DETAILED ACTION
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
Acknowledgment is made of the information disclosure statements filed on 3/13/2024. It is noted by the Examiner that all of the references were considered.
Priority
The instant application, filed 3/13/2024.
Drawings
The drawings submitted on 3/13/2024 have been considered and accepted.
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 1, 3-4, 8, 10-11, and 15, 17-18 is/are rejected under 35 U.S.C. 103 as being unpatentable over by Lee et al. (“Multi-SpectroGAN: High-Diversity and High-Fidelity Spectrogram Generation with Adversarial Style Combination for Speech Synthesis”) in view of Chen et al. (“FINE-GRAINED STYLE CONTROL IN TRANSFORMER-BASED TEXT-TO-SPEECH SYNTHESIS”)
Regarding Claim 1, Lee teaches:
A method of text to speech synthesis, the method comprising:
receiving, via a data interface, an input text, a first reference spectrogram, and a second reference spectrogram Figure 1 (a), Phoneme embedding (input text).(“Style Combination … … Figure 3 shows the mel-spectrograms and F0 contour (women, mixed and men style embedding) of GST(Left) and MSG (Right) model for the same sentence. … page 13204 , column 1, last paragraph”.
Lee further teaches:
generating, via a first encoder, a vector representation of the input text; Figure 1 (a), Phoneme embedding and encoder., Lee teaches (“Generator … We use the phoneme encoder …”Page 132000, column 1, para 2. Line 3.) by Lee et al. (“Multi-SpectroGAN: High-Diversity and High-Fidelity Spectrogram Generation with Adversarial Style Combination for Speech Synthesis”)
.generating, via a second encoder, a vector representation of the first reference spectrogram; Figure 1 and 3 Lee teaches women style embedding (i.e. first reference spectrogram) (“Extending to the multi-speaker model, we introduce a style encoder that can produce a fixed-dimensional style vector from a mel-spectrogram like Figure 1. …”) (“For the robustness of style transfer and control, we synthesize the mel-spectrogram with mixed style embedding which are interpolated style embedding of two speakers (1 male and 1 female). Figure 3 shows the mel-spectrograms and F0 contour (women, mixed and men style embedding) of GST (Left) and MSG (Right) model for the same sentence. …”) by Lee et al. (“Multi-SpectroGAN: High-Diversity and High-Fidelity Spectrogram Generation with Adversarial Style Combination for Speech Synthesis”)
generating, via a third encoder, a vector representation of the second reference spectrogram; Figure 1 and 3 Lee teaches male style embedding (i.e. second reference spectrogram) (“Extending to the multi-speaker model, we introduce a style encoder that can produce a fixed-dimensional style vector from a mel-spectrogram like Figure 1. …”) (“For the robustness of style transfer and control, we synthesize the mel-spectrogram with mixed style embedding which are interpolated style embedding of two speakers (1 male and 1 female). Figure 3 shows the mel-spectrograms and F0 contour (women, mixed and men style embedding) of GST (Left) and MSG (Right) model for the same sentence. …”) by Lee et al. (“Multi-SpectroGAN: High-Diversity and High-Fidelity Spectrogram Generation with Adversarial Style Combination for Speech Synthesis”)
generating a combined representation based on the vector representation of the first reference spectrogram and the vector representation of the second reference spectrogram; Figure 1-3 : Lee teaches the results demonstrate that the synthesis with the interpolated style embedding can generate a new style of mel-spectrogram by a combination of two styles (“which can learn latent representations of the combined style embedding from multiple mel-spectrograms. …” page 1, column 1, ) (“… an adversarial style combination which can learn latent representations of the combined speaker embedding from multiple mel-spectrograms. To learn the generated mel-spectrogram with randomly mixed styles which doesn’t have a ground truth mel-spectrogram, we propose an end-to-end learned frame-level conditional discriminator. …”page 13200 , column 1, first paragraph) (“The results demonstrate that the synthesis with the interpolated style embedding can generate a new style of mel-spectrogram by a combination of two styles. …” PAGE 13204 , Column 2, para 2) by Lee et al. (“Multi-SpectroGAN: High-Diversity and High-Fidelity Spectrogram Generation with Adversarial Style Combination for Speech Synthesis”)
