicable — Nytell has never had a Board decision on this patent to appeal.
• Defensive aggregator in the chain: I found no evidence that Unified Patents or any similar aggregator has touched this patent. The only Unified Patents hit in my searches was an unrelated 2017 conference write-up. Do not read a Unified connection into this record.
• Why might a litigating NPE's patent have zero IPR history? I can only offer hypotheses, explicitly flagged as such, not findings: (i) the patent has never been asserted, or has been asserted only in campaigns that settled or were dismissed before any defendant's § 315(b) one-year clock motivated a petition; (ii) the near-term 2027 expiration makes an IPR uneconomic; (iii) the narrowness of the anchor claim — claim 1 requires not merely bypassing the NLP but also "gradually reducing an attenuation gain of the noise suppressor to zero," a specific and awkward-to-prove encoder behavior — makes the claim harder to assert broadly, which in turn lowers the incentive to invalidate it. I could not confirm or refute any of these; treat them as questions for follow-up, not conclusions.

Procedural posture bullets: (a) The patent is pre-AIA (filed 2005-03-17), so PGR is unavailable — PGR reaches only patents with an effective filing date on or after 2013-03-16. (b) CBM review is unavailable both substantively and temporally — the patent's claims are directed to speech/audio signal processing, not a "financial product or service," and the CBM program sunset on 2020-09-20 under SAS/statutory expiry. IPR is the sole viable AIA vehicle. (c) A defendant served with a complaint alleging infringement has a one-year § 315(b) window to petition; the § 315(a)(1) bar also means a defendant who first files a declaratory-judgment action challenging validity forfeits the IPR.

Recommended next steps

1. Treat the PTAB as an empty, fully available battlefield — but run the economics first. Because no claim has been canceled, you cannot say "claim 1 is dead." You also have no estoppel and no adverse constructions. Before drafting, compute the remaining term (the 2027-06-18 vs. 2027-10-29 conflict should be resolved directly against USPTO Patent Center / PAIR for the application, 11/084,392) and decide whether an IPR is worth the spend. For a patent with ~9–26 months left, a stay-driven cost-delay strategy or a district-court invalidity case may dominate an IPR on pure ROI.

2. If you do petition, the art is wide open — start with the examiner's own citations. Build § 103 combinations from US 6,633,841 (Mindspeed, VAD for music signals), US 2007/0136053 (Acoustic Technologies, music detector for echo cancellation and noise reduction), US 6,760,435 (Lucent, network speech enhancement), and Tanrikulu et al., ICASSP 2003 (non-linear processor design). The examiner's having cited them does not immunize them from a better-framed § 103 theory.

3. Attack the claim-1 / claim-6 / claim-11 "gradual gain ramp" limitation specifically. It is the narrowest recited element on the noise-suppressor side and requires the gain be gradually reduced to zero, not merely bypassed. If the owner's infringement theory treats any look-ahead muting as a "gradual reduction," a § 112(b) indefiniteness theory on "gradually" (no recited rate, time constant, or frame count in the claims) and/or a § 112(a) written-description theory may be worth scoping in parallel.

4. No FWD or CAFC opinion exists to link to. Because there is no Board decision and no Federal Circuit appeal, there is nothing to cite on USPTO PTAB E2E or CourtListener. The only linkable source of record here is the Google Patents page for US 7,558,729, whose legal-events table I relied on for the fee and assignment history.

5. Verify the negative directly before relying on it. My web-search coverage of PTAB and district-court dockets is incomplete — searches are keyword-driven and will miss a petition captioned under a party name I did not guess. Pull the PTAB E2E "Proceedings" tab and the litigation tab on Docket Navigator / Lex Machina (or PACER) for the patent number itself, and check the sibling patents in its family — US 7,120,576 (Ser. No. 10/981,022) and US 7,130,795 (Ser. No. 11/156,874) — for any proceeding that might have named the family without naming this number. If any such proceeding surfaces, the "clean slate" conclusion above must be revisited.

Caveat on cross-references: the previously generated section's uncertainty flags carry forward — notably the unresolved adjusted-expiration conflict and the assignee-chain question (original assignee Mindspeed Technologies, Inc.; current assignee reported as Nytell Software LLC, reached via assignment to O'Hearn Audio LLC effective 2012-10-30 and a merger effective 2015-08-26). I found no contradiction between that section and anything in this one.

I'll reconstruct the assignment chain for US 7,558,729 by searching USPTO Assignment Center records and cross-referencing the parties.

Excellent — I found a related Mindspeed patent transfer pattern. Let me dig into the specific assignees and correspondents.

Search-scope caveat up front: I hit the tool-step limit partway through this task. I retrieved the assignment chain and party identities (from the patent's own recorded legal events plus third-party records), but I was not able to pull the recorded "correspondent of record" field for the reel/frame entries below, nor complete the RPX/Unified asserter-directory cross-checks. I flag every such gap inline rather than guess.

Inventors

Pattern check — no "departing-inventor" signal. All four inventors are Orange County–based and executed the assignment to Mindspeed on 2005-03-15 (two days before the 2005-03-17 filing). The same three of the four (Shlomot, Gao, Benyassine) reappear as named inventors on later Mindspeed speech-coding filings — e.g., US 8,032,359, "Embedded silence and background noise compression," filed 2007-12-14 — and on the sibling applications in this very family (US 7,120,576; US 7,130,795). That continuity indicates the inventor team stayed with the original assignee rather than leaving within 12 months. The usual fire-sale precursor (inventor exodus right after filing) is absent.

Original assignee

Mindspeed Technologies, Inc. (Newport Beach, CA), a Delaware corporation and former Conexant spin-off. It was a fabless semiconductor company — network/communications processors and, directly relevant here, voice-processing DSP and speech-codec silicon (multiservice access, VoIP gateway, and voice-over-packet chips). The claims at issue (embedded echo-canceller + speech-encoder music detection) map onto that voice-processing product line, so this was an operating company with shipping products, not a filing-only entity.

Current status: No longer independent. MACOM Technology Solutions Holdings acquired Mindspeed in a tender-offer merger that closed in 2014 (Merger Agreement disclosed in MACOM's 14D-9; MACOM's 10-Q confirms "the operations of Mindspeed have been included in MACOM's consolidated" results). Mindspeed's wireless business was separately divested in connection with that deal. Mindspeed did not file bankruptcy — so any patent transfers out of Mindspeed were portfolio monetization ahead of the acquisition, not a Chapter 7/11 sale.

