US 5819222

US Patent 5819222, titled "Task-constrained connected speech recognition of propagation of tokens only if valid propagation path is present," was assigned to British Telecommunications PLC. The inventors are Samuel Gavin Smyth and Simon Patrick Alexander Ringland. The patent was filed on March 31, 1994, and issued on October 6, 1998.

Abstract:
The patent describes a speech recognition system designed for connected speech. It uses a network with multiple vocabulary nodes, some of which have an associated signature. During operation, partial recognition paths are evaluated at decision nodes situated between the start and end of the recognition path. Each decision node has a predefined set of valid accumulated signatures. A "token" (representing a partial recognition path) is only allowed to proceed through a decision node if its accumulated signature matches one of the valid signatures associated with that node.

Plain-Language Overview of Independent Claims:

• Claim 1 (System Claim): This claim describes a token-passing speech recognition system. The system includes a network that models expected speech, containing various vocabulary nodes. At least one of these vocabulary nodes has a unique identifier called a "signature." The system also has mechanisms to examine partial recognition paths at specific "decision nodes" within the network. Each of these decision nodes is pre-programmed with a list of acceptable "accumulated signatures." A token (which carries information about a partial recognition path and its accumulated signature) is only allowed to move forward through a decision node if its accumulated signature is on that node's list of valid signatures.

• Claim 9 (System Claim): This claim describes a speech recognition system that processes input speech using a network of vocabulary nodes. Each vocabulary node has a model that measures how similar the input speech is to a recognized speech unit, producing a "first parameter" for similarity. The system compares these "first parameters" from different models to decide which ones should continue through the network. At the end of the recognition process, it identifies the "first parameter" with the highest similarity and the corresponding recognized speech. Crucially, the system assigns a "signature" to at least one vocabulary node and associates each "first parameter" with a "second parameter" that represents the accumulated signatures of its path through the network. A "validation means" checks if this "second parameter" is part of a valid set of "second parameters," allowing the "first parameter" to proceed only if its associated "second parameter" is valid.

• Claim 17 (Method Claim): This claim describes a method for recognizing connected speech. It involves creating a network to model expected speech, with vocabulary nodes each having a model to indicate similarity between input and recognized speech (a "first parameter"). The method then compares these "first parameters" to decide which ones should advance through the network. At the network's output, the method identifies the "first parameter" with the highest similarity and the recognized speech. The method also assigns a "signature" to at least one vocabulary node and links each "first parameter" to a "second parameter" representing the accumulated signatures of its path. These "first parameters" are only allowed to propagate if their associated "second parameter" is found within a valid set of "second parameters."

• Claim 22 (Method Claim): This claim describes a token-passing method for recognizing connected speech. It involves using a network with vocabulary nodes, where at least one node has a signature, to model expected input speech and partial recognition paths at decision nodes. Each decision node has a predefined set of valid accumulated signatures. A token that reaches a decision node is only propagated further if its accumulated signature, which reflects its path through the network, is one of the valid signatures for that specific decision node.

CAFC 2026 Dockets:
As of April 26, 2026, US Patent 5819222 was cited as a prior art reference in a U.S. Court of Appeals for the Federal Circuit (CAFC) case in 2026. The case involved Apple as a petitioner against Zentian Ltd., owner of U.S. Patent No. 10,839,789. Apple argued that claims of US10839789, specifically claim 1 and claim 29, were obvious over US5819222 (referred to as "Smyth") or a combination of US5819222 and US Patent No. 6,832,194. The CAFC vacated and remanded a decision from the Patent Trial and Appeal Board (PTAB) in favor of the patent owner, indicating errors in the PTAB's reading of the claims for US10839789. This means US5819222 was used as a foundational prior art reference in arguments concerning the patentability of another patent, rather than being directly litigated itself.

I'll research litigation involving US Patent 5819222 across patent litigation databases.

Let me check litigation-specific databases and search for direct assertions of this patent.

Litigation Search: US Patent 5,819,222

Bottom line

No district court litigation asserting US Patent 5,819,222 has been identified. Across patent-litigation aggregators (Unified Patents litigation portal, RPX Insight, Stanford NPE Litigation Database, CourtListener/RECAP, Docket Alarm, Justia/UniCourt dockets) and general web searches on the literal string "5819222," the patent does not appear as an asserted patent-in-suit, counterclaim patent, or declaratory-judgment patent in any case.

