PTAB

IPR2023-00033

Apple Inc v. Zentian Ltd

Log In
Key Events
Petition
petition Intelligence

1. Case Identification

2. Patent Overview

  • Title: Speech Recognition System Having Parallel Large Vocabulary Recognition Engines
  • Brief Description: The ’319 patent discloses a speech recognition circuit that uses a plurality of parallel processors to process speech parameters. The processors are arranged in groups, with each group connected to a dedicated lexical memory.

3. Grounds for Unpatentability

Ground 1: Obviousness over Thelen, Bailey, and Chen

Claims 46, 50, 54, 64, and 67 are obvious over Thelen in view of Bailey and in further view of Chen.

  • Prior Art Relied Upon: Thelen (Patent 6,526,380), Bailey (Patent 5,459,798), and Chen (Patent 5,428,803).
  • Core Argument for this Ground:
    • Prior Art Mapping: Petitioner argued that Thelen taught a base speech recognition system with parallel recognizers (processors) that use context-specific recognition models to process parts of a large vocabulary. However, Thelen lacked an express input buffer, results memory, and a plurality of distinct lexical memories. Petitioner asserted Bailey, which teaches a multiprocessor pattern recognition system, supplied these missing elements by disclosing an input buffer and results memory (RAM 512) and a plurality of private memories (615a-c), each connected to a processor. Finally, Petitioner mapped Chen’s disclosure of a parallel processing architecture with processor clusters to the ’319 patent’s limitation of arranging processors in groups, where each group is connected to a shared memory.
    • Motivation to Combine: A POSITA would combine Bailey’s input buffer with Thelen’s system to improve efficiency by decoupling the rate of feature vector generation from the rate of recognition, a known technique for performance enhancement. A POSITA would further combine this system with Chen’s clustered architecture to logically group Thelen's context-specific recognizers (e.g., all "sports" contexts), allowing for more efficient retrieval and sharing of lexical data and predictably speeding up recognition.
    • Expectation of Success: Petitioner contended a POSITA would have a high expectation of success, as the combination involved applying well-known hardware architectures (buffers, clustered processors) from Bailey and Chen to a speech recognition system (Thelen) to achieve the predictable results of improved efficiency and processing speed.

Ground 2: Obviousness over Thelen, Bailey, Chen, and Huang

Claims 47-49 are obvious over Thelen in view of Bailey and Chen, and in further view of Huang.

  • Prior Art Relied Upon: Thelen (Patent 6,526,380), Bailey (Patent 5,459,798), Chen (Patent 5,428,803), and Huang (Patent 6,076,056).
  • Core Argument for this Ground:
    • Prior Art Mapping: This ground built upon the base combination of Thelen, Bailey, and Chen to address the limitations of claims 47-49, which required storing lexical data in a lexical tree data structure. Petitioner argued Huang taught this missing element, disclosing a prefix tree structure to represent a lexicon for speech recognition. Huang further taught building monophone models (an "n phone model" where n=1) for the tree's components and using those monophone models to dynamically generate context-dependent phone models (triphones) during recognition.
    • Motivation to Combine: A POSITA would be motivated by Huang's express goal of "reducing the magnitude of computation required for speech recognition" to organize the lexical data in the primary combination’s memories into Huang’s efficient tree structure. This was a known technique to improve computational performance in HMM-based recognition systems.
    • Expectation of Success: The modification required only reprogramming data storage, a task well within the skill of a POSITA, to implement a known efficiency-improving technique (lexical trees) in a similar system, leading to a reasonable expectation of success.

Ground 3: Obviousness over Thelen, Bailey, Chen, and Jiang

Claims 53 and 58-59 are obvious over Thelen in view of Bailey and Chen, and in further view of Jiang.

  • Prior Art Relied Upon: Thelen (Patent 6,526,380), Bailey (Patent 5,459,798), Chen (Patent 5,428,803), and Jiang (Patent 6,374,219).
  • Core Argument for this Ground:
    • Prior Art Mapping: This ground addressed limitations in claims 53 and 58-59 related to determining scores at word recognition events and pruning the search process based on those scores. Petitioner asserted Jiang taught these features. Jiang disclosed computing "probability scores" while traversing a prefix tree, outputting a word when a "word boundary" is detected (the word recognition event), and performing a "pruning technique" by dropping low-scoring branches to reduce the search space.
    • Motivation to Combine: A POSITA would be motivated to incorporate Jiang's scoring and pruning methods into the base speech recognition circuit to gain the computational savings and efficiency that Jiang explicitly described. This would be a straightforward application of a known optimization technique to improve a similar system.
    • Expectation of Success: Implementing pruning based on intermediate scores was a known method for accelerating tree-based searches. A POSITA would expect this modification to predictably improve the performance of the combined system without undue experimentation.
  • Additional Grounds: Petitioner asserted an additional obviousness challenge against claim 51 based on the combination of Thelen, Bailey, and Chen in further view of Katz (Patent 4,831,550), which allegedly taught using a dedicated language model processor.

4. Arguments Regarding Discretionary Denial

  • Petitioner argued that the Board should not exercise discretionary denial under Fintiv. It contended that the co-pending district court litigation was in its nascent stages, with a trial date scheduled for approximately the same time as the Final Written Decision in the IPR would be due. Therefore, the Board's review would conclude before the court invested substantial resources in its own invalidity analysis, avoiding duplicative efforts.

5. Relief Requested

  • Petitioner requested the institution of an inter partes review and the cancellation of claims 46-51, 53-54, 58-59, 64, and 67 of the ’319 patent as unpatentable.