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 .
DETAILED ACTION
This action is responsive to patent application as filed on 7/12/2024 which claims priority to Chinese Pat. App. No: 2024109061377 filed 7/5/2024.
This action is made Non-Final.
Claims 1 – 20 are pending in the case. Claims 1, 10, and 16 are independent claims.
Drawings
The drawings filed on 7/12/2024 have been accepted by the Examiner.
Claim Rejections - 35 USC § 102
The following is a quotation of the appropriate paragraphs of 35 U.S.C. 102 that form the basis for the rejections under this section made in this Office action:
A person shall be entitled to a patent unless –
(a)(1) the claimed invention was patented, described in a printed publication, or in public use, on sale, or otherwise available to the public before the effective filing date of the claimed invention.
Claim(s) 1, 4-8, 10-12, 14-17 and 20 is/are rejected under 35 U.S.C. 102(a)(1) as being anticipated by Gurbani (USPAT 11,961,511 B2).
Claim 1:
Gurbani discloses A computer-implemented method, comprising: generating, using a speech recognition model, a text-based representation of speech encoded in input audio, the text-based representation including confidence scores for individual words (Col 7 ln 49-56 and Col 8 ln 33-46: A goal of the ASR engine 1 is to map spoken words or utterances in an audio signal containing speech to their corresponding text. The ASR engine 1 can include a feature extraction unit 2, an acoustic model 3, a decoder 4 and a language model 5. The ASR engine 1 is arrange to generate the best guess transcription of the words that were spoken or uttered in the speech signal, as text output... A lexicon or dictionary can be used to map each word to its phonetic representation, and it can be used by the decoder 4 in the ASR engine 1 to convert the predicted phonemes into words and eventually complete sentences... The word sequence with the highest probability can be output as the final text. The ASR engine 1 can be arranged to include word confidence scores in the output); determining that the confidence score for a lower confidence word, of the individual words in the text-based representation, falls below a confidence threshold; providing, as input to a language model, a sentence including the lower confidence word and an indication of the lower confidence word (Col 10 ln 47-58, Col 21 ln 61-66: The RAILS model can receive the ASR-generated transcript as input (including word confidence scores), analyze and process the transcript and supply the result to an input of the BERT model, which can then further analyze and process the results and send its output to the word selector in the Filter 2. The word selector (for example, trained decision tree model) can fit the output to the model and synthesize a final output automatically (that is, without any human intervention) that includes a corrected transcript, with all, or substantially all errors and ambiguities in the original ASR-generated transcript replaced or corrected...the RAILS model was applied to all words with a confidence of 0.40 or less, the optimal threshold determined for this example and data. This was observed to be the lowest word confidence threshold of any model); providing, as additional input to the language model, contextual text data retrieved from at least one knowledge base relevant to a knowledge domain associated with the speech (Col 11 ln 42-64: The RAILS model can be arranged to take as input a word embedding for each word in a transcript corpus received from an ASR engine, count for all tri-grams that appear in the corpus, and each mis-transcribed word and the surrounding context words. For example, in the statement (1) above, the RAILS model can be arranged to determine the context words to be “and very” and “it”, while determining the mis-transcribed word to be “consider... The RAILS model can be arranged to analyze the transcript and identify similarities between words in a corpus by predicting the co-occurrence of words. Word embeddings can be created by identifying the words that occur within a context window. The RAILS model can be arranged to calculate how often each word occurs next to every other word within the context window, after normalizing for overall word frequency. The RAILS model can be arranged to build word embeddings based on context windows with a length of, for example, eight words. Context windows having fewer, or more than eight words are contemplated here as well); and receiving, from the language model, a second version of the sentence including an alternative word in place of the lower confidence word, the alternative word determined using the contextual text data from the at least one knowledge base relevant to the knowledge domain (Col 10 ln 47-58: The RAILS model can receive the ASR-generated transcript as input (including word confidence scores), analyze and process the transcript and supply the result to an input of the BERT model, which can then further analyze and process the results and send its output to the word selector in the Filter 2. The word selector (for example, trained decision tree model) can fit the output to the model and synthesize a final output automatically (that is, without any human intervention) that includes a corrected transcript, with all, or substantially all errors and ambiguities in the original ASR-generated transcript replaced or corrected).