What the searches returned (and why the hits are all false positives)

Every literal "5819222" hit was an unrelated identifier, not this patent:

The one proceeding where '222 appears — but as prior art, not as a cause of action

Per the previously generated "CAFC 2026 Dockets" section of this analysis (which I build on rather than repeat), US 5,819,222 is recorded as having been cited as a prior-art reference in a 2026 Federal Circuit appeal involving Apple (petitioner) v. Zentian Ltd., owner of US 10,839,789. There, Apple argued claims 1 and 29 of the '789 patent were obvious over US 5,819,222 ("Smyth") or over Smyth in combination with US 6,832,194; the CAFC vacated and remanded the PTAB's decision in favor of the patent owner.

Important characterization: that is a PTAB appeal about the patentability of US 10,839,789, in which '222 served as an invalidating reference. US 5,819,222 was not a patent-in-suit and Zentian was not accused of infringing it. It is not "litigation involving US 5,819,222" in the sense of an assertion of its claims.

Verification caveat / flag: I was unable to independently confirm the Apple v. Zentian CAFC details through live search in this session (my targeted query on that case returned no corroborating results before the search budget was exhausted). I am therefore carrying that item forward as stated in the earlier section rather than as an independently verified fact. If precision matters — e.g., for a filing or opinion — the docket should be pulled directly from the CAFC docket or PACER.

Related BT activity that does not involve this patent

For completeness, and to avoid conflating them: British Telecommunications plc has asserted other patents, but never '222 as far as the record shows, e.g.:

• British Telecommunications plc v. Google Inc., No. 1:11-cv-01249 (D. Del., filed 2011-12-15; Judge Leonard P. Stark) — asserted US 6,151,309; 6,169,515; 6,397,040; 6,578,079; 6,650,284; 6,826,598. Disposed.
• Conexant Systems, Inc. v. British Telecommunications PLC, No. 3:07-cv-00573 (S.D. Cal.) and British Telecommunications plc v. Broadcom Corporation, No. 3:09-cv-03662 (D.N.J.) — involving US 5,153,591, not '222.
• Adaptive Spectrum and Signal Alignment Inc. v. British Telecommunications plc (UK, [2013] EWHC 3768 (Pat); [2014] EWCA Civ 1462) — DSL patents, not '222.

Case table

Supporting evidence and reasoning

• The patent is recorded on Google Patents as "Expired – Lifetime," with an anticipated expiration of 2015-10-06, and issued 1998-10-06 to British Telecommunications PLC. An expired patent owned continuously by an operating company (see the assignment section: single assignment, Reel 007822/0544, recorded 1995-10-11) is a strong structural reason to expect little or no assertion history.
• The previously generated "PTAB challenges" section reports that the USPTO ODP API returns no AIA trial proceedings for US 5,819,222, and the "NPE / troll-pattern signals" section concludes "Insufficient data" with no litigation found. My searches are consistent with both.

Caveats

1. Absence of results is not proof of absence. Litigation databases are incomplete for pre-2000s cases, sealed complaints, and cases that settled before service; PACER full-text search has coverage gaps. I could not complete a PACER/CAFC docket-level confirmation within this session's search budget.
2. I interpreted the identifier literally as "5819222" and did not auto-correct. Note in particular the near-miss US 5,819,220 (Sarukkai et al.), which appears in at least one IDS citation; a search for "US 5,819,220" returns different results and must not be substituted for this patent.
3. The Apple v. Zentian item should be treated as inherited from the earlier generated section and pending independent verification, not as a fresh finding.
4. For a definitive answer, the recommended sources are: the CAFC docket search, PACER Case Locator (party search "British Telecommunications" as plaintiff for patent causes), and Docket Navigator / Unified Patents litigation search filtered to patent number 5,819,222.

Proceedings overview

The USPTO ODP API returns no AIA trial proceedings for US Patent 5819222 as of the most recent ingest. Therefore, there is no PTAB activity on file for this patent.

Strategic summary

As there are no PTAB proceedings on file for US Patent 5819222, all claims (1-29) remain untested by AIA trial proceedings.

Since there have been no IPR, PGR, or CBM proceedings, there is currently no estoppel landscape established under § 315(e)(2) for this patent. All prior-art grounds remain available for potential future challenges.

