CTFR 17/979,471 CTFR 73899 Notice of Pre-AIA or AIA Status 07-03-aia AIA 15-10-aia The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA. Claim Rejections - 35 USC § 103 07-06 AIA 15-10-15 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. 07-20-aia AIA 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. 07-21-aia AIA Claim (s) 1,4-9,12-16 are rejected under 35 U.S.C. 103 as being unpatentable over Strimel (11887583) in view of Skarbovsky et al (20180143974) . Examiner notes, that the referred-to paragraphs in Strimel et al (11887583) are toward the text print of the patent (listed on the PTO-892 form, under NPL, dated 2/26/2025). As per claim 1, Strimel (11887583) teaches a method, executed in a computer system including a storage device and a processor of constructing an end-to-end speech recognition model (as, end-to-end, speech recognition – para 0094), capable of specializing a domain by using text data the method comprising: collecting, by the processor, domain text data (as, taking audio speech and transcribing into text – para 0091); the natural language unit determines phrases/statements from this text data – para 0092; of a domain that includes text that may appear in a domain to be specialized and comparing by the processor comparison candidate text (as, using general models or specific domain models, for the recognition process – para 0115; – examiner notes, that the speech is already converted to phonemes, then compared to ASR’s – both general and domain specific; the models are trained for each particular domain – para 0126); extracting, by the processor, a text from the collected domain text data, and constructing a domain text database (DB) through a normalization process for symbols, numbers, and foreign language notation in included in the extracted text; extracting, by the processor, a comparison candidate text from the constructed domain text DB ((as, updating/training the model for each domain – para 128-130, wherein the specialized target comparison candidate text is narrowed for each domain; and see last sentence in para 0208; and following, in para 0209, the text data includes other data such as linguistic representation, phoneme, data, syllable, emotion, speaker features, and language and context information – see all of para 209; in other words, the text to be speech synthesized contains the ‘specialization information’ to produce speech); with a basic transcript text database DB (hereinafter, “basic transcript text database (DB)”) included in the storage device, when a number of appearances of the comparison candidate text is less than or equal to a preset threshold value (as, looking at a threshold value of an n-best list, specialized for a certain domain – see para 0145), determining, by the processor, the comparison candidate text as a specialization target comparison candidate text, wherein the specialization target comparison candidate text is a domain text that is not included in the basic transcript text DB (as, adding/removing wakewords/intents /sentiments/words, etc. that are found during the usage recognition – para 0031); repeatedly performing, by the processor, the determining the comparison candidate text as the specialization target comparison candidate text, if it is determined that the comparison candidate texts remains in the domain text DB (as repeating the comparison/training – para 128-130, also including during the training, data curation – para 31, which includes the removal of wakewords/acoustic events/intents, and adjusting the probabilities – para 31; see further in para 0045, wherein the target may change/less significance based on the shifting of desired model features – see para 0045, after ‘the dataset may evolve…’) and requires additional training and constructing, by the processor, a specialization target domain text DB storing the specialization target comparison candidate text in the storage device (as, updating/training the model for each domain – para 128-130, wherein the specialized target comparison candidate text); generating, by the processor, a specialization target speech signal from the specialization target comparison candidate text, using the specialization target domain text DB, thereby constructing a specialization target speech DB ( as generating targeting speech for synthesis – para 0221; speech from a parametric speech database (Fig. 16, 1668a – 1668n; or alternatively, to save space, performing TTS without the TTS nit storage – see last sentence in para 0208; and following, in para 0209, the text data includes other data such as linguistic representation, phoneme, data, syllable, emotion, speaker features, and language and context information – see all of para 209; in other words, the text to be speech synthesized contains the ‘specialization information’ to produce speech); training, by the processor, a speech recognition neural network with the generated specialization target speech signal and after the training of the speech recognition neural network is completed, generating, by the processor, an end-to-end speech recognition model specialized for the domain to be specialized (the speech recognition process, and natural language process, as noted above, can be performed by neural networks – para 0028). As per claim 1, Strimel (11887583) teaches, in para 0208, 0209, tracking/matching