Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Response to Amendment
Applicant's amendment filed on December 2, 2025 has been entered. No claims have been amended. Claims 16 – 20 have been canceled. No claims have been added. Claims 1 – 15 and 21 – 25 are still pending in this application, with claims 1, 8 and 21 being independent.
Response to Arguments
Applicant's arguments filed December 2, 2025 have been fully considered but they are not persuasive.
The Applicant alleges: “
Applicant respectfully asserts that the cited references, taken alone or in
hypothetical combination, at least fail to teach or suggest "applying a skip-connection emission from a final unit frame to an end frame of a plurality of frames," as recited in claim 1. Accordingly, Applicant respectfully requests withdrawal of the rejection of claim 1.
As discussed during the Examiner Interview, Applicant respectfully submits that the cited references, alone or in hypothetical combination, do not teach or suggest a "skip-connection emission," as recited in claim 1. For example, the Office Action contends that Tian, with reference to section 2.2, teaches the claimed "skip connection emission." See Office Action, 3. However, the cited portion of Tian teaches an attempt to "force[] the transducer model to align the predicted blank position with the blank position of the CTC model." See Tian, 2.2. The fast- skip regularization of Tianis is not a "skip-connection emission from a final unit frame to an end frame of a plurality of frames," and instead, is intended to predict blank tokens within a sequence. Applicant respectfully submits that the prediction of a blank token cannot reasonably be interpreted as a "skip-connection emission" that will cause a skip from "a final unit frame to an end frame of a plurality of frames." Rather, Tian's Fast-Skip Regularization (FSR) is a loss function term used to penalize delays in blank predictions during training. Tian does not disclose a structural lattice emission that connects a final unit frame directly to an end frame. In Tian, the lattice topology remains sequential; the model still processes intermediate frames to determine if they are blanks. See Tian, FIG. 2(a). Moreover, Fainberg is not contended to teach this feature of claim 1. Accordingly, Applicant respectfully requests withdrawal of the rejection of claim 1.”
Examiner’s response:
The Examiner respectfully disagrees.
Tian discloses, in abstract, that:
we propose a method named fast-skip regularization, which tries to align the blank position predicted by a transducer with that predicted by a CTC model.
During the inference, the transducer model can predict the blank tokens in advance by a simple CTC project layer without many complicated forward calculations of the transducer decoder and then skip them, which will reduce the computation and improve the inference speed greatly. All experiments are conducted on a public Chinese mandarin dataset AISHELL-1. The results show
that the fast-skip regularization can indeed help the transducer model learn the blank position alignments. Besides, the inference with fast-skip can be speeded up nearly 4 times with only a little performance degradation.
in section 1:
Furthermore, we find the predicted blank tokens account for over 90% of all
the tokens. Based on this observation, we assume that the inference process can be speeded up by predicting the blank tokens in advance with some simple computation and then skipping them. In this paper, we propose a novel method named fast-skip regularization (FSR) to make the transducer model depend
on a CTC project layer to predict the blank tokens in advance. The FSR introduces a CTC module into the transducer model. And it forces the transducer model to align the blank position with that predicted by the CTC project layer. During the inference, the transducer model can skip the blank token based on
the prediction of the CTC project layer, which makes the output probability lattice be sparse and improve the inference speed greatly. All the experiments are conducted on a public Chinese mandarin dataset AISHELL-1. The results show that the FSR can align the blank position predicted by the CTC and transducer
accurately. The inference process can be accelerated up to 4 times with only a little performance degradation.
(emphasis added)
That means, Tian discloses, in section 2 and section 2.2, to determine blank tokens to skip them. Fig. 2 also shows FSR from a final unit frame to an end frame of a plurality of frames.
Thus, Tian teaches “applying a skip-connection emission from a final unit frame to an end frame of a plurality of frames.”
The Applicant further alleges: “
The Office Action also contends that Fainberg, with reference to page 2, section 4, teaches the claimed "zero-value cost" recited in claim 1. See Office Action at 3. Applicant respectfully disagrees at least because Fainberg describes initialization of an edit transducer and not skip-connections through a lattice. See Fainberg, Section 4. Fainberg's disclosure of setting "edit costs to 0" relates to initializing an edit-distance transducer for aligning transcripts (i.e., the penalty for an insertion or deletion operation is zero). This is fundamentally different from the claimed "skip-connection emission" having a "zero-value cost." The claimed cost regards a path in the lattice representation, representing the probability (or log-probability) of taking that skip action. Fainberg's editorial penalties are not lattice transition costs. For example, Fainberg describes an "edit transducer" that enables deletions and substitutions to combine lattices with inaccurate transcripts. See id. The edit transducer, however, does not have a zero cost. Instead, matches have a cost of -1, in an attempt to obtain a "most negative total cost." See id. Accordingly, Applicant respectfully submits that Fainberg cannot reasonably be interpreted as teaching that the claimed "skip-connection emission has a zero-value cost," at least because Fainberg does apply a cost to its edit transducer, and moreover, Fainberg does not include a skip- connection emission. For at least this reason, among others, Applicant respectfully requests withdrawal of the rejection of claim 1.”