generating, via a variance adaptor, a modified vector representation based on the vector representation of the input text; and Figure 1(a), generator, phoneme sequence (i.e input text). (“For details, we denote the phoneme-side FFT networks as phoneme encoder Ep(_), which produces the phoneme hidden representation …”) (“We use FastSpeech2 (Ren et al. 2020) as a generator consisting of a phoneme encoder with the variance adaptor denoted as f(_; _), and decoder g(_). We use the phoneme encoder and decoder which consists of 4 feed-forward Transformer (FFT) blocks. Extending to the multi-speaker model, we introduce a style encoder that can produce a fixed-dimensional style vector from a mel-spectrogram like Figure 1. … … We also use a gated recurrent unit (Cho et al. 2014) layer and take the final output to compress the length down to a single style vector. Before conditioning the length regulator and variance adaptor, the output is projected as the same dimension of the phoneme encoder output to add style information, followed by a tanh activation function. We denote the style encoder as Es(_), which produces the style embedding s = Es(y); (1) where s refers to the style embedding extracted from the mel-spectrogram y through the style encoder Es. Style-conditional Variance Adaptor With the exception of using style conditional information for learning the multi-speaker model, we use the same variance adaptor of Fast- Speech2 (Ren et al. 2020) to add variance information. … … By adding the style embedding predicted from the mel-spectrogram to the phoneme hidden sequence Hpho, the variance adaptor predicts each variance information with the unique style of each speaker. For details, we denote the phoneme-side FFT networks as phoneme encoder Ep(_), which produces the phoneme hidden representation” page 13200, entire column 1, Generator.) by Lee et al. (“Multi-SpectroGAN: High-Diversity and High-Fidelity Spectrogram Generation with Adversarial Style Combination for Speech Synthesis”)
generating, via a decoder, an audio waveform based on the modified vector representation and conditioned by the style vector. Figure 2. (a), (b), (c). and equation 19-23, Lee teaches The sum of each informational hidden sequence Htotal is passed to the decoder as a generator g(_) to generate a mel-spectrogram (i.e. an audio waveform). (“Style-conditional Variance Adaptor With the exception of using style conditional information for learning the multi-speaker model, we use the same variance adaptor of Fast- Speech2 (Ren et al. 2020) to add variance information. By adding the style embedding predicted from the mel-spectrogram to the phoneme hidden sequence Hpho, the variance adaptor predicts each variance information with the unique style of each speaker. For details, we denote the phoneme-side FFT networks as phoneme encoder Ep(_), which produces the phoneme hidden representation” page 13200, entire column 1) (“… The sum of each informational hidden sequence Htotal is passed to the decoder as a generator g(_) to generate a mel-spectrogram as …” page 13200, column 2 eqution 10) (“section Adversarial Style Combination … … the duration predictor may predict the wrong duration at the early training step. Each variance information is predicted by different ratios of mixed style embedding. We call it “style combination”, in which the final mixed hidden representation is the combination of each variance information from different mixed styles: …”) (“where pmix and emix are the pitch and energy embedding of the predicted value from mixed styles, respectively, and cmix is fed to discriminator as the frame-level condition for mel-spectrogram ^ymix generated by style combination. The discriminator is trained using the following objective: …” page 13201 -5) Note: Figure 2. (a), (b), (c). shows output is conditioned by discriminator as a frame level conditioned via decoder.”) by Lee et al. (“Multi-SpectroGAN: High-Diversity and High-Fidelity Spectrogram Generation with Adversarial Style Combination for Speech Synthesis”)
Lee does not explicitly teach generating a style vector based on cross attention between the combined representation and the vector representation of the input text.