Claim 2:
Gurbani discloses determining the knowledge base relevant to the knowledge domain; and retrieving, using a domain-adapted retriever model, the contextual text data determined to have at least a minimum probability of being relevant to the input audio (Col 8 ln 33-46: A lexicon or dictionary can be used to map each word to its phonetic representation, and it can be used by the decoder 4 in the ASR engine 1 to convert the predicted phonemes into words and eventually complete sentences. The language model 5 can determine which words or sequence of words are more likely, given the surrounding words as context. This context can be derived from neural networks or n-gram models trained on large textual datasets. The decoder 4 can use the acoustic model, the grammar, and the language model collectively to generate word sequences that are likely for a particular audio frame. The word sequence with the highest probability can be output as the final text. The ASR engine 1 can be arranged to include word confidence scores in the output).
Claim 4:
Gurbani discloses the knowledge base includes domain-specific examples in a plurality of different formats, the domain-specific examples included in the knowledge base without prior cleaning, labeling, or pre-processing (Col 2 ln 23-37: the second machine learning system can comprise a BERT model architecture or a GPT-3 model architecture. The BERT model architecture or the GPT-3 model architecture can be trained (or augmented) with, for example, domain specific data... The RAILS model can be constructed from a building block comprising: a dataset containing all unigrams, bigrams, trigrams and quadgrams present in a corpus of transcripts and their respective probabilities; and/or a phonetic encoding component for determining phonetic similarity between lexical items).
Claim 5:
Gurbani discloses applying a low confidence tag to the lower confidence word upon determining that the confidence score for the lower confidence word falls below the confidence threshold (Col 21 ln 61-66: the RAILS model was applied to all words with a confidence of 0.40 or less, the optimal threshold determined for this example and data. This was observed to be the lowest word confidence threshold of any model in this particular example, as it was the model with the lowest WER).
Claim 6:
Gurbani discloses fine-tuning the speech recognition model using at least the alternative word (Col 10 ln 47-58: The RAILS model can receive the ASR-generated transcript as input (including word confidence scores), analyze and process the transcript and supply the result to an input of the BERT model, which can then further analyze and process the results and send its output to the word selector in the Filter 2. The word selector (for example, trained decision tree model) can fit the output to the model and synthesize a final output automatically (that is, without any human intervention) that includes a corrected transcript, with all, or substantially all errors and ambiguities in the original ASR-generated transcript replaced or corrected).
Claim 7:
Gurbani discloses providing multiple domain-specific knowledge bases for use with the speech recognition model, wherein the speech recognition model is able to be adapted for use with multiple different domains without retraining of the speech recognition model (Col 6 ln 4-16: Most ASR engines are trained on acoustic and language models constructed from corpora that include news feeds, books, and blogs in order to achieve better generalization. Consequently, when such engines are presented an unseen dataset in a specific domain, the WER suffers by a non-trivial margin. While the drop in accuracy is acute when the domain is very specific (for example, health or medicine), the inventors have observed that the drop is sizable even when the domain is general enough, for example, hospitality. For such domains, the inventors have discovered that the systems and methods described herein can further lower WER values significantly).
Claim 8:
Gurbani discloses the speech model is an automatic speech recognition (ASR) model, and the language model is a large language model (LLM) (Col 7 ln 49-63 and Col 19 ln 16-21).
