DETAILED ACTION
Continued Examination Under 37 CFR 1.114
A request for continued examination under 37 CFR 1.114, including the fee set forth in 37 CFR 1.17(e), was filed in this application after final rejection. Since this application is eligible for continued examination under 37 CFR 1.114, and the fee set forth in 37 CFR 1.17(e) has been timely paid, the finality of the previous Office action has been withdrawn pursuant to 37 CFR 1.114. Applicant's submission filed on 12/10/2025 has been entered.
This communication is in response to the Amendments and Arguments filed on 11/10/2025.
Claims 1, 3-12, and 14-22 are pending and have been examined.
All previous objections/rejections not mentioned in this Office Action have been withdrawn by the examiner.
Notice of Pre-AIA or AIA Status
The present application is being examined under the pre-AIA first to invent provisions.
Response to Arguments
Applicant's arguments filed 11/10/2025 regarding the 101 rejection have been fully considered but they are not persuasive.
Applicant asserts on pgs 9-10 that the features of input and output notes, non-reduction and reduction cells connected between the input and output nodes, and the context-dependent operations of the non-reduction cells being based on past speech data and being independent of future speech data and including a convolution or pooling operation, reflect an improvement in the field of speech recognition. The Examiner respectfully disagrees with this assertion. The use of specific nodes and cells reads on a human using particular operations when putting together sets of equations for processing the data in a particular manner, and the use of past speech data and being independent of future speech data reads on the human only using particular portions of the data when performing calculations using the specified operations. There is no additional language present in the claims that clearly identifies how or why the process, as recited, provides a technological improvement to speech recognition.
Hence, Applicant’s arguments are not persuasive.
Applicant’s arguments with respect to claim(s) 1, 10, and 12, have been considered but are moot because the new ground of rejection does not rely on any reference applied in the prior rejection of record for any teaching or matter specifically challenged in the argument. Please see the updated mappings below citing Baruwa, Ding, and Yeh, for further detail.
Claim Rejections - 35 USC § 101
35 U.S.C. 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
Claims 1, 3-12, and 14-22 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Regarding claim(s) 1, 10, and 12, the limitation(s) of (claim 10) acquiring speech training sample information, obtaining a speech recognition model, receiving streaming speech data, processing the streaming speech data, and outputting the speech recognition text, as drafted, are processes that, under broadest reasonable interpretation, covers performance of the limitation in the mind and/or with pen and paper but for the recitation of generic computer components. More specifically, the mental process of a human having access to a person who will speak out loud and hold up a sign showing the text associated with what they are saying, constructing a set of specific operations to be performed in a particular order using specific sections of information to transcribe speech into text, hearing someone talking, considering the speech using the rules in order to determine the corresponding words as text, and writing down the corresponding words as text on a piece of paper. The neural architecture search reads to a set of rules for determining which elements of a model are beneficial, and the initial network and speech recognition model read to respective sets of steps/rules/operations understood by a human to follow in order to perform a specific task. If a claim limitation, under its broadest reasonable interpretation, covers performance of the limitation in the mind and/or with pen and paper but for the recitation of generic computer components, then it falls within the --Mental Processes-- grouping of abstract ideas. Accordingly, the claim(s) recite(s) an abstract idea.
This judicial exception is not integrated into a practical application because the recitation of processing circuitry of claims 1 and 10, and an apparatus and processing circuitry of claim 12 reads to generalized computer components, based upon the claim interpretation wherein the structure is interpreted using [0165-174] in the specification. Accordingly, these additional elements do not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea. The claim(s) is/are directed to an abstract idea.
The claim(s) do(es) not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to the integration of the abstract idea into a practical application, the additional element of using generalized computer components to acquire, obtain, receive, process, and output, amounts to no more than mere instructions to apply the exception using a generic computer component. Mere instructions to apply an exception using a generic computer component cannot provide an inventive concept. The claim(s) is/are not patent eligible.
With respect to claim(s) 3 and 14, the claim(s) recite(s) the connection of the unit networks, which reads on a human using sets of steps that are related in a particular way. No additional limitations are present.