Assignment timeline

The chain has three recorded links, with reel/frame identifiers taken from the patent's own recorded legal-events table (reproduced verbatim on Google Patents).

• 2005-03-15 (executed) / recorded 2005-03-17 — Reel 016414/0034

  • Conveyance: Assignment of assignors' interest
  • Assignor: Adil Benyassine; Yang Gao; Carlo Murgia; Eyal Shlomot
  • Assignee: Mindspeed Technologies, Inc. (California)
  • Correspondent: Not retrieved — I could not surface the filing attorney/firm for this record within the search budget.
  • Context: Initial employer assignment at filing (inventors → company).

• 2012-10-30 (executed) / recorded 2012-11-23 — Reel 029343/0322

  • Conveyance: Assignment of assignors' interest
  • Assignor: Mindspeed Technologies, Inc.
  • Assignee: O'Hearn Audio LLC (Delaware)
  • Correspondent: Not retrieved. This is the single most important gap in this report — the correspondent on the first link out of the operating company is the key repeat-player tell, and I could not pull it.
  • Context: Transfer of an operating company's speech-coding portfolio to a licensing-entity name. Corroborating context: O'Hearn Audio LLC appears as the recorded assignee across a large family of Mindspeed speech-coding patents (e.g., US 8,620,649 and much of the "Adaptive codebook gain control"/CELP-speech patent set then later reassigned to Nytell) — consistent with a bulk portfolio sale, not an isolated conveyance.

• 2015-08-26 (executed) / recorded 2015-11-24 — Reel 037136/0356

  • Conveyance: Merger (see document for details)
  • Assignor: O'Hearn Audio LLC
  • Assignee: Nytell Software LLC (Delaware)
  • Correspondent: Not retrieved.
  • Context: Internal restructuring by merger — the licensing LLC (O'Hearn Audio) was absorbed into another LLC (Nytell Software). Not an arm's-length sale to a new buyer; the same economic owner persists across the merger.

Related-but-separate observation (do not conflate with this chain): A different Mindspeed patent chain (e.g., US 6,633,945) runs Mindspeed → Takano Holdings LLC (2008) → Intellectual Ventures II LLC (merger, executed 2014-09-05, recorded 2014-09-12, Reel 033727/0751), with Correspondent: Schwabe, Williamson & Wyatt, P.C., 1420 5th Avenue, Suite 3400, Seattle, WA 98101. That is evidence that Mindspeed sold multiple speech/IP portfolios to different monetization vehicles around the same era. It does not establish the correspondent for the O'Hearn/Nytell chain, and I am not imputing it.

Timeline diagram

timeline
    title Ownership of US 7558729
    2005 : Filed by four inventors
         : Assigned to Mindspeed Technologies
    2009 : Patent issued as US 7558729
    2012 : Sold to OHearn Audio LLC
    2014 : MACOM acquires Mindspeed
    2015 : Merged into Nytell Software LLC

(Note: the entity is literally spelled "O'Hearn"; the apostrophe is elided above only to keep the diagram parser-safe, per the formatting constraint.)

NPE / troll-pattern signals

1. Shell-entity transfer — PRESENT (documented in part).
The patent moves from an operating chip company (Mindspeed) to a Delaware LLC with a licensing-flavored name (O'Hearn Audio LLC) via Reel 029343/0322 (executed 2012-10-30). The transfer covers a bulk speech-coding portfolio, reinforcing that this is a portfolio-level monetization vehicle rather than a product acquirer. Caveat: the strongest tells called for (no products in commerce, registered-agent service address, single-member LLC) could not be verified for O'Hearn Audio within the search budget — so this is a documented-transfer finding, not a fully evidenced shell finding.

2. Known asserter in the chain — NOT PRESENT (against the enumerated list).
Neither O'Hearn Audio LLC nor Nytell Software LLC appears on the enumerated NPE list (Acacia, Marathon, IV, IPNav, Wi-LAN, Mosaid/Conversant, Vringo, Pendrell, Innovatio, MPHJ, Lumen View, Round Rock, DGC, Spangenberg). Nytell is nonetheless a broad, high-volume non-practicing holder (third-party trackers tally it at ~34–487 patents across video, LCD, memory, networking, and speech coding; grants clustering 2015–2025; last activity 2024). I did not find a Nytell/O'Hearn entry on Unified Patents' or RPX's published high-frequency-plaintiff lists in the searches that completed. Call this not present against the named list, with the broader asserter-directory check incomplete.

3. Repeat correspondent across the chain — UNCLEAR (retrieval gap).
I could not retrieve the correspondent field for any of the three reel/frame entries. This is the signal most likely to convert an "unclear" to a "present," and it remains unresolved. One adjacent data point exists (Schwabe, Williamson & Wyatt as correspondent on the parallel Mindspeed→Takano→IV chain), but a single appearance is not a recurrence, and it is on a different chain, so it does not score here.

4. Cascading transfers (<24 months through chained LLCs) — NOT PRESENT.
The links span 2005 → 2012 → 2015 (Reels 016414/0034, 029343/0322, 037136/0356). The two post-issuance links are ~3 years apart, and the second link (2015) is a merger, not a fresh third-party sale. This is a slow two-step chain, not a rapid cascade.

5. Pre-litigation transfer — UNCLEAR / cannot establish.
No complaint, ITC action, PTAB proceeding, or CAFC appeal naming 7,558,729 was found (consistent with the earlier litigation section). With no identified suit, the six-months-before-filing test cannot be triggered either way. (The 2012 transfer is ~3 years before the 2015 merger, and no assertion date is known.)

6. Bankruptcy fire-sale — NOT PRESENT.
Mindspeed did not enter Chapter 7/11. Its value was realized through a sale to MACOM (2013–2014) and a pre-sale patent divestiture. The 2012 O'Hearn transfer is a monetization event, not a court-supervised estate sale.

7. Privateering — UNCLEAR.
There is a plausible shape (operating company sheds speech IP to an LLC), but I found no SEC filing, EFF/Patent Progress coverage, or Unified/RPX record showing O'Hearn Audio or Nytell asserting on Mindspeed/MACOM's behalf against competitors. Without that, privateering is not a finding.