Claim 10:
Gurbani discloses At least one processor comprising one or more processing units to: generate, using a first model, a text-based representation of speech encoded in input audio, the text-based representation including confidence scores for individual words (Col 7 ln 49-56 and Col 8 ln 33-46: A goal of the ASR engine 1 is to map spoken words or utterances in an audio signal containing speech to their corresponding text. The ASR engine 1 can include a feature extraction unit 2, an acoustic model 3, a decoder 4 and a language model 5. The ASR engine 1 is arrange to generate the best guess transcription of the words that were spoken or uttered in the speech signal, as text output... A lexicon or dictionary can be used to map each word to its phonetic representation, and it can be used by the decoder 4 in the ASR engine 1 to convert the predicted phonemes into words and eventually complete sentences... The word sequence with the highest probability can be output as the final text. The ASR engine 1 can be arranged to include word confidence scores in the output); determine that the confidence score for a lower confidence word, of the individual words in the text-based representation, falls below a confidence threshold; provide, as input to a language model, a prompt including a sequence of words from the text-based representation including the lower confidence word, an indication of the low-confidence word, and contextual text data extracted from a domain-specific knowledge base (Col 10 ln 47-58, Col 11 ln 42-64, Col 21 ln 61-66: The RAILS model can receive the ASR-generated transcript as input (including word confidence scores), analyze and process the transcript and supply the result to an input of the BERT model, which can then further analyze and process the results and send its output to the word selector in the Filter 2. The word selector (for example, trained decision tree model) can fit the output to the model and synthesize a final output automatically (that is, without any human intervention) that includes a corrected transcript, with all, or substantially all errors and ambiguities in the original ASR-generated transcript replaced or corrected... The RAILS model can be arranged to take as input a word embedding for each word in a transcript corpus received from an ASR engine, count for all tri-grams that appear in the corpus, and each mis-transcribed word and the surrounding context words. For example, in the statement (1) above, the RAILS model can be arranged to determine the context words to be “and very” and “it”, while determining the mis-transcribed word to be “consider... The RAILS model can be arranged to analyze the transcript and identify similarities between words in a corpus by predicting the co-occurrence of words. Word embeddings can be created by identifying the words that occur within a context window. The RAILS model can be arranged to calculate how often each word occurs next to every other word within the context window, after normalizing for overall word frequency. The RAILS model can be arranged to build word embeddings based on context windows with a length of, for example, eight words. Context windows having fewer, or more than eight words are contemplated here as well...the RAILS model was applied to all words with a confidence of 0.40 or less, the optimal threshold determined for this example and data. This was observed to be the lowest word confidence threshold of any model); and receive, from the language model, a second sequence of words including an alternative word in place of the lower confidence word, the alternative word determined using the contextual text data from the domain-specific knowledge base (Col 10 ln 47-58: The RAILS model can receive the ASR-generated transcript as input (including word confidence scores), analyze and process the transcript and supply the result to an input of the BERT model, which can then further analyze and process the results and send its output to the word selector in the Filter 2. The word selector (for example, trained decision tree model) can fit the output to the model and synthesize a final output automatically (that is, without any human intervention) that includes a corrected transcript, with all, or substantially all errors and ambiguities in the original ASR-generated transcript replaced or corrected).
Claim 11:
Gurbani discloses the first model is an automatic speech recognition (ASR) model, and the language model is a large language model (LLM) (Col 7 ln 49-63 and Col 19 ln 16-21).
Claim 12:
Gurbani discloses identify the domain-specific knowledge base; and retrieve, using a domain-adapted retriever model, the contextual text data determined to have at least a minimum probability of being relevant to the input audio (Col 8 ln 33-46: A lexicon or dictionary can be used to map each word to its phonetic representation, and it can be used by the decoder 4 in the ASR engine 1 to convert the predicted phonemes into words and eventually complete sentences. The language model 5 can determine which words or sequence of words are more likely, given the surrounding words as context. This context can be derived from neural networks or n-gram models trained on large textual datasets. The decoder 4 can use the acoustic model, the grammar, and the language model collectively to generate word sequences that are likely for a particular audio frame. The word sequence with the highest probability can be output as the final text. The ASR engine 1 can be arranged to include word confidence scores in the output).
Claim 14:
Gurbani discloses the knowledge base includes domain-specific examples in a plurality of different formats, the domain-specific examples included in the knowledge base without prior cleaning, labeling, or pre-processing (Col 2 ln 23-37: the second machine learning system can comprise a BERT model architecture or a GPT-3 model architecture. The BERT model architecture or the GPT-3 model architecture can be trained (or augmented) with, for example, domain specific data... The RAILS model can be constructed from a building block comprising: a dataset containing all unigrams, bigrams, trigrams and quadgrams present in a corpus of transcripts and their respective probabilities; and/or a phonetic encoding component for determining phonetic similarity between lexical items).
Claim 15:
Gurbani discloses 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 implemented at least partially in a data center; a system for performing hardware testing using simulation; a system for synthetic data generation; a system for performing generative AI operations using a large language model (LLM), a system for performing generative AI operations using a vision language model (VLM), a system for performing generative AI operations using a multi-modal language model, a collaborative content creation platform for 3D assets; or a system implemented at least partially using cloud computing resources (Col 30 ln 57-67 and claim 9).