The absence of PTAB activity suggests that this patent has not been subjected to the scrutiny of AIA trials, which can be a signal that it has not been extensively asserted in recent years, or that potential challengers have not yet found sufficient grounds or motivation to file an IPR.

Recommended next steps

If facing an assertion of US Patent 5819222, a defendant would not have the benefit of any claims being invalidated by PTAB proceedings. Given the lack of PTAB activity, a thorough prior art search would be a crucial first step to identify potential grounds for an IPR or other invalidity challenges.

Inventors

• Samuel Gavin Smyth (British Telecommunications PLC)
• Simon Patrick Alexander Ringland (British Telecommunications PLC)

Original assignee

The original assignee on US Patent 5819222 is British Telecommunications PLC. British Telecommunications PLC (now BT Group plc) is a major global telecommunications services company, and it is highly likely they shipped products embodying the claims related to speech recognition in telecommunications environments. BT Group plc is currently an operating company.

Assignment timeline

• 1995-09-11 (executed) / recorded 1995-10-11 — Reel 007822/0544
  • Conveyance: Assignment
  • Assignor: SMYTH, SAMUEL GAVIN; RINGLAND, SIMON PATRICK ALEXANDER
  • Assignee: BRITISH TELECOMMUNICATIONS PUBLIC LIMITED COMPANY
  • Correspondent: NOT RECORDED.
  • Context: Internal reorg (assignment from inventors to original assignee)

There are no further assignment records for US5819222 in the USPTO Patent Assignment Search database beyond the initial assignment from the inventors to British Telecommunications PLC.

Timeline diagram

timeline
    title Ownership of US 5819222
    1994 : Filed by British Telecommunications PLC
    1995 : Assigned to British Telecommunications PLC
    1998 : Issued to British Telecommunications PLC

NPE / troll-pattern signals

1. Shell-entity transfer — not present. The only recorded transfer is from the inventors to the original operating company assignee.
2. Known asserter in the chain — not present. British Telecommunications PLC (BT Group plc) is a known operating company, not a patent asserter.
3. Repeat correspondent across the chain — not present. Only one assignment is recorded, and no correspondent information is provided.
4. Cascading transfers — not present. Only one assignment is recorded.
5. Pre-litigation transfer — not present. No litigation involving this patent as a plaintiff has been found, and the only transfer is from the inventors to the original assignee well before its issue date.
6. Bankruptcy fire-sale — not present. British Telecommunications PLC is an active operating company.
7. Privateering — not present. No evidence of transfer to an NPE for assertion on behalf of an operating company.
8. Defensive aggregator (anti-NPE) — not present. The patent remains with the original operating company.

Verdict

Insufficient data. The USPTO Assignment Center search reveals only the initial assignment from the inventors to the original operating company, British Telecommunications PLC (Reel 007822/0544, recorded 1995-10-11). There are no subsequent recorded assignments or other legal events that would indicate a transfer to an NPE or a defensive aggregator.

USPTO Assignment Center Search for US5819222

To identify the most relevant prior art for US Patent 5819222, I will examine the "Citations" section of the patent on Google Patents, which typically includes "Prior Art Citations (U.S. Patent Documents)" and "Foreign Patent Documents" and "Other Publications". I will focus on the cited U.S. patents.

Here's an analysis of the patent citations listed for US5819222, focusing on their potential anticipation of the claims under 35 U.S.C. § 102:

Most Relevant Prior Art for US Patent 5819222

The most relevant prior art for US Patent 5819222 would be those documents that disclose, or render obvious, the core invention of using accumulated signatures for propagating tokens in a task-constrained connected speech recognition system. The following U.S. patents are cited and appear particularly relevant to various aspects of speech recognition, including continuous speech, Hidden Markov Models, and network-based processing, which form the technical context of US5819222.

Here's a breakdown of the cited US patents, with a brief description and potential anticipation:

• US4783804A: Hidden Markov model speech recognition arrangement

  • Full Citation: US4783804A (American Telephone And Telegraph Company, At&T Bell Laboratories)
  • Publication Date: 1988-11-08
  • Filing Date: 1985-03-21
  • Brief Description: This patent describes a speech recognition arrangement using Hidden Markov Models (HMMs). HMMs are explicitly mentioned in US5819222 as a suitable word representation model (description, "Definitions" section, "the word representation models"). Therefore, this patent would likely anticipate the use of HMMs in a speech recognition system, as broadly covered by claims such as Claim 4 (which specifies HMMs).
  • Potential Anticipation (35 U.S.C. § 102): Likely anticipates the use of Hidden Markov Models as a word representation model (e.g., in relation to Claim 4, Claim 25).