8. Defensive aggregator (anti-NPE) — NOT PRESENT.
The chain terminates at Nytell Software LLC (Bellevue, WA business address of record: 3150 139th Avenue SE, Building 4, Bellevue, WA 98005). Nytell is not RPX, AST, LOT, Unified, or OIN. The patent has not been neutralized by a defensive aggregator.

Verdict

NPE — moderate confidence.

Justification: the chain contains one strong, well-documented signal — an operating semiconductor company (Mindspeed) transferring a bulk speech-coding portfolio to a Delaware licensing LLC (O'Hearn Audio) at Reel 029343/0322 (executed 2012-10-30), which was then merged into another non-practicing holder (Nytell Software LLC) at Reel 037136/0356 (recorded 2015-11-24). That is the operating-company→licensing-vehicle pattern. Confidence is capped at moderate, not high, because the two corroborating tells that would confirm it — the correspondent of record (unretrieved) and any assertion campaign for this number (none found) — are both missing; and because O'Hearn/Nytell do not appear on the published asserter lists I could reach.

Verification link: USPTO Assignment Center — https://assignmentcenter.uspto.gov/ (and the legacy index at https://assignment.uspto.gov/patent/index.html) — search by Application 11/084,392 or Patent 7,558,729 and open Reels 016414/0034, 029343/0322, and 037136/0356 to read the correspondent field I could not retrieve.

Highest-value follow-ups to close the gaps: (a) pull the correspondent block from each of the three reel/frame records — this directly tests signal #3; (b) run Unified Patents' caselist and RPX's portal by patent number for 7,558,729 and its siblings (US 7,120,576 / US 7,130,795); (c) verify whether Nytell/O'Hearn share a common registered-agent address or principals with each other or with a larger aggregator, which would move the verdict toward high confidence.

I'll research each cited reference against US 7,558,729 to build the § 102 analysis.

Now let me research the remaining cited references.

Let me research the final cited references.

Prior-Art / § 102 Anticipation Analysis — US 7,558,729 B1

Scope, method, and limits

This section builds on the previously generated claim mapping, obviousness, and bibliographic sections, and does not repeat them. I re-verified the technical content of five of the six patent citations of record by pulling their specifications directly (Tellabs '705; Nortel '635; Mindspeed '841; Lucent '435; RealNetworks '506). I could not complete a fresh search on the sixth citation (US 2007/0136053 A1) or on either non-patent citation (Tanrikulu 2003; Zhu 2005) before hitting the step limit — for those I rely on the previously generated sections and the fetched '729 record, and I flag it.

Framing that governs everything below. The question asked is § 102 anticipation — so the controlling rule is that a single prior-art reference must disclose every element of a claim, arranged as recited. Net MoneyIN v. VeriSign; MPEP § 2131. Disclosures that must be assembled from two or more references are § 103 territory, not § 102. That distinction matters here more than usual because — as set out in the prior section — the '729's independent claims are combinations of a music detector with downstream echo-canceller and speech-encoder controls, and the cited art splits cleanly: some references teach music detection, others teach echo-canceller NLP control, but none teaches both.

Bottom line up front: No reference cited on the face of US 7,558,729 anticipates any of claims 1–15 under § 102. Each is at most a § 103 reference. The detail below explains, per reference, which claim elements it reads on and why complete anticipation nevertheless fails.

1. The six patent citations of record

A. US 5,274,705 (Tellabs) — the NLP "remove from the send path" reference

What it discloses (verified from the specification): An adaptive-filter echo canceller that subtracts an echo estimate to leave a residual e[n], feeding a center-clipper NLP under dual adaptive thresholds — the suppression ("NLP") threshold and a lower noise-transparency (NT) threshold. Critically for the '729, the reference expressly teaches that when the near-end signal must not be distorted, "the NLP threshold should be set to zero, effectively removing the NLP from the send path," and its three-region graded transfer function lets low-level residual through with spectral content maintained. It even names music among the background content that the NLP would otherwise modulate: "the background signals modulated by the NLP usually consist of environmental sounds such as people talking, a computer printer, music, etc.," and warns that substituting pseudorandom noise is "nearly as obtrusive as inserting silence."

Claims it potentially raises under § 102 — and why anticipation fails.
The '705 reads directly on the structural spine shared by every independent claim — an echo canceller with an adaptive filter producing an error/residual signal and a nonlinear processor that removes the residual non-linear echo when engaged (claims 1, 4, 5 method; 6, 9, 10 system; 11, 14, 15 CRM). It also reads on the "bypassing the nonlinear processor" element, and supplies the motivation (music) for doing so.
But it contains no music detector and does not analyze the error signal to determine the existence of a music signal. Every independent claim of the '729 recites that music-detecting step. A single missing element defeats § 102. '705 is a § 103 reference, not an anticipatory one.

B. US 6,424,635 (Nortel) — the software-module / CRM reference

What it discloses (verified): An echo canceller implemented as processor 450 executing an adaptive filter module 470, a summing circuit 490, and an NLP module 480 (modules may live in on-chip memory 460 or off-chip memory 440), with the NLP clipping threshold dynamically adjusted via ERLE. Critically for claim format, claims 24 and 28 recite "a machine readable medium having embodied thereon a computer program" and a communication device with "means for storing software … including a nonlinear processor module."

Claims it potentially raises under § 102 — and why anticipation fails.
This is the reference that reads on the claim-format limitations of the CRM claims (11, 14, 15) and the "processor configured to…" framing of the system claims (6, 9, 10) — i.e., it establishes that implementing an echo canceller as a programmed processor with an adaptive-filter module and an NLP module, embodied on a machine-readable medium, was known. That is fatal to the format, but not anticipatory of the claims, because '635 likewise lacks any music detector. Its value is confined to knocking out the § 112/format “weight” of the CRM and system claims once the underlying method is shown obvious. § 103 reference only.

