DETAILED ACTION
This office action is in response to Applicant’s submission filed on 2/13/2025. Claims 1-20 are pending in the application. As such, claims 1-20 have been examined.
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 .
Information Disclosure Statement
The information disclosure statement (IDS) was submitted on 8/19/2025. The submission is in compliance with the provisions of 37 CFR 1.97. Accordingly, the information disclosure statement is being considered by the examiner.
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.
Claims 1, 4, 9-11, 14, and 19-20 are rejected under 35 U.S.C. 103 as being unpatentable over Huang et al. (Huang, W. Ronny, et al. "Detecting unintended memorization in language-model-fused ASR." arXiv preprint arXiv:2204.09606 (2022).; hereinafter referred to as Huang) in view of Rosenberg et al. (US 20230013587 A1; hereinafter referred to as Rosenberg).
Regarding claim 1, Huang discloses: obtaining a plurality of training canary transcriptions ([3.2] For the speech model, the acoustics are further diversified via multi-style training [29], random down-sampling from 16 to 8 kHz, and SpecAug [30]. The LM uses only the transcripts of the training set. The canaries are merged into the LM’s training set), each training canary transcription comprising a predetermined number of words that are out-of-distribution from words of the un-transcribed speech utterances ([3.1] Each canary is a fixed-length sequence of random lower-case alphabetical characters with spaces between them. These canaries are implicitly out-of-distribution.);
for each training canary transcription, generating, using a text-to-speech (TTS) system, a corresponding synthetic training canary speech utterance ([2.1] we query an AM+LM recognizer to detect the level the unintended memorization of the canaries. This is done by first using a WaveNet TTS system [21] to generate synthetic utterances for the canaries) that recites the predetermined number of words of the training canary transcription… ([3.1] Each canary is a fixed-length sequence of random lower-case alphabetical characters with spaces between them);
and measuring an un-intended memorization of the pre-trained audio encoder based on encoder labels ([1] we design a simple classifier that can use the top transcripts for these utterances to conduct a membership inference (MI) task. We provide an analysis of how metrics like precision and recall (commonly used for evaluating MI attacks [16, 17, 18]) can be used to demonstrate the extent of unintended memorization in ASR LMs) predicted by the pre-trained audio encoder for the synthetic training canary speech utterances ([2.1] use two methods to measure unintended memorization: First, we compare the word error rate (WER) of the canaries from a canary-trained model, and contrast it with the same measurement from an extraneous-trained model. Second, as described later in §2.3, we also design a membership classifier that modifies the TTS utterances to accurately predict memorization on a per-example basis).
Huang does not explicitly, but Rosenberg teaches: a computer-implemented method executed on data processing hardware that causes the data processing hardware to perform operations comprising: receiving a training data set comprising un-transcribed speech utterances that each comprise audio-only data not paired with any corresponding transcription… ([0037] Each unspoken training text utterance 320 includes text-only data (i.e., unpaired data) such that each unspoken training text utterance 320 is not paired any corresponding spoken audio representation (speech) of the utterance. Each un-transcribed non-synthetic speech utterance 306 (also referred to as simply “un-transcribed speech utterance 306”) includes audio-only data (i.e., unpaired data) such that the un-transcribed speech utterance 306 is not paired with any corresponding transcription);
pre-training an audio encoder on a combination of the un-transcribed speech utterances and the synthetic training canary speech utterances… ([0004] pre-training an audio encoder on the synthetic speech representations generated for the unspoken textual utterances, the un-transcribed non-synthetic speech utterances, and the transcribed non-synthetic speech utterances to teach the audio encoder to jointly learn shared speech and text representations).
Huang and Rosenberg are considered analogous in the field of automatic speech recognition (ASR). 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 teachings of Huang to combine the teachings of Rosenberg because doing so would allow pre-training of an audio encoder using synthesized speech and audio-only speech for determining a contrastive loss, leading to improved accuracy of an ASR model when handling out-of-domain data (Rosenberg [0064] The text selection process 400 aims to select a subset of the available unspoken textual utterances 320 from the unspoken text corpus 402 as the data for TTS synthesis resulting in the synthetic speech representations 332 generated for pre-training the audio encoder 210 during the contrastive loss and supervised loss parts 300a, 300b of the training process 300 described above with reference to FIGS. 3A and 3B. Stated differently, the text selection process 400 aims to improve the match between the selected subset of the available unspoken textual utterances 320 and the specific domain being targeted, which in turn reduces the computational resources required to exploit a large amount of non-domain-specific data).