• US4888823A: System for continuous speech recognition through transition networks

  • Full Citation: US4888823A (Kabushiki Kaisha Toshiba)
  • Publication Date: 1989-12-19
  • Filing Date: 1986-09-29
  • Brief Description: This patent focuses on continuous speech recognition using transition networks. The concept of a "transition network" is fundamental to US5819222, which describes a network for modeling expected input speech and propagating tokens through it. This patent would likely anticipate the general architecture of continuous speech recognition systems employing networks.
  • Potential Anticipation (35 U.S.C. § 102): Likely anticipates the general concept of a network for modeling expected input speech in a continuous speech recognition system (e.g., in relation to claims 1, 9, 17, 22, which all involve a "network for modelling expected input speech").

• US5040127A: Continuous speech recognition system

  • Full Citation: US5040127A (Motorola, Inc.)
  • Publication Date: 1991-08-13
  • Filing Date: 1986-06-02
  • Brief Description: This patent describes a continuous speech recognition system. Similar to US4888823A, this patent contributes to the general state of the art in continuous speech recognition and network-based processing. The background of US5819222 discusses "connected recognition systems" and the challenges of "token passing" which is a core feature of this prior art.
  • Potential Anticipation (35 U.S.C. § 102): Likely anticipates aspects of connected or continuous speech recognition systems.

• US5228110A: Method for recognizing N different word strings in a speech signal

  • Full Citation: US5228110A (U.S. Philips Corporation)
  • Publication Date: 1993-07-13
  • Filing Date: 1989-09-15
  • Brief Description: This patent describes a method for recognizing multiple word strings. While not explicitly mentioning "signatures" or "decision nodes" as defined in US5819222, the recognition of "N different word strings" implies a need to manage and distinguish between multiple potential recognition paths, which is the problem US5819222 addresses with its signature mechanism.
  • Potential Anticipation (35 U.S.C. § 102): Potentially anticipates systems and methods for recognizing multiple word sequences, which could be relevant to the underlying problem addressed by the signature-based path validation.

• US5309547A: Method of speech recognition

  • Full Citation: US5309547A (Matsushita Electric Industrial Co., Ltd.)
  • Publication Date: 1994-05-03
  • Filing Date: 1991-06-19
  • Brief Description: This patent describes a general method of speech recognition. Without a more detailed description, it's hard to pinpoint specific anticipation. However, as it predates the issuance of US5819222, it could cover foundational elements of speech recognition.
  • Potential Anticipation (35 U.S.C. § 102): May anticipate general steps or components of speech recognition methods.

Key Distinguishing Feature of US5819222

The novelty of US5819222 appears to reside in the specific mechanism of propagating tokens only if their accumulated signature is valid at intermediate decision nodes within the network. While prior art dealt with continuous speech recognition and network structures, the explicit use of "signatures" assigned to vocabulary nodes, accumulated along a path, and then validated at "decision nodes" to prune invalid paths seems to be the core inventive step. This approach is claimed to reduce memory requirements and, in some cases, increase accuracy compared to dynamically building or severely pruning a full tree network.

The CAFC case from 2026, where US5819222 (Smyth) was used as prior art against US10839789 (Zentian Ltd.), further highlights that "Smyth" provided a foundation for speech recognition technologies. The dispute in that case centered on the "acoustic model memory" and "integrated circuit" limitations of the later patent, implying that "Smyth" taught significant aspects of speech recognition systems, including those that could be implemented on an integrated circuit with a calculating apparatus. This underscores its general relevance as foundational prior art in the field of speech recognition.

Obviousness Analysis of US Patent 5819222 Under 35 U.S.C. § 103

A person having ordinary skill in the art (PHOSITA) in 1993, in the field of connected speech recognition, would have been familiar with techniques such as Hidden Markov Models (HMMs), dynamic time warping, and network-based speech recognition systems, as well as the trade-offs between accuracy, computational load, and memory requirements.

The core innovation of US Patent 5819222 lies in its method and apparatus for applying grammar constraints to connected speech recognition by propagating "tokens" (representing partial recognition paths) only if their "accumulated signature" reflects a valid path within the network, as checked at "decision nodes." This mechanism aims to reduce memory requirements compared to a full tree network.