Claim 16:
Gurbani discloses A system (Claim 9) comprising: one or more processors to improve the accuracy of a transcript generated using a speech recognition model by, in part, providing at least a portion of the transcript including one or more lower confidence words to a language model along with contextual text data extracted from a domain-specific knowledge base, wherein the language model replaces at least one of the lower confidence words with one or more alternative words inferred from the contextual text data (Col 7 ln 49-56, Col 8 ln 33-46, Col 10 ln 47-58, Col 11 ln 42-64, Col 21 ln 61-66: A goal of the ASR engine 1 is to map spoken words or utterances in an audio signal containing speech to their corresponding text. The ASR engine 1 can include a feature extraction unit 2, an acoustic model 3, a decoder 4 and a language model 5. The ASR engine 1 is arrange to generate the best guess transcription of the words that were spoken or uttered in the speech signal, as text output... A lexicon or dictionary can be used to map each word to its phonetic representation, and it can be used by the decoder 4 in the ASR engine 1 to convert the predicted phonemes into words and eventually complete sentences... The word sequence with the highest probability can be output as the final text. The ASR engine 1 can be arranged to include word confidence scores in the output...The RAILS model can receive the ASR-generated transcript as input (including word confidence scores), analyze and process the transcript and supply the result to an input of the BERT model, which can then further analyze and process the results and send its output to the word selector in the Filter 2. The word selector (for example, trained decision tree model) can fit the output to the model and synthesize a final output automatically (that is, without any human intervention) that includes a corrected transcript, with all, or substantially all errors and ambiguities in the original ASR-generated transcript replaced or corrected...The RAILS model can be arranged to take as input a word embedding for each word in a transcript corpus received from an ASR engine, count for all tri-grams that appear in the corpus, and each mis-transcribed word and the surrounding context words. For example, in the statement (1) above, the RAILS model can be arranged to determine the context words to be “and very” and “it”, while determining the mis-transcribed word to be “consider... The RAILS model can be arranged to analyze the transcript and identify similarities between words in a corpus by predicting the co-occurrence of words. Word embeddings can be created by identifying the words that occur within a context window. The RAILS model can be arranged to calculate how often each word occurs next to every other word within the context window, after normalizing for overall word frequency. The RAILS model can be arranged to build word embeddings based on context windows with a length of, for example, eight words. Context windows having fewer, or more than eight words are contemplated here as well)the RAILS model was applied to all words with a confidence of 0.40 or less, the optimal threshold determined for this example and data. This was observed to be the lowest word confidence threshold of any model).
Claim 17:
Gurbani discloses the domain-specific knowledge base includes domain-specific examples in a plurality of different formats relevant to a knowledge base associated with speech used to generate the transcript, the domain-specific examples allowed to be included in the knowledge base without prior cleaning, labeling, or pre-processing (Col 2 ln 23-37: the second machine learning system can comprise a BERT model architecture or a GPT-3 model architecture. The BERT model architecture or the GPT-3 model architecture can be trained (or augmented) with, for example, domain specific data... The RAILS model can be constructed from a building block comprising: a dataset containing all unigrams, bigrams, trigrams and quadgrams present in a corpus of transcripts and their respective probabilities; and/or a phonetic encoding component for determining phonetic similarity between lexical items).
Claim 20:
Gurbani discloses 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 implemented at least partially in a data center; a system for performing hardware testing using simulation; a system for synthetic data generation; a system for performing generative AI operations using a large language model (LLM), a system for performing generative AI operations using a vision language model (VLM), a system for performing generative AI operations using a multi-modal language model, a collaborative content creation platform for 3D assets; or a system implemented at least partially using cloud computing resources (Col 30 ln 57-67 and claim 9).
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) 3 and 13 is/are rejected under 35 U.S.C. 103 as being unpatentable over Gurbani in view of Anderson (USPUB 20250173313 A1 which claims priority to foreign application (GB) No: 2318258.7 filed Nov. 29, 2023).
Claim 3:
Gurbani discloses every feature of claim 2.
Gurbani, by itself, does not seem to completely teach the domain-adapted retriever model generates an index of domain-relevant text content in a format appropriate for the language model.
The Examiner maintains that these features were previously well-known as taught by Anderson.