Regarding claim 4, the combination of Huang and Rosenberg teaches: the method of claim 1. Rosenberg further teaches: wherein the operations further comprise fine-tuning an automatic speech recognition (ASR) model on supervised training utterances ([0028] implementations are directed toward pre-training an audio encoder of the ASR model on training data that includes un-transcribed non-synthetic speech utterances, unspoken textual utterances for generating corresponding synthetic speech representations, and the transcribed non-synthetic speech utterances to jointly learn speech and text representations, and then fine-tuning (e.g., warm-start training) the pre-trained ASR model using the available transcribed non-synthetic speech utterances), the ASR model implementing the pre-trained audio encoder and a decoder ([0046] the supervised loss part 300b of the training process 300 is configured to inject lexical information into the audio encoder 210 during pre-training based on supervised loss terms 342, 344 derived from the transcribed non-synthetic speech utterances 304 and the synthesized speech representations 332 generated by the TTS system 330 for the unspoken textual utterances 320. Notably, the supervised loss part 300b leverages one or more auxiliary decoders 390 for generating the supervised loss terms).
Regarding claim 9, the combination of Huang and Rosenberg teaches: the method of claim 1. Huang further teaches: wherein the operations further comprise applying sensitivity-bounded training when pre-training the audio encoder ([2.2] large models trained using DP-SGD with strong privacy guarantees tend to have unacceptably low utility (i.e. generalization) compared with baseline (non-private) models. A possible step forward is to only perform per-example clipping, thus bounding the sensitivity of each gradient. A model trained this way no longer satisfies any meaningful differential privacy, but prior work [10, 12] has shown that such models tend to match the utility of baseline models while being empirically less prone to memorization).
Regarding claim 10, the combination of Huang and Rosenberg teaches: the method of claim 9. Huang further teaches: wherein the sensitivity-bounded training comprises per-example clipping wherein gradients of each example in a mini-batch is clipped before being averaged ([4.3] we report the results on different levels of per-example gradient clipping (discussed in §2.2). At a clip norm of 32, no gradients are clipped. Thus, the top row can be seen as a non-clipped baseline that is most prone to memorization, similar to the models in §4.2. At a clip norm of 0.5, 99% of the per-example gradients get clipped. At each clip norm level, we train an LM on the canaries (70M-CAN), along with a corresponding LM on the extraneous set (70M EXT) as a baseline. This allows us to make well-baselined measurements of canary memorization).
Regarding claim 11, Rosenberg teaches: data processing hardware; and memory hardware in communication with the data processing hardware and storing instructions that when executed on the data processing hardware cause the data processing hardware to perform operations… ([0071] The method 600 may execute on data processing hardware 710 (FIG. 7) using instructions stored on memory hardware 720 (FIG. 7). The data processing hardware 710 and the memory hardware 720 may reside on the remote computer/server 201 of FIG. 1 corresponding to a computing device 700 (FIG. 7)). The rest of the claim recites similar limitations to claim 1 and therefore is rejected similarly.
Regarding claim 14, it recites similar limitations to claim 1 and therefore is rejected similarly.
Regarding claim 19, it recites similar limitations to claim 9 and therefore is rejected similarly.
Regarding claim 20, it recites similar limitations to claim 10 and therefore is rejected similarly.
Claims 2 and 12 are rejected under 35 U.S.C. 103 as being unpatentable over Huang in view of Rosenberg, as applied to claims 1, 4, 9-11, 14, and 19-20 above, and further in view of Chiu et al. (Chiu, Chung-Cheng, et al. "Self-supervised learning with random-projection quantizer for speech recognition." International Conference on Machine Learning. PMLR, 2022.; hereinafter referred to as Chiu).