C. US 6,633,841 (Mindspeed) — the closest music-detection reference

What it discloses (verified from the specification and claim 1): A VAD correction/supervision circuitry 1120 containing music detection circuitry 1122 that overrides the VAD decision when it detects "at least one characteristic of the speech signal indicative of a music signal," switching coding to the G.729E high-rate extension instead of DTX. It performs statistical analysis (parameters incl. pitch, spectral difference, background-noise energy) over a window of frames (claim 25: sixty-four frames), compared against predetermined thresholds, with a running mean and stationary/hangover counters (the musdetect C routine, count_music, Mcount_music, count_consc, count_consc_rflag). Same assignee as the '729; one overlapping inventor (Benyassine).

Claims it potentially raises under § 102 — and why anticipation fails.
'841 reads on the "analyzing … using a music detector to determine existence of a music signal" element common to all nine independent claims, and it independently corroborates much of the threshold/statistical-analysis logic recited in dependent claims 3, 8, 13 (multi-frame window, running mean, comparison to thresholds, hangover counters). It also matches the specification's unclaimed "high bit rate on music" behavior.
But '841 is purely a coding/DTX reference — it has no echo canceller, no adaptive filter, no error signal, and no NLP. It therefore cannot anticipate any claim. § 103 reference only (and, as noted in the prior section, its own disclosure is what makes claim 3's algorithm look like a well-known statistical classifier).

D. US 6,760,435 (Lucent) — the noise-suppression gain / echo-control reference

What it discloses (verified): A single network speech-enhancement device 10 applying noise reduction 20, echo control 22, automatic level control 24, and noise compensation 26 to both directions; a per-band noise-reduction gain calculator, echo-control gain, master gain, and a thresholder; and a bi-directional speech detector 34 classifying near-end talk / far-end talk / double-talk / pauses. It notes users are "uncomfortable with periods of total silence" and discusses comfort noise.

Claims it potentially raises under § 102 — and why anticipation fails.
'435 reads on the noise-suppressor gain-control element of claims 1, 6, 11 (the encoder-side "attenuating … using the noise suppressor" half) and on the coexistence of echo control and noise reduction in one device — which is exactly the '729's FIG. 1/FIG. 3 integrated architecture. But it identifies signals only as "speech or noise," not music, and its gains are not gated on a music determination. The music element is again absent. § 103 reference only.

E. US 7,430,506 (RealNetworks) — the "music mis-coded as noise" reference

What it discloses (verified): Music fed to a voice-optimized LPC codec (EVRC) is treated as noise by the VAD/rate decision, causing "intermittent pause of music"; the remedy is preprocessing (AGC / pitch-harmonic enhancement) and "determining an encoding rate from a plurality of encoding rates based on a characteristic" of the signal (claim 1).

Claims it potentially raises under § 102 — and why anticipation fails.
'506 corroborates the problem statement the '729 itself admits (music mistaken for noise → destroyed/paused), and reads on the rate-selection concept discussed in the '729 specification — but rate selection 345 is not recited in any independent claim, so it maps onto no claim element. It has no echo canceller, no NLP, and no music-gated bypass. § 103 reference only.

F. US 2007/0136053 A1 (Acoustic Technologies) — the concept match that is not prior art on its dates

What it is: An application publication titled "Music detector for echo cancellation and noise reduction," Ebenezer (Acoustic Technologies, Inc.). Substantively it is the closest third-party art on the invention's core concept — music detection controlling both echo cancellation and noise reduction — and on its title alone it would read on the gist of claims 1, 6, 11.

Why it fails § 102 (and § 103) on the record: Its earliest date of record is a priority date of 2005-12-09, and it published 2007-06-14. Both are after the '729's priority date (2004-07-16) and after its filing date (2005-03-17). It is therefore:

• not § 102(a) art (not known/patented before the '729 invention), and
• not § 102(b) art (not published >1 yr before the '729 priority date), and
• not § 102(e) art (its U.S. filing, on the face of the record, postdates the '729 invention).

Its presence in the '729 "References Cited" table is consistent with a later-filed IDS (a reference may be listed even when it is not prior art). Net § 102 effect on the dates of record: none. ⚠️ Flag: I could not verify its U.S. filing/provisional chain before the tool limit. If an earlier provisional predating 2004-07-16 exists, this reference moves to the center of the analysis and could be near-dispositive. Treat its date as unresolved-but-probably-too-late, not settled.

2. Non-patent literature of record

Note on both NPL items: I was unable to re-verify these in this pass (tool limit). The ICASSP 2003 date is well established and safely § 102(b); the Zhu 2005 MMM date should be confirmed before reliance.

3. Synthesis — claim-by-claim § 102 result

Because anticipation requires one reference with all elements, and because the '729's independent claims each pair a music detector with a downstream control, the discipline of the analysis is to show the split. This table shows which references read on which element and confirms that no row closes on a single reference:

Design-of-record conclusion: the six citations divide into (i) echo-canceller NLP references — '705, '635 — which contain no music detector; and (ii) music/voice-classification and coding references — '841, '506, and (date-defective) '053 — which contain no echo canceller. '435 spans the two devices (echo control + noise reduction) but classifies only speech-vs-noise. That pattern — complementary disclosures, no single teaching the combination — is the classic signature of a § 103 case and the classic refutation of a § 102 case. Accordingly:

• No claim of US 7,558,729 is anticipated by any reference of record.
• The most probative references per element are: '841 for the music detector; '705 (and Tanrikulu) for the NLP-bypass/background-continuity; '435 for the noise-suppressor gain; '814/ '506 for the pitch tools; '635 for the CRM/processor claim format.
• The single reference that would be most dangerous — US 2007/0136053 A1 — is date-defective on the record and must have its priority chain resolved before it can be used at all.

4. Uncertainty flags specific to this section

1. US 2007/0136053 A1 date anomaly (highest priority). On its dates of record (priority 2005-12-09; published 2007-06-14) it is not § 102 or § 103 prior art against the '729, despite its appearance in the "References Cited" table. I could not verify its U.S. filing/provisional chain in this pass. If it claims an earlier provisional, the conclusion changes materially.
2. Zhu et al. (MMM 2005) exact publication date unverified — § 102(b) status unresolved; relevance marginal regardless.
3. '506 § 102(e) basis. I treated its date as the U.S. filing date (2004-01-08) because pre-AIA § 102(e) does not extend to a foreign priority claim. If any authority reads a § 102(e) date differently for a published application with a Korean priority, the reference is more dated, not less — the analysis is unaffected.
4. No fresh verification of US 2007/0136053 A1 or the two NPL items in this pass due to the tool-step limit; the four patents '705, '635, '841, '435 and '506 were verified against their specifications.
5. Effective filing date per claim (carried forward). The '729 is a CIP; the echo-canceller/encoder-coordination subject matter is new matter added 2005-03-17, while the FIG. 5 threshold algorithm traces to the parent (US 7,120,576). This affects the critical date for § 102(e) references but does not change the outcome here, since all four firm references predate even the later 2005-03-17 date.