data in the construction of the specialized database, the data including other data such as linguistic representation, phoneme, data, syllable, emotion, speaker features, and language and context information; however, does not explicitly teach using symbols, numbers, and foreign language notation; Skarbovsky et al (20180143974) teaches providing a transcription of multiple languages – para 0004, wherein data/content includes the language, language characters, and tracking numbers – see Figure 2D, and para 0050, and para 0062, reflecting back on para 0061. Therefore, it would have been obvious to one of ordinary skill in the art of speech/language transcription to further define the data/context tracking parameters found in Strimel (11887583) with symbols/numbers/foreign language information, as taught by Skarbovsky et al (20180143974) , because it would advantageously allow for the user to choose/select a section of the transcript of interest, based on the tracked information of the transcript – see Skarbovsky et al (20180143974) para 0055. As per claim 4, the combination of Strimel (11887583) in view of Skarbovsky et al (20180143974) teaches the method of claim 1, wherein the specialization target speech signal is generated using one of a single-speaker speech synthesizer and a multi-speaker speech synthesizer (see Strimel (11887583) ,see para 102 – “The TTS component – the speech output can match the multi-user speech input, or an artificial waveform). As per claim 5, the combination of Strimel (11887583) in view of Skarbovsky et al (20180143974) teaches the method of claim 1, wherein the training of the speech recognition neural network with the specialization target speech signal includes training the speech recognition neural network from the beginning with the generated specialized speech (see Strimel (11887583) ,as training the models, with the generated specialized/domainized speech – see para 0021, 0022 – starting with “Computer systems may employ machine learning algorithms…”. As per claim 6, the combination of Strimel (11887583) in view of Skarbovsky et al (20180143974) teaches the method of claim 1, wherein the training of the speech recognition neural network with the specialization target speech signal includes additionally training an existing general speech recognition neural network using one of connection learning and transfer learning (see Strimel (11887583) ,as using a fully connected neural network to determine sentiment categories during the network learning process – para 0108 “the systems 120 may also include a sentiment detection,,,). As per claim 7, the combination of Strimel (11887583) in view of Skarbovsky et al (20180143974) teaches the method of claim 1, further comprising generating a specialized language model that adjusts a weight of specialization target domain text of the specialization target domain text DB by changing an amount of specialization target domain text of the specialization target domain text DB (see Strimel (11887583) ,as, measuring the OTA budget – data size, and adjusting the updating of the models based on an accuracy/size weighting trade-off – para 0023, “Offered is a system, method, and other new technology to improve….and continued in para 0024). As per claim 8, the combination of Strimel (11887583) in view of Skarbovsky et al (20180143974) teaches the method of claim 1, further comprising extracting a specialized user vocabulary from the specialization target domain text DB in order to adjust a weight of specialization target domain text of the specialization target domain text DB by changing an amount of specialization target domain text of the specialization target domain text DB (see Strimel (11887583) ,as, using weights on the already generated model to update the model with new information – see para 0028, “In some systems 100”, see also para 0029, 0031 – adjusting weight to recognized object/faces, adjusting probabilities, etc.) , and constructing a specialized user vocabulary DB (see Strimel (11887583) ,as developing a domain/gazetteer for a particular user – see para 134, “Each recognizer 1463, and more specifically each NER component…”). Claims 12-16 are apparatus claims that perform the method steps of claims 1,4-9 above and as such, claims 12-16 are similar in scope and content to claims 1,4-9 above; therefore, claims 12-16 are rejected under similar rationale as presented against claims 1,4-9 above . Response to Arguments 07-37 AIA Applicant's arguments filed 2/13/2026 have been fully considered but they are not persuasive. Starting on pp 9 of the response, bottom, examiner argues that Strimel teaches, in para 0094, “uses an end-to-end model that combinate ASR/NLU w audio input and text output” clearly, 1) end-to-end processing is being taught, 2) the combining of the ASR + NLU IS, the model, in Strimel, and 3) Strimel’ s model is constantly trained and updated – see para 0030 for training, as well as para 0031, 0045. On page 10 of the response, applicant presents Fig. 2, and argues that Strimel “neither discloses nor suggests, the process in Fig. 2”; This is reinforce on p11 of the response. On pp 12-14 of the response, applicants present Fig 12,1 of Strimel and associated discussion, of the drawings. On pp 12-14 of the response, applicant present that Strimel is an OTA (Over -The – Airways) model updater, with efficiently updating a user device with speech recognition functionality over the air, then presents a general allegation o, “is unrelated to the series of process of the present embodiments. On page 16 of the response, Applicants argue “is not related to determining domain text that are no included in the based transcript text DB and require additional training. Then on pp 17, continuing on the same line-of-thought, Applicants quote partial sections of the referred to para 0031. Examiner notes, the most-relevant section of para 0031, is not duplicated by the applicant ; this most noteable missing section of para 0031, is: “ The usage data may additionally or alternatively include other data 113 that may represent interactions with other users 5 of the system 100 . In some implementations, the training dataset may include curated and/or annotated data. At a stage 140 , the system 100 may train the model update object 141 using the usage data and the previous-generation model 131 a “ Clearly, Strimel is teaching model updating. Further, in applicants response, on the bottom of pg 17 of the response, applicant’s argue that “’the model for each domain’ of Strimel is updating – as noted above, the newly-updated model includes new information and discarding old information – ie, the database is constantly updated/constructed – see further in para 0045 of Strimel. On to pp 18 of the response, examiner argues that the claim states ‘specialization target speech signal’ – para 221 explicitly states: First a unit selection engine 1630 determines what speech units to use and then it combines them so that the particular combined units match the desired phonemes and acoustic features and create the desired speech output. Units may be selected based on a cost function which represents how well particular units fit the speech segments to be synthesized. The cost function may represent a combination of different costs representing different aspects of how well a particular speech unit may work for a particular speech segment. For example, a target cost indicates how well an individual given speech unit matches the features of a desired speech output (e.g., pitch, prosody, etc.). A join cost represents how well a particular speech unit matches an adjacent speech unit (e.g., a speech unit appearing directly before or directly after the particular speech unit) for purposes of concatenating the speech units together in the eventual synthesized speech. The overall cost function is a combination of target cost, joint cost , and other costs that may be determined by the unit selection engine 1630 . As part of unit selection, the unit selection engine 1630 chooses the speech unit with the lowest overall combined cost. For example, a speech unit with a very low target cost may not necessarily be selected if its join cost is high. On pages 19-20 of the response, toward applicants arguments that ‘it is noted that Strimel does not disclose or suggest a set of processes for constructing a domain-specific speech recognition model using text data – examiner agues that Strimel teaches the operating on text (e.g., para 222, 2223) and text on domain based models – see para 0133-0138, as a minor sample. On pp 20-21 toward the arguments against the combination of Strimel and Skarbovsky, examiner argues that the commonality to both references, is the operating on text data, with the motivation added benefit of interpreting the transcript (see motivational statement provided in the 103 rejection) . Conclusion 07-40 AIA Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL . See MPEP § 706.07(a). 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. 07-96 AIA The prior art made of record and not relied upon is considered pertinent to applicant's disclosure. Sypniewski et al (20200035222) teaching the use of neural networks for end-to-end speech recognition. Aharoni et al (20220044684) teaches end-to-end speech recognition (para 0068) with domain specific databases (para 0069) Zhu et al (20220230628) teaches “end-to-end” speech recognition (para 0010) with a knowledge graph using multi-separate domains (para 0192-0194), which are more specific. Zhao et al (20200357388) teaches end-to-end speech recognition using acoustic/pronunciation/language models via neural network, para 0088, using topical sub-domains (para 0041). Any inquiry concerning this communication or earlier communications from the examiner should be directed to Michael Opsasnick, telephone number (571)272-7623, who is available Monday-Friday, 9am-5pm. If attempts to reach the examiner by telephone are unsuccessful, the examiner's supervisor, Mr. Richemond Dorvil, can be reached at (571)272-7602. 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). /Michael N Opsasnick/Primary Examiner, Art Unit 2658 05/28/2026 Application/Control Number: 17/979,471 Page 2 Art Unit: 2658 Application/Control Number: 17/979,471 Page 3 Art Unit: 2658 Application/Control Number: 17/979,471 Page 4 Art Unit: 2658 Application/Control Number: 17/979,471 Page 5 Art Unit: 2658 Application/Control Number: 17/979,471 Page 6 Art Unit: 2658 Application/Control Number: 17/979,471 Page 7 Art Unit: 2658 Application/Control Number: 17/979,471 Page 8 Art Unit: 2658 Application/Control Number: 17/979,471 Page 10 Art Unit: 2658
Prosecution Insights