Regarding claim 2, the combination of Huang and Rosenberg teaches: the method of claim 1. Rosenberg further teaches: after masking a subset of the audio features in the sequence of audio features associated with the corresponding un-transcribed non-synthetic speech utterance or the corresponding synthetic training canary speech utterance ([0043-0044] generates, as output, for each of a plurality of output steps, an encoded feature 211 that corresponds to a respective one of one of the synthesized speech representations 332, one of the transcribed non-synthetic speech utterances 304, or one of the un-transcribed non-synthetic speech utterances 306. The encoded features 211 output from the convolution subsampling block 212 may be fed to a masking module 218 where some of the encoded features 211 are randomly chosen and replaced with a trained feature vector shared between all masked time steps to provide corresponding masked encoded features 211m), generating, by the audio encoder, contrastive context vectors from corresponding masked audio features ([0044] After masking is applied, the linear layer 214 and the Conformer blocks 216 of the context network receives the masked encoded features 211m and outputs corresponding contrastive context vectors 215 from masked encoded features 211m);
and deriving a loss term between the contrastive context vectors at the masked positions and the target token index ([0045] The contrastive loss 316 is optimized between the contrastive context vectors 215 at the masked positions and the target context vectors 219);
and pre-training the audio encoder based on the loss terms derived for each of the un- transcribed speech utterances and the synthetic training canary speech utterances ([0045] the training process 300 pre-trains the audio encoder 210 on the derived contrastive loss 316 applied on the corresponding encoded features 211 associated with each synthesized speech representation 332, each transcribed non-synthetic speech utterance 304, and each un-transcribed non-synthetic speech utterance 306 provided as input to the audio encoder 210. Pre-training the audio encoder 210 may include updating parameters of the audio encoder based on the contrastive losses).
The combination of Huang and Rosenberg does not explicitly, but Chiu teaches: wherein pre-training the audio encoder on the combination of the un-transcribed speech utterances and the synthetic training canary speech utterances comprises: for each corresponding un-transcribed speech utterance and each synthetic training canary speech utterance: generating, at each of a plurality of output steps, using a random- projection quantizer, a target quantized vector token and a target token index for a corresponding audio feature in a sequence of audio features associated with the corresponding un-transcribed speech utterance or the corresponding synthetic training canary speech utterance ([1] The algorithm masks speech signals and feeds them to the encoder part of the speech recognition model, and the encoder learns to predict the masked region based on the unmasked speech signals where the learning targets are labels provided by a random-projection quantizer. The random projection quantizer projects speech signals to a randomly initialized matrix, and finds a nearest vector in a randomly initialized codebook. The index of that vector is the target label), wherein the target token index maps the corresponding audio feature to the target quantized vector token stored in one or more codebooks… ([3] The quantizer randomly initializes a matrix and a codebook, and uses the matrix to project the input speech signals and the codebook to find the nearest vector where the index of the vector is the label).
Huang, Rosenberg, and Chiu are considered analogous in the field of automatic speech recognition. 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 teachings of Huang and Rosenberg to combine the teachings of Chiu because doing so would allow for the use of BERT-based Speech pre-Training with Random-projection Quantizer (BEST-RQ) to perform self-supervised learning for speech recognition models, improving speech recognition by using a simpler framework to perform a target task (Chiu [5] untangle the quantizer from the speech recognition model and also eliminates the requirement of representation learning. This simpler framework makes it easier to find a good recipe for the target task. The improvement on streaming models shows that the separation of the quantizer from the model makes the algorithm more effective for architectures that can be less effective for representation learning. The improvement on multilingual tasks shows that complicated tasks can benefit more from a simpler framework).
Regarding claim 12, it recites similar limitations to claim 2 and therefore is rejected similarly.
Claims 3 and 13 are rejected under 35 U.S.C. 103 as being unpatentable over Huang in view of Rosenberg, as applied to claims 1, 4, 9-11, 14, and 19-20 above, and further in view of Chiu and Liu et al. (Liu, Zhe, Xuedong Zhang, and Fuchun Peng. "Mitigating unintended memorization in language models via alternating teaching." ICASSP 2023-2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2023.; hereinafter referred to as Liu).