I'll verify the one unverified prior-art reference and check a couple of legal/factual anchors before writing the § 103 analysis.

Let me verify the pivotal date question on the Acoustic Technologies reference — whether it carries an earlier benefit date that would make it available § 102/§ 103 art.

Obviousness Analysis under 35 U.S.C. § 103 — US 7,558,729 B1

Scope: Built strictly on the prior-art record established in the Prior Art section of this analysis (the six U.S. citations of record, the two non-patent publications, and the family members), and on the claim decomposition in the earlier Patent Summary. I do not re-describe the claims or the references here; I assume them.

One verification result materially changes the analysis, so it leads:

The single most on-point reference is definitively unavailable. I re-pulled the priority record for Acoustic Technologies' US 2007/0136053 A1 / US 8,126,706 B2 ("Music detector for echo cancellation and noise reduction"). Its only priority application is US 11/298,865, priority date 2005-12-09, filed 2005-12-09. There is no provisional and no earlier benefit claim. Against the '729's 2005-03-17 filing date (and 2004-07-16 priority), that reference fails § 102(a), § 102(b), and § 102(e). Because pre-AIA § 103(a) operates only on § 102 prior art, the one reference that discloses the complete concept — music detection → turn off residual echo suppression and turn off noise estimation — is also unusable in a § 103 combination. The earlier section's "interference/derivation" hypothesis is therefore not just a curiosity; it is the pivot on which the entire validity picture turns.

1. Governing law and the obviousness frame

The '729 was filed 2005-03-17, so pre-AIA § 103(a) applies (AIA first-to-file reaches only applications with effective filing on/after 2013-03-16). The test is Graham v. John Deere, 383 U.S. 1 (1966), as glossed by KSR Int'l Co. v. Teleflex Inc., 550 U.S. 398 (2007):

• KSR holds that a combination is obvious where the prior-art elements were known, the combination uses known methods to yield predictable results, or where the improvement is a predictable variation or a "simple substitution of one known element for another." 550 U.S. at 416–17, 421.
• KSR authorizes a "design incentive" / "market pressure" motivation and the "obvious to try" rationale where there is "a finite number of identified, predictable solutions." Id. at 421.
• Critical hedge: KSR did not abolish the motivation requirement. A combination must still be supported by an articulated reason with rational underpinning traced to the references, the problem, or the ordinary skill of the artisan — not to hindsight reconstruction of the claims.

Two pre-AIA gates that bound what art may be used at all:

• § 103(c) (pre-AIA): art qualifying only under § 102(e)/(f)/(g), and commonly owned or under common assignment obligation at the time the invention was made, cannot be used in a § 103 combination. Because at filing the '729, the parent '576, and US 6,633,841 were all Mindspeed property, this gate matters — but it does not rescue the patent, for the reason in § 2 below.
• § 103 uses only § 102 prior art. This is the sentence that kills the best reference (Acoustic).

2. Availability of each item as § 103 art (the gate analysis)

Net effect of the gate: the patent owner's single best validity answer is a date defense. Every reference that reads on the complete concept is off the table; the § 103 case must be assembled from references that each supply at most half of the claimed subject matter, welded by an articulated motivation. That is a contestable but winnable case under KSR — and it is exactly the posture the earlier sections flagged.

3. The person having ordinary skill in the art (PHOSITA)

Pre-AIA, the level of skill is a factual predicate. On this record: a B.S./M.S. in electrical engineering or computer science with roughly 2–5 years in telephony voice processing, specifically familiar with (i) adaptive transversal-filter echo cancellers and their NLPs (as taught by '705 and '635), (ii) ITU-T speech codecs G.711/G.729/G.729A/B/E and CELP/ACELP structures (as taught by '841 and '506), (iii) VAD/DTX, comfort-noise generation, and SID (as taught by '841 and '506), and (iv) gain-based noise reduction (as taught by '435). Note the field-defining fact that '841 itself establishes: the '729's four inventors and '841's co-inventor Benyassine were the same Mindspeed voice-DSP team. The PHOSITA is, functionally, that team — which cuts against the patent on the "would have been motivated" prong, because the artisan was already working the intersection of echo cancellation and music-aware coding.

4. Element map — what each reference actually supplies

Using the core-element labels from the prior sections (a = receive near-end input with echo; b = adaptive filter → error signal; c = music detector on the signal; d = bypass NLP on music; e = apply NLP when no music; and the three encoder-side families):

Read the table's diagonal: no row is fully populated by any single column. That is the anticipation defeat — and, symmetrically, the reason the § 103 case must be built from explicitly motivated joints, not from a single teaching.

5. The § 103 combinations

Combination I — the core (claims 1, 4, 5, 6, 9, 10, 11, 14, 15, all pre-encoder-side)

US 5,274,705 (Tellabs) + US 6,633,841 (Mindspeed) [optionally + US 6,424,635 (Nortel)]

• What the combination yields: '705 supplies (a), (b), (e) in full — adaptive-filter echo canceller, residual error signal e[n], NLP that "eliminates nonlinear components." '841 supplies (c) in full — a "voice activity detection correction and supervision module" that overrides the existing decision signal "if [it] detects at least one characteristic of the speech signal indicative of a music signal," together with dedicated music-detection circuitry. The joint (d)/(c)→NLP is completed by '705's own express teaching that the NLP threshold is set to zero, "effectively removing the NLP from the send path," in defined operating conditions, and by '705's NT-threshold structure that keeps the NLP "essentially transparent … generally without change in the spectral content."

