Ex parte Hannun

Informative decision only: persuasive authority, not binding precedent. Applying the 2019 Revised Patent Subject Matter Eligibility Guidance, when a claim recites a specific implementation of speech transcription (normalizing, jitter set, spectrogram frames, neural network character probabilities, language-model decoding) but does not itself recite the mathematical algorithm, the Board will find that no mathematical concept, mental process or method of organizing human activity is recited at Step 2A prong one. Even if a mathematical concept were recited, a practical application that improves the technical field (speech recognition) means the claim is not directed to an abstract idea, and an examiner's conclusory Step 2B finding lacks the factual support needed. No prior precedent directly addressed this issue in the text; the Board applies the January 7, 2019 Memorandum (84 Fed. Reg. 50), which separates mental processes from methods of organizing human activity and requires that a mathematical concept actually be recited. For an examiner, the decision means that identifying a mathematical formula only in the specification is not enough, and a general assertion that a human could transcribe audio does not make steps like jitter-set generation and neural-network probability output a mental process. The examiner also needs factual support for any Step 2B finding of no significantly more. For appellants, the decision shows that arguing a specific implementation and a technical improvement can defeat an abstract-idea rejection on computer-implemented machine learning claims, and that an unrecited algorithm is not a recited mathematical concept. The Board relied on Alice, Mayo, Diehr, Benson, Flook and Bilski for the two-step framework, and on Berkheimer v. HP Inc., 881 F.3d 1360 (Fed. Cir. 2018), for the need for factual support at Step 2B. LEGAL TEST OR STANDARD: Under the Memorandum, the Board first looks to whether the claim recites (1) any judicial exceptions, including certain groupings of abstract ideas (i.e., mathematical concepts, certain methods of organizing human activity such as a fundamental economic practice, or mental processes); and (2) additional elements that integrate the judicial exception into a practical application. Only if a claim (1) recites a judicial exception and (2) does not integrate it into a practical application does the Board look to whether the claim (3) adds a specific limitation beyond the judicial exception that is not well-understood, routine, conventional in the field; or (4) simply appends well-understood, routine, conventional activities previously known to the industry, specified at a high level of generality, to the judicial exception. TRIGGERING FACTS: Claims 11-20 recited a computer-implemented method of transcribing speech: normalizing input audio, generating a jitter set of audio files, generating spectrogram frames, inputting them to a trained neural network, obtaining predicted character probabilities, and decoding with a language model. The Examiner characterized this as a mathematical formula and as human activity or mental work, but the algorithm for the probabilities appeared only in the specification and not in the claims. Appellants argued a specific improvement in automatic speech recognition, and the specification described higher performance than traditional methods. The Examiner's Step 2B finding was conclusory. KEY QUOTE: "The mathematical algorithm or formula, however, is not recited in the claims." (Paper 1, p. 10) PRIOR PRECEDENT AFFECTED: No prior precedential or informative decision is directly affected. The Board applies the January 7, 2019 Memorandum, 2019 Revised Patent Subject Matter Eligibility Guidance, 84 Fed. Reg. 50, and its Subject Matter Eligibility Examples: Abstract Ideas (Example 38). It also cites Berkheimer v. HP Inc., 881 F.3d 1360 (Fed. Cir. 2018). PRACTICE TIP: In an appeal brief, show that the claims recite no mathematical formula, calculation or mental step, and tie the recited steps to a specific technological improvement documented in the specification, such as the performance gains over traditional speech recognition. Also challenge any Step 2B finding that lacks factual support.