Examiner’s response:
The Examiner again respectfully disagrees.
Fainberg discloses, in section 4, that the cost for matched nodes in lattice is set to -1, and the cost for rest nodes (e.g. blank nodes) is set to 0. Thus, the path is selected to have the most negative total cost (skip-connection). Thus, Fainberg teaches the claimed "zero-value cost".
The Applicant still further alleges: “
Furthermore, Applicant respectfully submits that the Office Action has not
provided an articulated rationale for combining Tian and Fainberg. The proposed motivation advanced in the Office Action is "to take an advantage of the path selected retain [sic] with minimum cost." Office Action at 4. Applicant respectfully submits that "rejections on obviousness grounds cannot be sustained by mere conclusory statements; instead, there must be some articulated reasoning with some rational underpinning to support the legal conclusion of obviousness." See KSR, 550 U.S. at 418. The Office Action does not sufficiently explain why the alleged reason for combining the references would lead a person of ordinary skill in the art to make the specific proposed modification. A general motivation to improve or reduce a path length cost cannot sustain a rejection under § 103. Accordingly, Applicant respectfully requests withdrawal of the rejection of claim 1.”
Examiner’s response:
One of ordinary skill in the art will all know that the path selected retain with minimum cost would optimize the path for a fast and efficient performance.
Therefore, claim 1 is unpatentable over Tian in view of Fainberg (similar as to claims 8 and 21).
Regarding to the rest of claims, the Applicant does not argue about the rest of claims.
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claim(s) 1, 5 – 15 and 21 – 25 is/are rejected under 35 U.S.C. 103 as being unpatentable over Tian et al. (“FSR: Accelerating the Inference Process of Transducer-Based Models by Applying Fast-Skip Regularization”, arXiv:2104.02882v1), hereinafter referred as Tian, in view of Fainberg et al. (“Lattice-based lightly-supervised acoustic model training”, arXiv:1905.13150v2), hereinafter referred as Fainberg.
Regarding claim 1, Tian discloses a method (abstract), comprising:
generating a lattice representation corresponding to an auditory utterance (Fig. 2);
traversing the lattice representation based, at least, on log probabilities of emissions for units of the auditory utterance (section 2.1. Transducer);
determining a final unit of the auditory utterance is emitted (pages 2 – 3, last frame before blank token);
applying a skip-connection emission from a final unit frame to an end frame of a plurality of frames (section 2.2. Fast-skip Regularization); and
determining a cost of a path through the lattice representation (section 2.2. Fast-skip Regularization, transducer loss).
However, Tian fail to explicitly disclose the method wherein the skip-connection emission has a zero-value cost.
However, in a similar field of endeavor Fainberg discloses a method for lattice-based lightly-supervised acoustic model training (abstract). In addition, Fainberg discloses the method wherein the skip-connection emission has a zero-value cost (page 2, section 4. Lattice combination, “we set every edit cost to 0”).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the invention of Tian, and the skip-connection emission has a zero-value cost. The motivation for doing this is to take an advantage of the path selected retaining with minimum cost so that to optimize the path for a fast and efficient performance.
Regarding claim 5 (depends on claim 1), Tian discloses the method further comprising: adding a blank emission to the path; and adding an associated blank cost to the cost (section 2.2, equation (7), the third term talked about blank position and associated blank cost).
Regarding claim 6 (depends on claim 1), Tian discloses the method further comprising: applying the skip-connection emission from a start frame to an intermediate frame between the start frame and the end frame (section 2).
Regarding claim 7 (depends on claim 1), Tian discloses the method wherein the skip-connection emission traverses a time axis of the lattice representation for two or more frames (Fig. 2, horizontal axis (1 .. 11) is time axis; also see section 2.3).
Regarding claim 8, Tian discloses a system (Fig. 1) comprising:
at least one processor (abstract, transducer-based models) to:
generate a lattice representation corresponding to an auditory utterance over a plurality of frames (Fig. 2);
apply a skip-connection emission from a first frame of the plurality of frames to a selected second frame of the plurality of frames (section 2.2. Fast-skip Regularization);
traverse the lattice representation based, at least, on log probabilities of emissions for units of the auditory utterance (section 2.1. Transducer);
determine a final unit of the auditory utterance is emitted (pages 2 – 3, last frame before blank token); and
determine a cost of a path through the lattice representation (section 2.2. Fast-skip Regularization, transducer loss).
However, Tian fail to explicitly disclose the method wherein the skip-connection emission has a zero-value cost.
However, in a similar field of endeavor Fainberg discloses a method for lattice-based lightly-supervised acoustic model training (abstract). In addition, Fainberg discloses the method wherein the skip-connection emission has a zero-value cost (page 2, section 4. Lattice combination, “we set every edit cost to 0”).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the invention of Tian, and the skip-connection emission has a zero-value cost. The motivation for doing this is to take an advantage of the path selected retaining with minimum cost so that to optimize the path for a fast and efficient performance.