• The four-part motivation (this is the crux, and it is strong):

  1. The references supply the motivation in their own text. '705's background expressly enumerates "environmental sounds such as people talking, a computer printer, music" as the very content that a conventional NLP destroys, and states that "replacing the background sounds with pseudorandom noise is nearly as obtrusive as inserting silence." Motivation is not reconstructed — it is quoted.
  2. The NPL corroborates identically. Tanrikulu et al., ICASSP 2003 — an examiner-cited reference — frames the same problem: "the removal of the residual echo through a conventional NLP will also remove the background signal. For the listener, the background signal will be modulated creating an annoying artifact." A reference's own articulation of the problem is a KSR-recognized motivation.
  3. '841 supplies the missing means for a problem '705 has already diagnosed. '705 identifies "music" as a protected class but provides only a level/energy-based NT threshold — an indirect, non-discriminating proxy. '841 converts the problem into a classification problem and supplies a detector. Where the reference of record diagnoses the deficiency and the second reference supplies the specific mechanism to cure it, the combination is KSR-compliant. In re ICON Health & Fitness, 496 F.3d 1374 (Fed. Cir. 2007).
  4. The joint is a known control-port substitution, not a redesign. '705's own figure architecture already exposes the NLP to external control inputs — an NLP threshold control port driven by the receive-path energy and by a double-talk-detector signal, plus a separately estimated NT threshold controlled by the background-noise estimate. Routing a third control signal (music-detect) to that same port, and asserting it to force the NLP transparent, is "the simple substitution of one known element for another" to yield a predictable result — precisely KSR's threshold formulation. In re Harza, 274 F.2d 669 (CCPA 1960) ("the mere addition of a known element to a known structure").

• Field-of-endeavor bridge (defeats a "different art" argument). '705 itself notes that "signal processing delays through new digital network elements, such as through the speech coding circuits for digital cellular systems, are becoming a significant source of transmission delays and, therefore, of new echo canceller applications." '705 thus itself ties the echo-cancellation art to the speech-coding art. '841 operates on the identical G.729 Annex A/B/E codec family that the '729's own FIG. 1 and FIG. 3 describe. The two references are in the same field of endeavor and the same field of use.

• Claim 3 / 8 / 13 (the three-threshold logic): add '705's own dual/multi-threshold structure — literally a "first, second and third transfer function" regime with an intermediate NT band between the NLP threshold and transparency ('705 claim 10) — combined with '841's multi-frame statistical analysis ("Statistical Analysis of Backward LPC Usage," FIG. 9 of '841). The '729's three thresholds (T1 noise, T2 music, T0 the intersection/unsure point) plus the cnt_mus/cnt_nomus multi-frame counter are then the routine merger of (i) a prior-art multi-zone threshold classifier with (ii) prior-art multi-frame statistical smoothing to suppress the exact instability the '729's own Background concedes in the prior art: conventional VADs "exhibit rapidly fluctuating outputs for music signals." An applicant's admission of the problem in the specification is usable for what the art taught. The residual-line designations (100–200 frames; 30-frame hangover) are optimization of a disclosed range, i.e., routine experimentation — not patentable weight.

Assessment: Strong as to the shared-echo-canceller core. This combination is the strongest single § 103 theory available on the record, and it survives the "teaching away" test because nothing in '705, '841, '435, or '506 teaches away from bypassing suppression on music — every one of them moves in the same direction (preserve coherent non-speech content).

Combination II — claims 1 / 6 / 11 (the noise-suppressor family)

Combination I + US 6,760,435 (Lucent, Etter & Chuang) [optionally + '506]

• What '435 adds: (F1). '435 discloses a network speech-enhancement device that computes a noise-reduction gain and applies it to the near-end signal, that cooperates with an external echo canceller, and that expressly reduces the residual echo not removed by that echo canceller — i.e., '435 is architecturally the same upstream-echo-canceller/downstream-gain-suppressor pipeline that claims 1/6/11 recite, one stage short of the '729's music control input.
• Motivation: (i) '841 establishes as a known proposition that music must not be processed as background noise (its entire raison d'être is the inadequacy of the DTX/noise path for music); (ii) '435's noise-reduction gain is derived from a noise estimate — so the art's own control variable is defeating for a music-bearing signal; (iii) a POSITA optimizing the pipeline would therefore bias the gain toward unity (0 dB) when music is present. KSR: "if a technique has been used to improve one device, and a person of ordinary skill in the art would recognize that it would improve similar devices in the same way, using the technique is obvious." 550 U.S. at 417.
• The "gradually reducing … to zero" limitation (the element the earlier section identified as the narrowest). This is where the § 103 theory is thinnest but not defeated. Two independent rationales:
  • Predictable variation / design choice. Ramping a gain to unity rather than stepping it is the textbook anti-click technique in a codec pipeline; the '729 itself gives the only stated reason — "to provide a smooth transition and avoid discontinuities" — which is a restatement of the well-known design constraint, not an inventive insight. Under KSR, "[a] person of ordinary skill is also a person of ordinary creativity." A hard-vs-gradual rate choice, with no rate, time constant, or frame count recited in the claim, is the paradigm of a predictable mechanical variation.
  • '435's own gain architecture. '435 computes and combines per-band gains; the art of gain-smoothing/soft-switching across frames is inherent to avoid modulation artifacts (the very "noise modulation"/"noise pumping" artifact '705 names and Tanrikulu names). A POSITA would not need the '729 to be told to ramp rather than to click.
• Honest counter: if the owner successfully construes "gradually reducing" to require a specific time-scheduled decay, a defendant must produce art showing soft-transition gain control in a noise-suppression path. '435 as characterized is the weakest-supported leg of this combination — see § 8.

