Apple Inc v. Andrea Electronics Corp

1. Case Identification

• Case #: IPR2017-00626
• Patent #: 6,363,345
• Filed: January 9, 2017
• Petitioner(s): Apple Inc.
• Patent Owner(s): Andrea Electronics Inc.
• Challenged Claims: 1-25 and 38-47

2. Patent Overview

• Title: System, Method and Apparatus for Cancelling Noise
• Brief Description: The ’345 patent discloses systems and methods for reducing noise from an audio signal using a spectral subtraction technique. The claimed invention identifies noise in individual frequency bins by comparing the magnitude of each bin to an adaptive threshold, which purportedly improves upon prior art methods that required a "voice switch" to detect pauses in speech.

3. Grounds for Unpatentability

Ground 1: Anticipation of Claims 1-3, 12-13, 21, 23, and 38 under 35 U.S.C. §102

• Prior Art Relied Upon: Hirsch (H. G. Hirsch and C. Ehrlicher, “Noise estimation techniques for robust speech recognition,” 1995).
• Core Argument for this Ground:
  • Prior Art Mapping: Petitioner argued that Hirsch, a 1995 article, describes a complete noise reduction scheme that anticipates every limitation of the independent claims. Hirsch teaches converting a noisy audio signal into the frequency domain ("frequency bins" or "sub-bands"), and then, for each bin, setting an "adaptive threshold" based on a noise estimate. It detects noise by comparing the signal magnitude in each bin to this threshold. If the magnitude is less than the threshold, the signal is considered noise, and the noise estimate is updated. The estimated noise is then subtracted from the signal. This process, Petitioner contended, directly maps to the elements of independent claims 1 (apparatus) and 38 (method).
  • Key Aspects: Petitioner asserted that a central feature of both the ’345 patent and Hirsch is the elimination of the need for explicit speech pause detection, which is accomplished by using the bin-by-bin adaptive threshold. Hirsch's disclosure of a running, weighted average for noise estimation was also argued to anticipate claim 12's "averaging unit."

Ground 2: Obviousness of Claims 4-11, 25, 39-42, and 46 over Hirsch in view of Martin

• Prior Art Relied Upon: Hirsch (1995 publication) and Martin (Rainer Martin, “An Efficient Algorithm to Estimate the Instantaneous SNR of Speech Signals,” 1993).
• Core Argument for this Ground:
  • Prior Art Mapping: Petitioner argued that Hirsch does not expressly disclose the specific threshold maintenance algorithm of claims 4-11 and 39-41, which involves tracking "current minimum" and "future minimum" values. However, Martin teaches a nearly identical noise floor estimation process that tracks the minimum signal power over a predetermined period using corresponding "current minimum" (Pn(i)) and "future minimum" (PMmin) values. Martin's algorithm periodically sets the current minimum based on the future minimum tracked during the preceding window. Petitioner also argued that Martin discloses using adaptive microphone arrays, as required by claims 25 and 46.
  • Motivation to Combine: A Person of Ordinary Skill in the Art (POSITA) would combine Hirsch with Martin to improve Hirsch's performance in non-stationary noise environments. Hirsch itself acknowledged that speech pause detection is difficult in such environments and cited Martin as a known solution. Martin's algorithm was specifically designed to track non-stationary noise without needing an explicit speech detector, providing a clear motivation to integrate its superior noise floor tracking into Hirsch's adaptive threshold framework.
  • Expectation of Success: A POSITA would have had a reasonable expectation of success because both references operate in the same field of spectral subtraction for noise reduction. Integrating Martin's well-defined algorithm for noise floor estimation into Hirsch's analogous threshold calculation would have been a straightforward modification to achieve a predictable improvement.

Ground 3: Obviousness of Claims 13-14, 17-21, 23, and 47 over Hirsch in view of Boll

• Prior Art Relied Upon: Hirsch (1995 publication) and Boll (Steven F. Boll, “Suppression of Acoustic Noise in Speech Using Spectral Subtraction,” 1979).

• Core Argument for this Ground:

  • Prior Art Mapping: Petitioner argued that while Hirsch discloses a complete spectral subtraction system, it does not specify certain implementation details that are recited in the challenged dependent claims. Boll, a foundational reference in the field (cited by the ’345 patent itself), provides these missing details. Specifically, Boll teaches performing the subtraction step using a "filter multiplication" (as required by claim 14) and implementing a "residual noise processor" to reduce remaining noise after the primary subtraction (as required by claims 17 and 47). Boll also teaches using a "speech activity detector" (voice switch) and a second threshold for its residual noise processing, mapping to claims 19 and 20.
  • Motivation to Combine: A POSITA implementing Hirsch's high-level noise estimation scheme would have naturally looked to a foundational and well-known reference like Boll for guidance on conventional aspects of the spectral subtraction process. Hirsch explicitly states its techniques "can be combined with well known spectral subtraction techniques." Since Boll was the seminal work describing such techniques, it would have been an obvious reference to consult for standard implementation methods like filter multiplication and residual noise reduction.
  • Expectation of Success: The combination would have been a simple integration of a conventional, well-understood technique (from Boll) into a system (from Hirsch) designed to accommodate such techniques, leading to a predictable result.

• Additional Grounds: Petitioner asserted additional obviousness challenges, including combining Hirsch, Martin, and Boll for claim 43; Hirsch, Boll, and Arslan (Patent 5,706,395) for claims 15-16 and 24; Hirsch and Uesugi (Patent 5,459,683) for claim 22; and Hirsch, Martin, and Uesugi for claims 44-45. These grounds relied on similar motivations to combine, using Arslan to add teachings of a Wiener filter and two-dimensional smoothing, and Uesugi to add a computationally efficient method for estimating signal magnitude.

4. Key Claim Construction Positions

• "magnitude": Petitioner proposed that the broadest reasonable interpretation includes both the signal's actual magnitude and an approximation of its magnitude. This construction is based on the specification's disclosure of estimating magnitude "using an approximation formula" and is important because some prior art references calculate signal power (the square of magnitude) or use other approximations.
• "frequency spectrum generator" / "generating the frequency spectrum": Petitioner argued these terms should be construed to include hardware or software that converts an audio signal from the time domain to the frequency domain, such as by performing a Fast Fourier Transform (FFT). This construction ensures that standard FFT-based systems taught in the prior art fall within the scope of the claims.

5. Relief Requested

• Petitioner requested the institution of an inter partes review and the cancellation of claims 1-25 and 38-47 of the ’345 patent as unpatentable.