Regarding claim 3, the combination of Huang and Rosenberg teaches: the method of claim 1. Rosenberg further teaches: after applying a word-level mask masking a subset of the audio features in the sequence of audio features associated with a word in the corresponding unseen canary speech utterance or the corresponding synthetic training canary speech utterance ([0043-0044] generates, as output, for each of a plurality of output steps, an encoded feature 211 that corresponds to a respective one of one of the synthesized speech representations 332, one of the transcribed non-synthetic speech utterances 304, or one of the un-transcribed non-synthetic speech utterances 306. The encoded features 211 output from the convolution subsampling block 212 may be fed to a masking module 218 where some of the encoded features 211 are randomly chosen and replaced with a trained feature vector shared between all masked time steps to provide corresponding masked encoded features 211m), generating, by the audio encoder, contrastive context vectors from corresponding masked audio features ([0044] After masking is applied, the linear layer 214 and the Conformer blocks 216 of the context network receives the masked encoded features 211m and outputs corresponding contrastive context vectors 215 from masked encoded features 211m), wherein the contrastive context vectors comprise the predicted encoder labels ([0029] the supervised losses may be derived from speech recognition labels predicted by one or more auxiliary decoders based on the latent speech representations generated by the audio encoder from corresponding ones of the synthetic speech representations and the transcribed non-synthetic speech utterances);
and deriving a loss term between the contrastive context vectors at the masked word positions and the target token index… ([0045] The contrastive loss 316 is optimized between the contrastive context vectors 215 at the masked positions and the target context vectors 219).
The combination of Huang and Rosenberg does not explicitly, but Chiu teaches: for each corresponding unseen canary speech utterance and each corresponding synthetic training canary speech utterance: generating, at each of a plurality of output steps, using a random- projection quantizer, a target quantized vector token and a target token index for a corresponding audio feature in a sequence of audio features associated with the corresponding unseen canary speech utterance or the corresponding synthetic training canary speech utterance ([1] The algorithm masks speech signals and feeds them to the encoder part of the speech recognition model, and the encoder learns to predict the masked region based on the unmasked speech signals where the learning targets are labels provided by a random-projection quantizer. The random projection quantizer projects speech signals to a randomly initialized matrix, and finds a nearest vector in a randomly initialized codebook. The index of that vector is the target label), wherein the target token index maps the corresponding audio feature to the target quantized vector token stored in one or more codebooks… ([3] The quantizer randomly initializes a matrix and a codebook, and uses the matrix to project the input speech signals and the codebook to find the nearest vector where the index of the vector is the label).
Huang, Rosenberg, and Chiu are considered analogous in the field of automatic speech recognition. 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 teachings of Huang and Rosenberg to combine the teachings of Chiu because doing so would allow for the use of BERT-based Speech pre-Training with Random-projection Quantizer (BEST-RQ) to perform self-supervised learning for speech recognition models, improving speech recognition by using a simpler framework to perform a target task (Chiu [5] untangle the quantizer from the speech recognition model and also eliminates the requirement of representation learning. This simpler framework makes it easier to find a good recipe for the target task. The improvement on streaming models shows that the separation of the quantizer from the model makes the algorithm more effective for architectures that can be less effective for representation learning. The improvement on multilingual tasks shows that complicated tasks can benefit more from a simpler framework).
The combination of Huang, Rosenberg, and Chiu does not explicitly, but Liu teaches: wherein measuring the un-intended memorization of the pre-trained audio encoder comprises: obtaining a plurality of unseen canary speech utterances, each unseen canary speech utterance comprising words that are out-of-distribution from the words of the un- transcribed speech utterances… ([3.2] The procedure illustrated above is intended to simulate real-world scenarios where any occurrences of user-specific unique or rare out of-distribution canaries are typically limited to a very small fraction of users, but these users can exhibit either low or high usage of those canaries);
and measuring the un-intended memorization of the pre-trained audio encoder based on a comparison between the loss terms derived for the unseen canary speech utterances and the loss terms derived for the synthetic training canary speech utterances ([4.4] we measure the performance of (student) LMs trained on LibriSpeech text-only corpus with inserted canaries. The test split of LibriSpeech ASR corpus is used for evaluating the PPL and WER results. Seen from Table 3, the utility gaps are relatively small over all methods with different numbers of teachers, which can be explained by the large training corpus of 40M sentences. Particularly, WERs only differ in less than 1% comparing Agg and Alt with Baseline. Thus, all these approaches tend to match the baseline utility while being empirically less prone to memorization).