With respect to claim(s) 4, 15, and 21, the claim(s) recite(s) reduction cells include an operation dependent on future speech data, which reads on a human using specific sections of received information. No additional limitations are present.
With respect to claim(s) 5, 16, and 22, the claim(s) recite(s) the two non-reduction cells and reduction cells having particular features, which reads on a human evaluating different sets of steps that have specific actions to follow. No additional limitations are present.
With respect to claim(s) 6, 8, 17, and 19, the claim(s) recite(s) the operation element including a specific type of operation, which reads on a human performing specific actions when encountering the associated step in the set of rules to follow. The recitation of an LSTM and GRU in claims 8 and 19 read to specific steps.
With respect to claim(s) 7 and 18, the claim(s) recite(s) nodes performing a specific operation, which reads on a human performing specific actions when encountering the associated step in the set of rules to follow. No additional limitations are present.
With respect to claim(s) 9, 11, and 20, the claim(s) recite(s) performing NAS via a speech training sample to obtain a speech recognition model with specific features, processing speech data to obtain acoustic recognition information, and processing the acoustic recognition information to obtain speech recognition text, which reads on a human using training information to evaluate different sets of rules, choosing a set of rules that has a specific sequence of steps, using the first part of the set of rules to evaluate the speech of another human to write down a specific set of information, and using the second part of the set of rules to evaluate the information and write down the text associated with the heard speech. The acoustic model, decoding graph, network search model, and initial network, all read to different sets of rules/steps for a human to follow to produce a desired result.
These claims further do not remedy the judicial exception being integrated into a practical application and further fail to include additional elements that are sufficient to amount to significantly more than the judicial exception.
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
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, 3-12, and 14-22 is/are rejected under 35 U.S.C. 103 as being unpatentable over Baruwa et al. (“Leveraging End-to-End Speech Recognition with Neural Architecture Search”, arXiv:1912.05946v1, 11 Dec 2019), hereinafter Baruwa, in view of Ding et al. (“AutoSpeech: Neural Architecture Search for Speaker Recognition”, arXiv:2005.03215v2, 31 Aug 2020), hereinafter Ding, and further in view of Yeh et al. (“TRANSFORMER-TRANSDUCER: END-TO-END SPEECH RECOGNITION WITH SELF-ATTENTION”, arXiv:1910.12977v1, 28 Oct 2019), hereinafter Yeh.
Regarding claims 1, 10, and 12, Baruwa teaches
(claim 1) A computer-implemented method for streaming speech recognition (an NAS approach to finding hyper-parameters for a neural architecture for ASR (Abstract and Intro)), comprising:
(claim 10) A speech recognition method comprising (an NAS approach to finding hyper-parameters for a neural architecture for ASR (Abstract and Intro)):
(claim 12) A speech recognition apparatus, the apparatus comprising: processing circuitry configured to…(see Intro)
(claim 10) acquiring, by processing circuitry, speech training sample information, the speech training sample information including speech training sample data and a speech recognition tag corresponding to the speech training sample data (the LibriSpeech and TIMIT corpora are used as training sets, i.e. acquiring by the processing circuitry speech training sample information, where the TIMIT corpus includes phonetic speech data, i.e. including speech training sample data, and acoustic phonetic labels used for transcribing the audio data, where a model scores how a speech frame matches its own label, i.e. a speech recognition tag corresponding to the speech training sample data (Abstract, Intro, Sec. 3, Sec. 4.1)),
((claim 10) performing, by the processing circuitry, a neural architecture search on an initial network using the speech training sample information to) obtaining, by processing circuitry, a speech recognition model constructed based on a neural architecture search performed on an initial network, the initial network including an input node, an output node, --and other set operations--, –the other set operations-- being connected between the input node and the output node (a controller RNN samples child networks with a limited search space of specific operations and blocks, i.e. based on a neural architecture search performed on an initial network, and using the training corpora, i.e. using the speech training sample information, to determine the choice of parameters for each child network and find the best architecture for a speech recognition model, i.e. obtaining by processing circuitry a speech recognition model constructed based on a neural architecture search, where the architecture begins with an input block, ends with a softmax block, and has a