Prior Art References

The following prior art references, as cited or discussed within US Patent 5819222, are relevant to an obviousness analysis:

1. Young et al., "Token Passing: a Simple Conceptual Model for Connected Speech Recognition Systems," Cambridge University Engineering Department, 1989. This publication describes a token-passing model for connected speech recognition, where tokens containing partial path information and an accumulated score are propagated through a transition network. Crucially, Young et al. states that "If more than one token is presented to a node, only the best scoring token is propagated onwards through the network."
2. Austin et al., "A Unified Syntax Direction Mechanism for Automatic Speech Recognition Systems Using Hidden Markov Models," ICASSP, vol. 1, 1989. The title of this non-patent citation suggests a mechanism for applying grammar or syntax constraints in HMM-based speech recognition systems.
3. General Knowledge in the Art: The patent acknowledges that constructing networks (e.g., tree networks) for enumerating all possible word sequences can become "problematic as the recognition task size grows large" due to significant memory requirements (e.g., "a million nodes" for 250,000 7-digit sequences could occupy "50 Mb" of storage). This highlights a known problem that the claimed invention sought to address. The concept of "grammar constraints" or "task constraints" in speech recognition was also well-known for limiting the search space to a predefined set of valid utterances (e.g., account numbers or product codes).

Obviousness Argument

Claims 1, 9, 17, and 22 of US Patent 5819222, as independent claims, all center on the use of accumulated signatures for path validation.

A PHOSITA would have been motivated to combine the teachings of Young et al. with the general knowledge regarding grammar constraints and the problem of large memory requirements in tree-based recognition systems.

Combination of References:

• Young et al. (1989) teaches a token-passing speech recognition system for connected speech, utilizing a network with vocabulary nodes, where tokens carry partial path information and accumulated scores. It also describes a method of pruning by propagating only the "best scoring token" at a node.
• General knowledge in the art (including implicit teachings of Austin et al. (1989) on syntax direction and the problem statement in US5819222): A PHOSITA would recognize the critical need to apply grammar constraints in task-constrained connected speech recognition to limit the search space and improve accuracy, especially for predefined sequences like digit strings. The PHOSITA would also be acutely aware of the memory and computational inefficiencies inherent in explicitly representing large grammars as full tree networks. Furthermore, the general concept of uniquely identifying sequences or paths with codes or "signatures" and performing lookups against a list of valid identifiers is a fundamental principle in computer science and data processing.

Motivation to Combine:

A PHOSITA, aiming to improve the memory efficiency and reduce the computational burden of applying grammar constraints within a token-passing speech recognition system (as taught by Young et al.) for large, task-constrained vocabularies, would have been motivated to replace the explicit, memory-intensive tree grammar structure with a more compact and dynamic method of grammar enforcement.

Given the existing token-passing framework from Young et al., it would have been an obvious design choice to embed grammar validation directly into the token propagation process. To achieve this, a PHOSITA would consider:

1. Assigning a unique identifier (a "signature") to each vocabulary unit or node: This would allow for tracking the sequence represented by a token's path.
2. Accumulating these signatures along the token's path: This would create a "second parameter" (as in Claim 9) or "accumulated signature" (as in Claim 1 and 22) that uniquely represents the partial recognized utterance.
3. Storing a compact representation of valid grammar paths: Instead of a full tree, this could be a list of valid accumulated signatures.
4. Implementing a validation step at intermediate points (e.g., "decision nodes"): By checking if a token's accumulated signature is present in the list of valid signatures, invalid paths could be discarded early in the recognition process. This contrasts with Young et al.'s score-based pruning and directly addresses the memory problem by eliminating the need to explicitly represent and traverse large portions of an invalid grammar tree.

The patent itself emphasizes that "Such a system replaces a full tree network with a very much smaller one... the memory requirements for the network are likely to be under 2 Mbytes, a considerable saving," explicitly stating the problem and the benefit of its solution over prior art tree networks. This motivation to reduce memory and computational overhead for grammar-constrained speech recognition, by using a compact signature-based validation within an existing token-passing framework, would have been apparent to a PHOSITA.

Therefore, the combination of Young et al.'s token-passing system with general knowledge regarding the limitations of large tree grammars and the utility of unique identifiers for path validation would render the claimed invention, particularly the signature-based token propagation mechanism in claims 1, 9, 17, and 22, obvious to a PHOSITA.