Huang, Rosenberg, Chiu, and Liu are considered analogous in the field of automatic speech recognition. 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 teachings of Huang, Rosenberg, and Chiu to combine the teachings of Liu because doing so would allow for the use of out-of-distribution training canary speech utterances to be compared to synthetic speech utterances to determine a loss comparison, leading to improved unintended memorization detection and mitigation using the loss comparison (Liu [5] we propose the alternating teaching method to mitigate unintended memorization in sequential modeling. With experiments on LibriSpeech datasets, we show this approach achieves stronger mitigation than other counterparts and significantly reduces memorized sequences. Compared with the baselines without protections for memorizing private data, the overall quality of proposed method is not compromised when there exists sufficient training data).
Regarding claim 13, it recites similar limitations to claim 3 and therefore is rejected similarly.
Claims 5, 7, 15, and 17 are rejected under 35 U.S.C. 103 as being unpatentable over Huang in view of Rosenberg, as applied to claims 1, 4, 9-11, 14, and 19-20 above, and further in view of Vishnoi et al. (US 20210082400 A1; hereinafter referred to as Vishnoi).
Regarding claim 5, the combination of Huang and Rosenberg teaches: the method of claim 1. The combination of Huang and Rosenberg does not explicitly, but Vishnoi teaches: wherein obtaining the plurality of training canary transcriptions comprises: sampling, from a training text corpus, the top-N most frequent words that appear in the training text corpus ([0112] augmenting training data with stop words (e.g., non-stop words or phrases of non-stop words may be randomly replaced with one or more stop words or one or more phrases of stop words)… The stop words may include a list of words for a particular domain or intent, or may include high-frequency words or any user selected words (e.g., a set of commonly used words in any language, not just English));
and generating each training canary transcription by randomly selecting the predetermined number of words from the top-N most frequent words sampled from the training text corpus ([0138] for each selected utterance, the existing stop words within the utterance are preserved and all of the non-stop words within the utterance are replaced with the stop word or the stop word phrase selected from the list of stop words. In other instances, the at least one non-stop word is one, two, three, four, or five of the non-stop words within the utterance. In other instances, for each selected utterance, the existing stop words within the utterance are preserved and at least one of the non-stop words within the utterance are randomly replaced with the stop word or the stop word phrase selected from the list of stop words. The list of stop words may be generated from a list of common words).
Huang, Rosenberg, and Vishnoi are considered analogous in the field of natural language processing. 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 teachings of Huang and Rosenberg to combine the teachings of Vishnoi because doing so would allow for training an encoder model using training data that represents out-of-distribution information, leading to improved detection of unintended memorization in an encoder model by providing relevant training examples (Vishnoi [0114] models and chatbots achieve a better accuracy using the SWAT because many utterances may include 20-50% stop words and the machine learning models may incorrectly pick up these stop word patterns as described above. By adding OOD sentences generated using the SWAT, the machine learning models may be prevented from picking up on the stop word patterns. The SWAT may also help to define the boundary of an intent. Another benefit is that a large number of utterances for the unresolved intent may help the machine learning models to perform better on OOD sentences).