series of other blocks in between, i.e. including an input node an output node and other set operations the other set operations being connected between the input node and the output node Fig. 3.1,(Abstract, Intro, Sec. 2.4-3.1)), the speech recognition model being configured based on –the other set operations-- connected between the input node and the output node (a controller RNN samples child networks with a limited search space of specific operations and blocks, i.e. other set operations, to determine the choice of parameters for each child network and find the best architecture for a speech recognition model, i.e. the speech recognition model being configured based on the other set operations, where the architecture begins with an input block, ends with a softmax block, and has a series of other blocks in between, i.e. including an input node an output node and other set operations the other set operations connected between the input node and the output node Fig. 3.1,(Abstract, Intro, Sec. 2.4-3.1)), and
(claims 1 and 12) receiving, by the processing circuitry, streaming speech data (audio data is processed in frames, i.e. receiving streaming speech data, to evaluate the models (Abstract, Intro, Sec. 4));
processing, by the processing circuitry, the streaming speech data using the speech recognition model to generate recognition outputs at each time step … (audio data is processed in frames of 10ms, i.e. processing by the processing circuitry the streaming speech data, by the model to provide a transcription of the audio data, i.e. using the speech recognition model to generate recognition outputs at each time step (Abstract, Intro, Sec. 4)); and
(claims 1 and 12) outputting speech recognition text corresponding to the processed streaming speech data (audio data is processed in frames of 10ms by the model to provide a transcription of the audio data, i.e. using the speech recognition model to generate recognition outputs at each time step (Abstract, Intro, Sec. 4)),
wherein the … operations … include a convolution operation or a pooling operation (the neural architecture includes convolution and maxpool operations Fig. 3.1,(Sec. 3.1)).
While Baruwa provides performing a neural architecture search with specific parameters for child models, Baruwa does not specifically teach reduction and non-reduction cells, and thus does not teach
the initial network including an input node, an output node, one or more non-reduction cells, and one or more reduction cells, the one or more non-reduction cells and the one or more reduction cells being connected between the input node and the output node,
the speech recognition model being configured based on at least a portion of the one or more non-reduction cells and at least a portion of the one or more reduction cells connected between the input node and the output node,
wherein the context-dependent operations within each of the one or more non-reduction cells include a convolution operation or a pooling operation.
Ding, however, teaches the initial network including an input node, an output node, one or more non-reduction cells, and one or more reduction cells, the one or more non-reduction cells and the one or more reduction cells being connected between the input node and the output node (the NAS search space, i.e. initial network, includes neural cells that are stacked to construct a CNN, where a cell can be either a normal cell or a reduction cell, i.e. one or more non-reduction cells and one or more reduction cells, where the output of the last cell is fed to an average pooling layer followed by a softmax layer that provides an output, i.e. output node, and each cell has an input node, i.e. an input node…the one or more non-reduction cells and the one or more reduction cells being connected between the input node and the output node (Sec. 3)),
the speech recognition model being configured based on at least a portion of the one or more non-reduction cells and at least a portion of the one or more reduction cells connected between the input node and the output node (neural cells that are stacked to construct a CNN, i.e. speech recognition model being configured based on, where a cell can be either a normal cell or a reduction cell, i.e. one or more non-reduction cells and one or more reduction cells, where the output of the last cell is fed to an average pooling layer followed by a softmax layer that provides an output, i.e. output node, and each cell has an input node, i.e. an input node…the one or more non-reduction cells and the one or more reduction cells being connected between the input node and the output node (Sec. 3)),
wherein the context-dependent operations within each of the one or more non-reduction cells include a convolution operation or a pooling operation (normal cells, i.e. non-reduction cells, can include operations such as convolution , average pooling, and max pooling, i.e. context-dependent operations within each of the one or more non-reduction cells include a convolution operation or a pooling operation (Sec. 3.1)).
Where Baruwa specifically teaches that the model is for speech recognition (Abstract, Intro).