Combination III — claims 4 / 9 / 14 (pitch-interpolation family)

Combination I + US 7,430,506 (Nam) + US 6,507,814 (admitted art)

• What the added art supplies: '506 establishes the operative technical premise: because voice-optimized LPC/EVRC codecs "obtain a pitch and a formant frequency" from the input, and "it is difficult to extract a pitch from a polyphonic music signal, whereas it is easy in case of a human voice," the formant/pitch-based processing chain degrades music — producing "intermittent pause[s]." '506's whole solution is to alter the encoder's rate decision using pitch-harmonic enhancement (PHE); it therefore expressly instructs an artisan that the pitch-processing branch is the branch that breaks on music, and that the encoder's pitch-based control path must be re-directed when music is present. US 6,507,814 (admitted into the '729 by incorporation) supplies the complementary teaching that pitch determination is itself driven by speech classification — i.e., that a classifier can gate the pitch stage.
• Motivation: with (c) music detection already supplied by '841 and the pitch stage already known to be speech-specific from '506, disabling the pitch stage on music is the direct application of the classification to the stage it invalidates. KSR: predictable use of a prior-art technique on a prior-art problem, with the solution (bypass the stage) being one of a finite set (bypass / reduce / substitute).
• The decoder-signaling limitation — handle with care. Claims 4/9/14 require transmitting information to the decoder to disable the decoder's pitch interpolation. The line of attack is that this is a functional necessity, not an invention: where the encoder suppresses a pitch stage whose output the decoder's interpolation assumes, failure to signal the decoder produces encoder/decoder mismatch. Both '506 and '814 are codec-pair disclosures (encoder + decoder), and '841's mode-selection architecture presupposes a paired decoder. A POSITA must keep the pair synchronized — an obvious consequence of the disabling step, not an added inventive contribution. In re Boesch, 617 F.2d 272 (CCPA 1980) (discovery of an inherent but unrecognized property does not confer patentability). This is the weakest element in the entire claim set, but the theory is defensible.

Combination IV — claims 5 / 10 / 15 (pitch pre-processing family)

Combination I + US 6,633,841 + US 7,430,506 / US 6,507,814

• Motivation is the tightest here, because '841 does the heavy lifting itself. '841 claims and discloses an extended codec with a "music speech signal coding module" distinct from the "background noise speech signal coding module," selected by overriding the VAD decision on music detection. When the art already teaches routing music to a different coding path, withholding speech-optimized pre-processing (pitch pre-processing "used to modify the speech characteristics or parameters … to ease the encoding process") on that path is the routine consequence of having separate paths. '506 independently teaches that the voice-optimized preprocessing chain is what damages music, and that preprocessing should be conditioned on signal type.

Assessment: Moderate-to-strong. Note the asymmetry the earlier section identified — rate selection (345) is not recited in any independent claim, so '506's strongest teaching cannot be leveraged directly against the claims; it functions as motivation and technical-premise evidence instead. That is a real limitation on this combination's power.

Combination V — claim 2 / 7 / 12 (hard-bypass alternative)

Combination II, minus the "gradual" limitation.
Claim 2 is the broader alternative to claim 1's gradual ramp ("bypassing the noise suppressor"). Once Combination II's motivation is accepted, a fortiori the hard-bypass variant is obvious and — paradoxically — it is supported by the art more directly than claim 1 is (the reference art teaches switching off, not ramping). This is important strategically: claim 1's narrow "gradual" element is harder to invalidate, but claim 2 is easier — and claim 2 is what a plaintiff would likely assert against a straightforward "mute the suppressor" implementation. Claim 2 therefore carries greater invalidity risk than claim 1 despite depending from it.

6. Why a POSITA would combine — consolidated KSR factors

1. Identical problem, stated identically in the art. '705 and Tanrikulu both name music/coherent background as the content a residual-echo processor must not destroy. The '729's own Background concedes "there is need in the art."
2. Complementary, not overlapping, teachings. '841 cleans up the decision side (what is music?); '705/'635 clean up the execution side (what does the NLP do?); '435 supplies the suppression stage; '506/'814 supply the pitch premise. Each teaches what the other lacks — the classic In re ICON pairing.
3. Predictable result / reasonable expectation of success. Each added element is a control-signal input to an existing stage or a bypass of an existing stage. Nothing requires redesign; the architecture of FIG. 2/3 of the '729 (echo canceller → error signal → encoder with suppressor/pitch modules) is the architecture '705+'435 and '841 disclose in combination.
4. Design incentives and market pressure (KSR-recognized). '506 documents the market reality: "Recently, several services for providing music to wireless phone users became available … transmission of music data to a cellular phone has been increasing." Where the industry is moving music through speech-optimized pipelines, an artisan has a strong incentive to make the pipeline music-safe. "Market forces" is an expressly sufficient KSR motivation.
5. Finite, predictable solution space. Once music is detected upstream of a suppressor, the artisan's options are exhausted by: (i) bypass, (ii) ramp down, (iii) substitute the gain. The '729 claims (i) and (ii). "Obvious to try" applies because the field presented no more than these options.
6. The specification supplies its own admissions. "music detector 235 can use any music detection algorithm and is not limited to the algorithm described"; the pitch stage is described as merely easing encoding; the gradual ramp's stated purpose is only to avoid discontinuities. Each admission narrows the distance to the art.

7. Claim-by-claim § 103 verdict

Aggregate: the patent is not vulnerable on anticipation (established in the prior section), and it is not vulnerable on the primary reference a defendant would most want (Acoustic — date-defective). It is vulnerable to a well-constructed, KSR-framed § 103 attack built on '705 + '841 + '435 + ('506/'814) — strongest against claims 2, 7, 12 (hard bypass), then 3, 8, 13 (thresholds), then 5, 10, 15 (pitch pre-processing), and weakest against 1, 6, 11 (gradual ramp) and 4, 9, 14 (decoder signaling).

8. Candid weaknesses in the § 103 case (what the patent owner will say)

I am flagging these because a § 103 opinion that only builds the prima facie case is not an opinion.

