IPR2023-00036
Apple Inc v. Zentian Ltd
1. Case Identification
- Case #: IPR2023-00036
- Patent #: 10,839,789
- Filed: November 15, 2022
- Petitioner(s): Apple Inc.
- Patent Owner(s): Zentian Limited
- Challenged Claims: 1-2, 4, 6-14, 16-18, 20-29, 35, 37-39, 42-45
2. Patent Overview
- Title: Acoustic Coprocessor
- Brief Description: The ’789 patent is directed to an acoustic coprocessor for processing audio data. The system calculates distances indicating the similarity between a feature vector derived from an audio signal and various acoustic states of a stored acoustic model, with the purported novelty being that the calculating apparatus and acoustic model memory are fabricated on a single integrated circuit.
3. Grounds for Unpatentability
Ground 1: Obviousness over Smyth and Mozer - Claims 1, 10-11, 14, and 17 are obvious over Smyth in view of Mozer.
- Prior Art Relied Upon: Smyth (Patent 5,819,222) and Mozer (Patent 6,832,194).
- Core Argument for this Ground:
- Prior Art Mapping: Petitioner argued that Smyth, which relates to speech recognition circuits, disclosed all elements of independent claim 1 except for the final limitation requiring the calculating apparatus and memory to be on a single integrated circuit (IC). Smyth’s speech recognition system was shown to include a classifier 34 (the "acoustic coprocessor") comprising a classifier processor 341 (the "calculating apparatus") and a state memory 342 (the "acoustic model memory"). Petitioner contended Smyth’s classifier receives feature vectors and calculates probabilities (distances) based on stored acoustic states. To meet the single IC limitation, Petitioner asserted that Smyth’s teaching of using a Motorola DSP56000—a device known to combine a processor and memory on a single chip—renders the claim obvious. Alternatively, Petitioner argued Mozer explicitly taught an "audio recognition peripheral integrated circuit" containing both a vector processor (calculating apparatus) and memory (acoustic model memory) on a single IC for the same purpose of speech processing.
- Motivation to Combine: A POSITA would combine Mozer's single-IC architecture with Smyth's system to achieve the known advantages of integration, such as offloading burdensome processing from a main CPU to a dedicated peripheral, reducing system complexity, and saving space. This was argued to be the application of a known technique (single-chip integration) to a similar system (Smyth) to yield predictable results.
- Expectation of Success: A POSITA would have had a reasonable expectation of success because both references describe similar speech recognition hardware, and integrating a processor and memory onto a single IC was a well-established practice in the art for improving performance and efficiency.
Ground 2: Obviousness over Smyth, Mozer, and Nguyen - Claims 2, 4, 6-9, 12, 16, 18, and 20-21 are obvious over Smyth, Mozer, and Nguyen.
Prior Art Relied Upon: Smyth (Patent 5,819,222), Mozer (Patent 6,832,194), and Nguyen (Patent 6,879,954).
Core Argument for this Ground:
- Prior Art Mapping: This ground built upon the base combination of Smyth and Mozer to address limitations in various dependent claims. For claim 2, which requires the acoustic model to contain "Gaussian parameters," Petitioner argued that while Smyth’s model is probability-based, it does not explicitly disclose Gaussian parameters. Nguyen was introduced to supply this teaching, as it described using Hidden Markov Models (HMMs) where complex probability values for acoustic states are represented by Gaussian distributions to create a more "robust model." For claim 9, which requires overlapping time periods for distance calculation and word search, Nguyen was cited for teaching contemporaneous processing, wherein a similarity measure (distance calculation) is computed for a subsequent group of feature vectors while a search process is performed on a previous group.
- Motivation to Combine: A POSITA would combine Nguyen’s teachings with the Smyth/Mozer system to improve the system’s performance and accuracy. The express motivations stated in Nguyen—to provide a more robust acoustic model and to improve processing efficiency through contemporaneous operations—would have driven a POSITA to incorporate these known techniques into Smyth’s similar architecture.
- Expectation of Success: Success was reasonably expected because using Gaussian parameters in HMMs and implementing pipelined or contemporaneous processing were common, well-understood techniques in the field of speech recognition that would have been straightforward to implement.
Additional Grounds: Petitioner asserted eight additional obviousness grounds (Grounds 3-10) that built upon the primary references by adding further prior art to meet specific dependent claim limitations. These additional references and their primary teachings included Houser for using FLASH memory, Vensko for generating an interrupt upon completion of a task and for Direct Memory Access (DMA) controllers, Comeau for using a status register, Boike for fabricating multiple distinct processors on a single IC, Whittaker for representing acoustic model parameters with integer data, Ichikawa for using a plurality of microphones, Seshadri for implementing the system on various devices like laptops and mobile phones, and Steinbiss for implementing the system in a consumer appliance.
4. Arguments Regarding Discretionary Denial
- Petitioner argued that the Board should not exercise discretionary denial under 35 U.S.C. §314(a) based on Fintiv factors. The co-pending district court litigation was described as being in a nascent stage, with a trial date set no earlier than April 22, 2024, which is contemporaneous with the projected Final Written Decision deadline. Petitioner asserted that the court had not yet invested substantial resources in invalidity analysis, as a Markman hearing was still months away and final invalidity contentions were not yet due. Therefore, instituting the IPR was presented as a more efficient path to resolving the patentability dispute.
5. Relief Requested
- Petitioner requests institution of IPR and cancellation of claims 1-2, 4, 6-14, 16-18, 20-29, 35, 37-39, and 42-45 as unpatentable.