Baruwa and Ding are analogous art because they are from a similar field of endeavor in developing speech processing models using NAS. Thus, 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 performing a neural architecture search with specific parameters for child models teachings of Baruwa with the use of stacked reduction and non-reduction cells as taught by Ding. It would have been obvious to combine the references to enable improved speaker recognition with lower model complexity (Ding (Abstract)).
While Baruwa in view of Ding provides processing frames of audio data, Baruwa in view of Ding does not specifically teach being independent of future speech data and using only corresponding portions of streaming speech data up to the respective time step, and thus does not teach
all context-dependent operations within each of the one or more non-reduction cells being based on past speech data and being independent of future speech data;
(claims 1 and 12) processing…the streaming speech data…using only corresponding portions of the streaming speech data up to the respective time step.
Yeh, however, teaches all context-dependent operations within each of the one or more non-reduction cells being based on past speech data and being independent of future speech data (to prevent future information from leaking into the computation at the current time step, i.e. being independent of future speech data, causal convolution, i.e. operations within each of the one or more non-reduction cells, is used in which all contexts required are pushed to the history, i.e. all context-dependent operations…being based on past speech data (Sec. 4.1));
(claim 10) Where Baruwa teaches the speech data is training data. (see Abstract, Intro, Sec. 3, Sec. 4.1)
(claims 1 and 12) processing…the streaming speech data…using only corresponding portions of the streaming speech data up to the respective time step (to prevent future information from leaking into the computation at the current time step, i.e. up to the respective time step, causal convolution is used in which all contexts required are pushed to the history, i.e. processing…the streaming speech data…using only corresponding portions of the streaming speech data up to the respective time step (Abstract, Intro, Sec. 4.1, Sec. 5.1)).
Baruwa, Ding, and Yeh are analogous art because they are from a similar field of endeavor in developing speech processing models. Thus, 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 processing frames of audio data teachings of Baruwa, as modified by Ding, with the use of causal convolution as taught by Yeh. It would have been obvious to combine the references to incorporate positional information and reduce frame rate for efficient inference (Yeh (Abstract)).
Regarding claims 3 and 14, Baruwa in view of Ding and Yeh teaches claims 1 and 12, and Ding further teaches
the one or more non-reduction cells and the one or more reduction cells are connected based on at least one of: a bi-chain-styled connection mode, a chain-styled connection mode (the neural cells are stacked 8 times to form the backbone CNN, where the cells are a mix of normal and reduction cells (Sec. 3.1)), or a densely-connected connection mode (the neural cells are stacked 8 times to form the backbone CNN, where the cells are a mix of normal and reduction cells, and the input nodes are set equal to the outputs of previous two cells (Sec. 3.1)).
Where the motivation to combine is the same as previously presented.
Regarding claims 4, 15, and 21, Baruwa in view of Ding and Yeh teaches claims 1, 12, and 10, and Baruwa further teaches
the one or more reduction cells include at least one operation dependent on the future speech data (the search space involves convolution, maxpool, batch normalization, and recurrent blocks, where a BiLSTM block learns information in both forward and backward directions, i.e. at least one operation dependent on the future speech data (Sec. 3.1 and 3.3)).
Where Ding teaches that both normal and reduction cells are determined based on searching the same set of operations (Sec. 3.1).
And where the motivation to combine is the same as previously presented.
Regarding claims 5, 16, and 22, Baruwa in view of Ding and Yeh teaches claims 4, 15, and 21, and Ding further teaches
at least one of the one or more non-reduction cells shares a topology and network parameters with at least another one of the one or more non-reduction cells (all the normal cells share the same architecture and parameters (Sec. 3.1-3.2)), and
at least one of the one or more reduction cells shares a topology and network parameters with at least another one of the one or more reduction cells (all the reduction cells share the same architecture and parameters (Sec. 3.1-3.2)).
Where the motivation to combine is the same as previously presented.