Chen teaches:
generating a style vector based on cross attention between the combined representation and the vector representation of the input text; Fig.1, Chen teaches fusion between linguistic content (i.e. text) and speaking style (i.e. melspectrogram) by cross-attention module. prosody (i.e melspectrogram). we develop a cross-attention module to fuse linguistic content and style information into a mixed representation and also teaches prosody embeddings with a soft attention mechanism for the fusion between text and prosody. (“As fixed-length representation fails to model fine-grained styles, Lee et al. [8] extend the work to extract variable-length prosody embeddings with a soft attention mechanism for the fusion between text and prosody.”) (“Adding the style control with our
method does not degrade but improves the naturalness, suggesting suggesting
the success of the fusion between linguistic content and speaking style (i.e. mel-spectrogram) by cross-attention module. Compared to [10] which applied fine-grained style control on Tacotron2, our method outperforms in both naturalness and style similarity.” Page 4, column 2, lines 1-5 ) (“We conjecture that with our style network and content-style cross-attention module, the content representation is adjusted to contain the prosody information. …” page 4, column 1, lines 3-6) (“we develop a cross-attention module to fuse linguistic content and style information into a mixed representation.” Page 1, column 2, ) (“As fixed-length representation fails to model fine-grained styles, Lee et al. [8] extend the work to extract variable-length prosody embeddings with a soft attention mechanism for the fusion between text and prosody.”) by Chen (“FINE-GRAINED STYLE CONTROL IN TRANSFORMER-BASED TEXT-TO-SPEECH SYNTHESIS”)
Chen is considered to be analogous to the claimed invention because it relates to realize fine-grained style control on the transformer-based text-to-speech synthesis (TransformerTTS).
Therefore, it would have been obvious for someone of ordinary skill in the art before the effective filing date of the claimed invention to modify Lee to incorporate the teachings of Chen in order to include Vowel feature in the system.
One could have been motivated to do so because system can effectively scaled, thereby increasing the processing speed of the image generation system (i.e. MFCC is image) .(“
our system performs better in terms of naturalness, intelligibility, and style transferability.”1st page 1st column, 1st paragraph, last two lines. ..”) by Chen et al. (“FINE-GRAINED STYLE CONTROL IN TRANSFORMER-BASED TEXT-TO-SPEECH SYNTHESIS”)
Claim 8 is a system claim with a limitation similar to the limitation of method Claim 1 and is rejected under similar rationale. Additionally,
Lee further teaches:
8. A system for text to speech synthesis, the system comprising: a memory that stores a plurality of processor executable instructions; Lee teaches (“We used a Nvidia Titan V to train the single-speaker model with the LJ-speech dataset and the multi-speaker model with the VCTK dataset …” page 13202, column 1. Last paragraph. (“To convert the mel-spectrogram to audio, we use the pretrained PWG vocoder (Yamamoto, Song, and Kim 2020) consisting of 30-layers of dilated residual convolution blocks. …”) by by Lee et al. (“Multi-SpectroGAN: High-Diversity and High-Fidelity Spectrogram Generation with Adversarial Style Combination for Speech Synthesis”)
Claim 15 is a non-transitory machine-readable medium claim with a limitation similar to the limitation of method Claim 1 and is rejected under similar rationale.
Lee further teaches:
15. A non-transitory machine-readable medium comprising a plurality of machine-executable instructions which, when executed by one or more processors, are adapted to cause the one or more processors to perform operations comprising: Leet teaches a Nvidia Titan V (i.e. product). (“We used a Nvidia Titan V to train the single-speaker model with the LJ-speech dataset and the multi-speaker model with the VCTK dataset …” page 13202, column 1. Last paragraph. (“To convert the mel-spectrogram to audio, we use the pretrained PWG vocoder (Yamamoto, Song, and Kim 2020) consisting of 30-layers of dilated residual convolution blocks. …”) by by Lee et al. (“Multi-SpectroGAN: High-Diversity and High-Fidelity Spectrogram Generation with Adversarial Style Combination for Speech Synthesis”)Lee further teaches:
Regarding Claim 3, the combination teaches method claim 1 as identified above.