1. No reference places a music detector in the echo-canceller residual-error path. '841 detects on the encoder input; '705 has no detector. Owner's argument: the specific situs is the invention. Rebuttal: the '729's own architecture collapses the distinction — FIG. 2's error signal 219 is the input to the encoder of FIG. 3, and the spec expressly contemplates that "music detection can be performed outside of echo canceller 200, and a music detection signal can be received by echo canceller 200." The patent therefore disclaims any siting significance. Also: the claim says only "analyzing the error signal using a music detector" — a functional recitation satisfied by a detector operating on the same signal path.
2. The reference that reads on the complete concept is unavailable. This is the owner's best day-in-court argument, and it is purely a date argument — which means it is fragile in exactly one respect: if Acoustic's 2005-12-09 priority is ever displaced by an earlier benefit claim (a provisional, a parent, a foreign priority in an earlier-filed US case), the entire analysis inverts. I searched and found none — only US 11/298,865 and a stray "2005-12-05" entry. But the file wrapper for 11/298,865 should be pulled to confirm no § 119/§ 120 benefit claim was omitted from the published record. This is the highest-value open item.
3. '435 is cited here for its gain-based noise suppression, not for music control. If '435 is shown to lack any teaching of dynamic reduction of the suppression gain on classification, the owner will argue the F1 joint is supplied by attorney argument rather than by disclosure. Mitigation: '506 independently teaches that the encoder's rate/suppression decision is driven by a signal classification, and that music triggers the wrong decision — so '435's gain need not be the sole source of the bypass teaching.
4. § 103(c) contamination risk. If a defendant's combination leans on the CIP parent US 7,120,576 (or on any Mindspeed-invented disclosure that qualifies only under § 102(e)), § 103(c) may knock it out given common ownership at the time of invention. Combination I–IV above deliberately do not rely on the parent. Keep it that way.
5. The '841 "different inventive entity" / teaching-direction issue. '841's mode-override is framed around DTX/VAD coding-mode selection (SID/CNG), not echo cancellation. An owner will argue a POSITA reading '841 would think of coding mode, not residual-echo suppression, and that '841 teaches away from NLP bypass because it addresses a different downstream decision. Rebuttal: '841's contribution is the generic proposition that a music classification should override a pre-existing downstream decision; '705 supplies the downstream decision (NLP) and expressly identifies music as the thing it must not destroy. The motivation is the overlap of the two problems, which KSR permits even where the references are from different sub-fields of the same art.
6. No secondary-considerations record either way. No evidence of commercial success, copying, licensing-acquiescence, or long-felt-but-unmet need beyond the patent's own self-serving "there is need in the art." Conversely — and importantly for a defendant — there is likewise no § 103-swallowing objective evidence, and the patent is near expiry (~2027), which depresses the expected value of any invalidity challenge. See the earlier PTAB section's ROI point; it applies identically to a district-court § 103 case.
7. Non-§ 103 validity hazards worth noting (outside this task's scope, flagged because they interact). The patent's own specification and family create (a) a potential obviousness-type double-patenting exposure against US 7,120,576 / US 7,130,795 (different claims, same family, common ownership), and (b) a § 112(b) exposure on "gradually reducing … to zero" (no rate/time constant/frame count recited). Neither is § 103, but a validity challenge that pairs a § 112 indefiniteness count on "gradually" with the § 103 combination in Combination II is materially stronger than § 103 alone.

9. Bottom line

• Anticipation: no. (Confirmed in the prior-art section, and reinforced here.)
• Obviousness under § 103: a genuine, litigable, but non-overwhelming case — and, critically, one the patent owner can only defeat by dates.
  • The strongest combination on this record is US 5,274,705 (Tellabs) + US 6,633,841 (Mindspeed) + US 6,760,435 (Lucent), optionally + US 7,430,506 (Nam) and the admitted US 6,507,814, with Tanrikulu (ICASSP 2003) supplying corroborating problem-recognition.
  • It rests on express teachings rather than attorney-created motivation: '705 names music as the background the NLP spoils; Tanrikulu names the artifact; '841 supplies a music detector that overrides a downstream decision; '435 supplies the downstream suppression stage; '506 supplies the pitch-domain premise.
  • Best targets: claims 2, 7, 12 (hard bypass) ≫ 3, 8, 13 (threshold logic) ≈ 5, 10, 15 (pitch pre-processing) > 1, 6, 11 (gradual ramp) ≈ 4, 9, 14 (decoder signaling).
  • The decisive variable is not the merits — it is one date. If US 11/298,865 can be shown to carry any benefit claim earlier than 2005-03-17, Acoustic's US 8,126,706 becomes available, and claims 1, 2, 6, 7, 11, 12 move from "Moderate" to serious § 102 exposure. If not, the owner's date defense holds and the § 103 case must be won the harder way — through KSR motivation on a four-reference combination, which is doable but not clean.

10. Uncertainty flags (this section)

1. ⚠️ Verified this pass: Acoustic US 8,126,706 / US 2007/0136053 A1 has no priority earlier than 2005-12-09 per the published record → not § 102 art, therefore not § 103 art. Residual risk: check the file wrapper of application 11/298,865 for any omitted § 119/§ 120 benefit claim.
2. ⚠️ Verified this pass: US 7,430,506 is a pitch-harmonic-enhancement / rate-decision reference (Nam et al., RealNetworks Asia Pacific), US filing 2004-01-08, foreign priority 2003-01-09. It does not disclose disabling pitch interpolation or pitch pre-processing. Its § 103 role is motivational and premise-establishing only — it is not a disclosure of the F2/F3 elements. Treat my earlier "unverified" flag on this reference as now resolved, with the caveat above.
3. ⚠️ Verified this pass: US 6,633,841's inventors are Jes Thyssen and Adil Benyassine, assignee Mindspeed; it includes Music Detection Circuitry 1122 and a VAD correction/supervision override of a no-voice-activity decision. Its § 103 exposure survives § 103(c) only because it also qualifies under § 102(b) (issued 2003-10-14). If a court were to treat it as qualifying "only" under § 102(e) for some reason, the § 103(c) gate would reopen — I regard that as unlikely but not impossible.
4. ⚠️ US 6,760,435 (Lucent): I rely on the characterization in the prior-art section (per-band noise-reduction gain, cooperation with an external echo canceller). I did not independently re-pull its text this pass; the "gradual gain ramp" leg of Combination II is the least-verified link in this analysis. Direct retrieval recommended before relying on it.
5. ⚠️ US 5,274,705 (Tellabs): independently confirmed this pass — dual adaptive thresholds, an NT threshold below the NLP threshold, a first/second/third transfer-function regime ('705 claim 10), explicit enumeration of "music" as NLP-destroyed background, and express teaching to set the NLP threshold to zero to remove the NLP from the send path.
6. ⚠️ Date inconsistency persists in the prompt: the system header gives today as 2026-09-29; this task states April 26, 2026. I have not used either date in any computation — every date calculation above uses only application and reference filing/issue dates.
7. ⚠️ Statutory frame: all of the above is pre-AIA § 102/§ 103, correct for a 2005-03-17 filing. PGR and CBM are unavailable (as the earlier PTAB section established); IPR is the only AIA vehicle, and it can reach § 103 grounds but not § 112.