Regarding claims 6 and 17, Baruwa in view of Ding and Yeh teaches claims 1 and 12, and Yeh further teaches
the context-dependent operations independent of the future speech data include a causality-based operation or a mask-based operation (to prevent future information from leaking into the computation at the current time step, i.e. independent of future speech data, causal convolution is used in which all contexts required are pushed to the history and ensuring the convolution is purely causal, i.e. the context-dependent operations…include a causality-based operation (Sec. 4.1)).
Where the motivation to combine is the same as previously presented.
Regarding claims 7 and 18, Baruwa in view of Ding and Yeh teaches claims 1 and 12, and Ding further teaches
the one or more non-reduction cells include nodes configured to perform at least one of a summation operation, a concatenation operation, or a product operation (the output node of each cell, i.e. the one or more non-reduction cells include nodes, performs concatenation of all intermediate nodes, i.e. configured to perform at least one of…a concatenation operation Fig. 1,(Sec. 3.1)).
Where the motivation to combine is the same as previously presented.
Regarding claims 8 and 19, Baruwa in view of Ding and Yeh teaches claims 1 and 12, and Baruwa further teaches
the context-dependent operations within each of the one or more non-reduction cells correspond to an operation based on a long short-term memory (LSTM) or an operation based on a gated recurrent unit (GRU) (the search space involves convolution, maxpool, batch normalization, and recurrent blocks, where the recurrent block is a BiLSTM block, i.e. an operation based on a long short-term memory (LSTM) (Sec. 3.1 and 3.3)).
Where Ding teaches that both normal and reduction cells are determined based on searching the same set of operations (Sec. 3.1).
And where the motivation to combine is the same as previously presented.
Regarding claims 9, 11, and 20, Baruwa, Ding, and Yeh teaches claims 1 and 12, and Baruwa further teaches
the speech recognition model includes an acoustic model and a decoding graph, the acoustic model being constructed based on a network search model obtained by performing the neural architecture search on the initial network using a speech training sample (NAS is used to determine a child neural architecture for a speech recognition model, i.e. the speech recognition model includes an acoustic model, where the search space is limited to specific operations/blocks and using the libriSpeech and TIMIT training corpora to determine the respective best child networks, i.e. the acoustic model being constructed based on a network search model obtained by performing the neural architecture search on the initial network using a speech training sample, where the inference to determine a transcript includes a beam search decoding, i.e. the speech recognition model includes a decoding graph (Abstract, Intro, Sec. 3 Intro-3.2)), and
(claim 10) the speech recognition tag includes acoustic recognition information of the speech training sample data, the acoustic recognition information including a phoneme, a syllable, or a semi-syllable (the LibriSpeech and TIMIT corpora are used as training sets, where the TIMIT corpus includes phonetic speech data, i.e. speech training sample data, and acoustic phonetic labels used for transcribing the audio data, i.e. speech recognition tag includes acoustic recognition information of the speech training sample data, where a model scores how a speech frame matches its own label, and the outputs can be character-level, i.e. acoustic recognition information including…a semi-syllable (Abstract, Intro, Sec. 3, Sec. 4.1)),
the processing the streaming speech data includes:
processing the streaming speech data using the acoustic model to obtain acoustic recognition information that includes a phoneme, a syllable, or a semi-syllable (audio data is processed in frames of 10ms by the child model, i.e. processing the streaming speech data using the acoustic model, to determine phonetic information used in transcribing the audio data, such as sets of characters, i.e. to obtain acoustic recognition information that includes a semi-syllable (Abstract, Intro, Sec. 3 Intro-3.2, Sec. 4)); and
processing the acoustic recognition information using the decoding graph to obtain the speech recognition text (character-level beam search decoding is performed to evaluate the candidate transcript sequence of the child network, i.e. processing the acoustic recognition information using the decoding graph, to improve the performance, where the final output is a transcript, i.e. obtain the speech recognition text (Abstract, Intro, Sec. 3 Intro-3.2, Sec. 4)).
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to NICOLE A K SCHMIEDER whose telephone number is (571)270-1474. The examiner can normally be reached 8:00 - 5:00 M-F.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Pierre-Louis Desir can be reached at (571) 272-7799. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/NICOLE A K SCHMIEDER/Primary Examiner, Art Unit 2659