Anderson teaches the domain-adapted retriever model generates an index of domain-relevant text content in a format appropriate for the language model (0138: techniques such as Retrieval Augmented Generation (RAG) can be applied to avoid the need to process all records within each search. In RAG, a pre-generated index of LLM-derived embedding vectors is constructed. This enables a subset of files to be identified that are relevant to a given query. The subset is typically selected using distance metrics in the embedding vector space. This indexing approach allows a much larger number of files to be interrogated but has the limitation that the query is evaluated only on a subset of documents chosen using those embedding vectors, not on the entire set of files).
Gurbani and Anderson are analogous art because they are from the same problem-solving area, automatic speech recognition.
Before the effective filing date of the claimed invention, it would have been obvious to one of ordinary skill in the art, having the teachings of Gurbani and Anderson before him or her, to combine the teachings of Gurbani and Anderson. The rationale for doing so would have been to not have to make the ASR system more efficient.
Therefore, it would have been obvious to combine Gurbani and Anderson to obtain the invention as specified in the instant claim(s).
Claim 13:
Gurbani discloses every feature of claim 10.
Gurbani, by itself, does not seem to completely teach the domain-adapted retriever model generates an index of domain-relevant text content in a format appropriate for the language model.
The Examiner maintains that these features were previously well-known as taught by Anderson.
Anderson teaches the domain-adapted retriever model generates an index of domain-relevant text content in a format appropriate for the language model (0138: techniques such as Retrieval Augmented Generation (RAG) can be applied to avoid the need to process all records within each search. In RAG, a pre-generated index of LLM-derived embedding vectors is constructed. This enables a subset of files to be identified that are relevant to a given query. The subset is typically selected using distance metrics in the embedding vector space. This indexing approach allows a much larger number of files to be interrogated but has the limitation that the query is evaluated only on a subset of documents chosen using those embedding vectors, not on the entire set of files).
Gurbani and Anderson are analogous art because they are from the same problem-solving area, automatic speech recognition.
Before the effective filing date of the claimed invention, it would have been obvious to one of ordinary skill in the art, having the teachings of Gurbani and Anderson before him or her, to combine the teachings of Gurbani and Anderson. The rationale for doing so would have been to not have to make the ASR system more efficient.
Therefore, it would have been obvious to combine Gurbani and Anderson to obtain the invention as specified in the instant claim(s).
Claim(s) 9, 18 and 19 is/are rejected under 35 U.S.C. 103 as being unpatentable over Gurbani in view of Bachura (USPUB 20250046334 A1 filed Aug. 1, 2024 claiming priority to U.S. Prov. App. No: 63/530,304 filed on Aug. 2, 2023).
Claim 9:
Gurbani discloses every feature of claim 1.
Gurbani further teaches generating, as input to the language model, a prompt including the sentence including the lower confidence word, an indication of the low-confidence word, and the contextual text data (Col 10 ln 47-58, Col 11 ln 42-64, Col 21 ln 61-66: The RAILS model can receive the ASR-generated transcript as input (including word confidence scores), analyze and process the transcript and supply the result to an input of the BERT model, which can then further analyze and process the results and send its output to the word selector in the Filter 2. The word selector (for example, trained decision tree model) can fit the output to the model and synthesize a final output automatically (that is, without any human intervention) that includes a corrected transcript, with all, or substantially all errors and ambiguities in the original ASR-generated transcript replaced or corrected...The RAILS model can be arranged to take as input a word embedding for each word in a transcript corpus received from an ASR engine, count for all tri-grams that appear in the corpus, and each mis-transcribed word and the surrounding context words. For example, in the statement (1) above, the RAILS model can be arranged to determine the context words to be “and very” and “it”, while determining the mis-transcribed word to be “consider... The RAILS model can be arranged to analyze the transcript and identify similarities between words in a corpus by predicting the co-occurrence of words. Word embeddings can be created by identifying the words that occur within a context window. The RAILS model can be arranged to calculate how often each word occurs next to every other word within the context window, after normalizing for overall word frequency. The RAILS model can be arranged to build word embeddings based on context windows with a length of, for example, eight words. Context windows having fewer, or more than eight words are contemplated here as well... the RAILS model was applied to all words with a confidence of 0.40 or less, the optimal threshold determined for this example and data. This was observed to be the lowest word confidence threshold of any model).
Gurbani, by itself, does not seem to completely teach generating, as input to the language model and using a retrieval augmented generation (RAG) pipeline.
The Examiner maintains that these features were previously well-known as taught by Bachura.