Regarding claim 9 (depends on claim 8), Tian discloses the system wherein the system comprises at least one of: a system for performing simulation operations; a system for performing simulation operations to test or validate autonomous machine applications; a system for performing digital twin operations; a system for performing light transport simulation; a system for rendering graphical output; a system for performing deep learning operations; a system implemented using an edge device; a system for generating or presenting virtual reality (VR) content; a system for generating or presenting augmented reality (AR) content; a system for generating or presenting mixed reality (MR) content; a system incorporating one or more Virtual Machines (VMs);a system for performing operations for a conversational AI application; a system for performing operations for a generative AI application; a system for performing operations using a language model; a system implemented at least partially in a data center (section 4, “Chinese mandarin dataset AISHELL-1”); a system for performing hardware testing using simulation; a system for synthetic data generation; a collaborative content creation platform for 3D assets; or a system implemented at least partially using cloud computing resources.
Regarding claim 10 (depends on claim 8), Tian discloses the system wherein the lattice representation is associated with a recurrent neural network transducer (RNN-T) (section 1. Introduction).
Regarding claim 11 (depends on claim 8), Tian discloses the system wherein the determining the cost of the path further causes the at least one processor to: determine forward weights for a path from a start to an end (section 2.1. Transducer; forward variable); determine backward weights for the path (section 2.1. Transducer; backward variable); and determine a path cost based at least on one or more loss functions (section 2.2. Fast-skip Regularization, transducer loss; equation (7 - 9)).
Regarding claim 12 (depends on claim 8), Tian discloses the system wherein the at least one processor is further to: apply the skip-connection emission from the final unit to a near-last frame (section 2.1. Transducer, and Fig. 2).
Regarding claim 13 (depends on claim 12), Tian discloses the system wherein the near-last frame is an immediately preceding frame of final output frame (section 2.1. Transducer, and Fig. 2).
Regarding claim 14 (depends on claim 12), Tian discloses the system wherein the at least one processor is further to: apply a blank emission after the near-last frame; and add a blank cost, for the blank emission, to the cost (section 2.1. Transducer, and Fig. 2).
Regarding claim 15 (depends on claim 8), Tian discloses the system wherein the lattice representation includes a respective end state for each frame of the plurality of frames (section 2.1. Transducer, and Fig. 2, end state of last blank token).
Regarding claim 21, Tian discloses a processor (abstract, transducer-based models) comprising:
one or more processing units to determine a cost of a path through a lattice representation (section 2.2. Fast-skip Regularization, determine transducer loss) traversed skip-connection emission from a first frame of a plurality of frames to a selected second frame of the plurality of frames (section 2.1. Transducer, and Fig. 2).
However, Tian fail to explicitly disclose the system wherein the skip-connection emission using at least one zero-value cost.
However, in a similar field of endeavor Fainberg discloses a method for lattice-based lightly-supervised acoustic model training (abstract). In addition, Fainberg discloses the method wherein the skip-connection emission using at least one zero-value cost (page 2, section 4. Lattice combination, “we set every edit cost to 0”).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the invention of Tian, and the skip-connection emission using at least one zero-value cost. The motivation for doing this is to take an advantage of the path selected retaining with minimum cost so that to optimize the path for a fast and efficient performance.
Regarding claim 22 (depends on claim 21), Tian discloses the processor wherein the wherein the lattice representation is associated with a recurrent neural network transducer (RNN-T) (section 1. Introduction).
Regarding claim 23 (depends on claim 21), Tian discloses the processor wherein the one or more processing units determine the cost of the path, based at least in part, on forward weights from a start to an end (section 2.1. Transducer; forward variable); backward weights (section 2.1. Transducer; backward variable), and one or more loss functions (section 2.2. Fast-skip Regularization, transducer loss; equation (7 - 9)).
Regarding claim 24 (depends on claim 21), Tian discloses the processor wherein zero-value skip-connection emission is applied from a determined final frame of the plurality of frames (Section 2, Fig. 2, determines final frame at time axis 11).
Regarding claim 25 (depends on claim 21), Tian discloses the processor wherein the lattice representation includes a respective end state for each frame of the plurality of frames (section 2.1. Transducer, and Fig. 2, end state of last blank token).
Allowable Subject Matter
Claims 2 – 4 are objected to as being dependent upon a rejected base claim, but would be allowable if rewritten in independent form including all of the limitations of the base claim and any intervening claims.
Conclusion
THIS ACTION IS MADE FINAL. Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to QIAN YANG whose telephone number is (571)270-7239. The examiner can normally be reached on Monday-Thursday 8am-6pm.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Andrew Bee can be reached on 571-270-5183. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300. Information regarding the status of an application may be obtained from the Patent Application Information Retrieval (PAIR) system. Status information for published applications may be obtained from either Private PAIR or Public PAIR. Status information for unpublished applications is available through Private PAIR only. For more information about the PAIR system, see http://pair-direct.uspto.gov. Should you have questions on access to the Private PAIR system, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative or access to the automated information system, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/QIAN YANG/
Primary Examiner, Art Unit 2677