Chen further teaches:
3. The method of claim 1, wherein the cross attention is performed using: the combined representation for a key input and a value input, and the vector representation of the input text for a query input. Chen teaches (“Specifically, we replace the original content encoder blocks in Transformer TTS with cross-attention blocks. As shown in Figure 1, phone representations are first processed by a typical transformer encoder block. The output is then used as the query for a multi-head cross-attention module with style representation as both key and value. The resulting aligned style representation will be added back to the content representation with the residual design in transformer blocks. …” section 2.2, column 2, 2nd paragraph) by Chen et al. (“FINE-GRAINED STYLE CONTROL IN TRANSFORMER-BASED TEXT-TO-SPEECH SYNTHESIS”)
Chen is considered to be analogous to the claimed invention because it relates to realize fine-grained style control on the transformer-based text-to-speech synthesis (TransformerTTS).
Therefore, it would have been obvious for someone of ordinary skill in the art before the effective filing date of the claimed invention to modify Leeto incorporate the teachings of Chen in order to include Vowel feature in the system.
One could have been motivated to do so because system can effectively scaled, thereby increasing the processing speed of the image generation system (i.e. MFCC is image) .(“
our system performs better in terms of naturalness, intelligibility, and style transferability.”1st page 1st column, 1st paragraph, last two lines. ..”) by Chen et al. (“FINE-GRAINED STYLE CONTROL IN TRANSFORMER-BASED TEXT-TO-SPEECH SYNTHESIS”)
Claim 10 is a system claim with a limitation similar to the limitation of method Claim 3 and is rejected under similar rationale.
Claim 17 is a non-transitory machine-readable medium claim with a limitation similar to the limitation of method Claim 3 and is rejected under similar rationale.
Regarding Claim 4, the combination teaches method claim 1 as identified above.
Chen further teaches
4. The method of claim 1, wherein the decoder includes at least one multi-head attention layer and at least one conditional layer normalization layer. Fig. 1, Chen teaches (“We decompose our system into four sub-modules: content network, style network, decoder network, and the content-style cross-attention blocks. The decoder network follows an architecture similar to MultiSpeech [14], while the content network is a simplified version containing only phoneme embedding and layer normalization. Here we focus on the design of the style network and content-style cross-attention blocks.” Page 2, method section, first paragraph) (“The output is then used as the query for a multi-head cross-attention module with style representation as both key and value. …” page 2, section 2.2, 3rd para, column 2)(“speaker embedding … …To obtain the speaker embedding, the query vector will be fed to a multi-head attention [17], with a trainable codebook of global style tokens as both key and value.” Page 2, column 1, section speaker embedding) (“Each time step of the output will be fed as a query to a multi-head attention with another trainable codebook as key and value to produce a sequence of preliminary style embeddings. …”) by Chen et al. (“FINE-GRAINED STYLE CONTROL IN TRANSFORMER-BASED TEXT-TO-SPEECH SYNTHESIS”)
Chen is considered to be analogous to the claimed invention because it relates to realize fine-grained style control on the transformer-based text-to-speech synthesis (TransformerTTS).
Therefore, it would have been obvious for someone of ordinary skill in the art before the effective filing date of the claimed invention to modify Lee to incorporate the teachings of Chen in order to include Vowel feature in the system.
One could have been motivated to do so because system can effectively scaled, thereby increasing the processing speed of the image generation system (i.e. MFCC is image) .(“
our system performs better in terms of naturalness, intelligibility, and style transferability.”1st page 1st column, 1st paragraph, last two lines. ..”) by Chen et al. (“FINE-GRAINED STYLE CONTROL IN TRANSFORMER-BASED TEXT-TO-SPEECH SYNTHESIS”)
Claim 11 is a system claim with a limitation similar to the limitation of method Claim 4 and is rejected under similar rationale.
Claim 18 is a non-transitory machine-readable medium claim with a limitation similar to the limitation of method Claim 4 and is rejected under similar rationale.