Regarding claim 7, the combination of Huang and Rosenberg teaches: the method of claim 1. The combination of Huang and Rosenberg does not explicitly, but Vishnoi teaches: wherein obtaining the plurality of training canary transcriptions comprises: sampling, from a non-native language text corpus, the top-N most frequent words that appear in the non-native language text corpus ([0112] augmenting training data with stop words (e.g., non-stop words or phrases of non-stop words may be randomly replaced with one or more stop words or one or more phrases of stop words), in order to improve accuracy of the intent classification. Stop words typically refer to the most common words in a language, however, there is no single universal list of stop words used by all natural language processing tools. The stop words may include a list of words for a particular domain or intent, or may include high-frequency words or any user selected words (e.g., a set of commonly used words in any language, not just English));
and generating each training canary transcription by randomly selecting the predetermined number of words from the top-N most frequent words sampled from the non-native language text corpus ([0113] populates OOD training samples associated with an unresolved intent using existing in-domain training samples, where existing stop word patterns in the existing in-domain training samples are preserved, while non-stop words or phrases of non-stop words are randomly replaced with one or more stop words or one or more phrases of stop words. For example, a training sentence “do you offer a one day trial membership?” may be transformed into a training sample “do you to a one say a likely” associated with an unresolved intent. As such, the original stop words “do you . . . a one . . . ” are preserved and the non-stop words “offer . . . day trial membership” are replaced with random stop words “to . . . say a likely” selected from a list of words for the particular domain or intent, or include high-frequency words or any user selected words).
Regarding claim 15, it recites similar limitations to claim 5 and therefore is rejected similarly.
Regarding claim 17, it recites similar limitations to claim 7 and therefore is rejected similarly.
Claims 6, 8, 16, and 18 are rejected under 35 U.S.C. 103 as being unpatentable over Huang in view of Rosenberg, as applied to claims 1, 4, 9-11, 14, and 19-20 above, and further in view of Liu.
Regarding claim 6, the combination of Huang and Rosenberg teaches: the method of claim 1. The combination of Huang and Rosenberg does not explicitly, but Liu teaches: wherein obtaining the plurality of training canary transcriptions comprises generating each training canary transcription by selecting a non- repeating permutation of digits ([3.1.2] at each time step we only leverage the prediction output from one teacher rather than using all teachers’ aggregation, and alternate the choices of teachers over different steps. This can be performed through randomization or fixed permutation. In the randomization based teacher selection, for each step t, we randomly generate random(t) ∈ {1,2,...,M} and the corresponding teacher model is chosen as the supervisor. It aims to disconnect consecutive words in private sequences but generally has no issues in learning common and non-sensitive sequences that are present in majority of teachers. The randomized teacher selection happens in each step at every batch during training. A more restricted teacher selection strategy is through permutation but kept the chosen order fixed over the entire training process. Specifically, let π(M) be a random permutation of the sequence {1,2,...,M}).
Huang, Rosenberg, and Liu are considered analogous in the field of automatic speech recognition. 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 teachings of Huang and Rosenberg to combine the teachings of Liu because doing so would allow for the generation of out-of-distribution training canary speech utterances using different randomized permutation of digits/words, leading to improved unintended memorization detection and mitigation by using the randomly generated canaries as training examples (Liu [5] we propose the alternating teaching method to mitigate unintended memorization in sequential modeling. With experiments on LibriSpeech datasets, we show this approach achieves stronger mitigation than other counterparts and significantly reduces memorized sequences. Compared with the baselines without protections for memorizing private data, the overall quality of proposed method is not compromised when there exists sufficient training data).
Regarding claim 8, the combination of Huang and Rosenberg teaches: the method of claim 1. The combination of Huang and Rosenberg does not explicitly, but Liu teaches: wherein obtaining the plurality of training canary transcriptions comprises generating each training canary transcription as a non-verbal utterance ([4.2] For each user, a random 5-word canary is generated which simulates the “private sequence” from that specific user. No canary is shared by different users. Note that each word in any generated canary is among the vocabulary set of LMs).
Regarding claim 16, it recites similar limitations to claim 6 and therefore is rejected similarly.
Regarding claim 18, it recites similar limitations to claim 8 and therefore is rejected similarly.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure:
Dernoncourt et al. (US 12412038 B2) – discloses a system for training language models to preserve privacy by using a training corpus including sensitive and non-sensitive training samples to update model parameters.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to Nathan Tengbumroong whose telephone number is (703)756-1725. The examiner can normally be reached Monday - Friday, 11:30 am - 8:00 pm EST.
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, Hai Phan can be reached at 571-272-6338. 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.
/NATHAN TENGBUMROONG/Examiner, Art Unit 2654
/HAI PHAN/Supervisory Patent Examiner, Art Unit 2654