Bachura teaches generating, as input to the language model and using a retrieval augmented generation (RAG) pipeline (0021: The large language model evaluation process 210 includes behavioral machine learning model input features 211 and an ensemble of LLMs 213 that are a mixture of fine-tuned and retrieval-augmented generative (RAG) variants. This ensemble of LLM models 213 evaluates final inference model assumption requirements and generates a behavior and personality measures output 214, alleviating the need for human-in-the-loop interaction prior to the inferencing of the specialized behavioral models).
Gurbani and Bachura are analogous art because they are from the same problem-solving area, automatic speech recognition.
Before the effective filing date of the claimed invention, it would have been obvious to one of ordinary skill in the art, having the teachings of Gurbani and Bachura before him or her, to combine the teachings of Gurbani and Bachura. The rationale for doing so would have been to obtain the benefit of training the language model in order to have a more accurate ASR system.
Therefore, it would have been obvious to combine Gurbani and Bachura to obtain the invention as specified in the instant claim(s).
Claim 18:
Gurbani discloses a domain-adapted retriever to retrieve the contextual text data determined to be potentially relevant to a content of the transcript (Col 8 ln 33-46: A lexicon or dictionary can be used to map each word to its phonetic representation, and it can be used by the decoder 4 in the ASR engine 1 to convert the predicted phonemes into words and eventually complete sentences. The language model 5 can determine which words or sequence of words are more likely, given the surrounding words as context. This context can be derived from neural networks or n-gram models trained on large textual datasets. The decoder 4 can use the acoustic model, the grammar, and the language model collectively to generate word sequences that are likely for a particular audio frame. The word sequence with the highest probability can be output as the final text. The ASR engine 1 can be arranged to include word confidence scores in the output).
Gurbani, by itself, does not seem to completely teach the language model is part of a retrieval augmented generation (RAG) pipeline.
The Examiner maintains that these features were previously well-known as taught by Bachura.
Bachura teaches the language model is part of a retrieval augmented generation (RAG) pipeline (0021: The large language model evaluation process 210 includes behavioral machine learning model input features 211 and an ensemble of LLMs 213 that are a mixture of fine-tuned and retrieval-augmented generative (RAG) variants. This ensemble of LLM models 213 evaluates final inference model assumption requirements and generates a behavior and personality measures output 214, alleviating the need for human-in-the-loop interaction prior to the inferencing of the specialized behavioral models).
Gurbani and Bachura are analogous art because they are from the same problem-solving area, automatic speech recognition.
Before the effective filing date of the claimed invention, it would have been obvious to one of ordinary skill in the art, having the teachings of Gurbani and Bachura before him or her, to combine the teachings of Gurbani and Bachura. The rationale for doing so would have been to obtain the benefit of training the language model in order to have a more accurate ASR system.
Therefore, it would have been obvious to combine Gurbani and Bachura to obtain the invention as specified in the instant claim(s).
Claim 19:
Gurbani teaches the domain-specific knowledge base includes domain-specific examples in a plurality of different formats, the domain-specific examples included in the knowledge base without prior cleaning, labeling, or pre-processing (Col 2 ln 23-37: the second machine learning system can comprise a BERT model architecture or a GPT-3 model architecture. The BERT model architecture or the GPT-3 model architecture can be trained (or augmented) with, for example, domain specific data... The RAILS model can be constructed from a building block comprising: a dataset containing all unigrams, bigrams, trigrams and quadgrams present in a corpus of transcripts and their respective probabilities; and/or a phonetic encoding component for determining phonetic similarity between lexical items).
Note
The Examiner cites particular columns, line numbers and/or paragraph numbers in the references as applied to the claims below for the convenience of the Applicant(s). Although the specified citations are representative of the teachings in the art and are applied to the specific limitations within the individual claim, other passages and figures may apply as well. It is respectfully requested that, in preparing responses, the Applicant fully consider the references in their entirety as potentially teaching all or part of the claimed invention, as well as the context of the passage as taught by the prior art or disclosed by the Examiner. See MPEP 2123.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure and is listed in the attached PTOL-892 form.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to MOHAMMED-IBRAHIM ZUBERI whose telephone number is (571)270-7761. The examiner can normally be reached on M-Th 8-6 Fri: 7-12/OFF.
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, Steph Hong can be reached on (571) 272-4124. 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.
/MOHAMMED H ZUBERI/Primary Examiner, Art Unit 2178