Claim 5, 12, and 19, is/are rejected under 35 U.S.C. 103 as being unpatentable over Lee and Chen, and further in view of Xue et al. (“Improving Speech Recognition with Augmented Synthesized Data and Conditional Model Training”)
Regarding Claim 5, the combination teaches method claim 4 as identified above.
Lee further teaches (, we introduce a style encoder that can produce a fixed-dimensional
style vector from a mel-spectrogram like Figure 1…. … Before conditioning the length regulator and variance adaptor, the output is projected as the same dimension of the phoneme encoder output to add style information, followed by a tanh activation function. We denote the style encoder as Es(_), which produces the style embedding … … where s refers to the style embedding extracted from the mel-spectrogram y through the style encoder Es. Style-conditional Variance Adaptor With the exception of using style conditional information for learning the multi-speaker model, we use the same variance adaptor of Fast-Speech2 (Ren et al. 2020) to add variance information….”) by Chen et al. (“FINE-GRAINED STYLE CONTROL IN TRANSFORMER-BASED TEXT-TO-SPEECH SYNTHESIS”)
The combination does not explicitly teach at least one conditional layer normalization layer is conditioned based on the style vector.
Xue teaches
5. The method of claim 4, wherein the at least one conditional layer normalization layer is conditioned based on the style vector. Xue teaches the conditional network consists of two simple linear layers Wlα and Wlβ that take condition embedding Ec as input (i.e. style vector). (“(“Fig.2. In the added conditional block of each CRL, the condition embedding vectors of real and synthetic data are used as the the auxiliary input. It is transferred by a linear layer, a layer normalization and a sigmoid function to produce condition vectors for this block. The input is passed through a linear layer and multiplied with the condition vectors. …” page 3 , column 2, 1st paragraph) (“section 3.2.1 TTS [14] and NLP [19] tasks. Since layer normalization is adopted in each self-attention and feed-forward network in the Transformer block, it can greatly influence the hidden activation and output prediction with a learnable scale vector α and bias vector β: … … The conditional network consists of two simple linear layers Wlα and Wlβ that take condition embedding Ec as input and output the scale and bias vector respectively: …” section 3.2.1, page 3, column 1, 3rd. para.) by Xue et al. (“Improving Speech Recognition with Augmented Synthesized Data and Conditional Model Training”)
Xue is considered to be analogous to the claimed invention because it relates to speech recognition systems using synthetic data.
Therefore, it would have been obvious for someone of ordinary skill in the art before the effective filing date of the claimed invention to modify Lee and Chen to incorporate the teachings of Xue in order to include multi head attention layer in the system.
One could have been motivated to do so because system will achieve best validation accuracy .(“ … Each transformer layer uses 256 attention dimension and 2048 feed forward dimension. The dimension of conditional embedding layer is set to 256. The model checkpoint of each epoch is saved, and the final model is produced by averaging the 10 checkpoints with the best validation accuracy....”) by Xue et al. (“Improving Speech Recognition with Augmented Synthesized Data and Conditional Model Training”)
Claim 12 is a system claim with a limitation similar to the limitation of method Claim 5 and is rejected under similar rationale.
Claim 19 is a non-transitory machine-readable medium claim with a limitation similar to the limitation of method Claim 5 and is rejected under similar rationale.
Claim 7 and 14 is/are rejected under 35 U.S.C. 103 as being unpatentable over Lee and Chen and further in view of Mingjian Chen Mingjian Chen (“ADASPEECH: ADAPTIVE TEXT TO SPEECH FOR CUSTOM VOICE”) {IDS provided}
Regarding Claim 7, the combination teaches method claim 1 as identified above.
The combination does not explicitly teach updating parameters.
Mingjian Chen teaches:
7. The method of claim 1, further comprising: updating parameters of at least one of the first encoder, the second encoder, the third encoder, the variance adaptor, the cross attention, or the decoder via backpropagation based on a comparison of the audio waveform to a ground truth audio waveform Mingjian Chen teaches decoder parameter fine tuned. (“Abstract To better trade off the adaptation parameters and voice quality, we introduce conditional layer normalization in the mel-spectrogram decoder of AdaSpeech, and fine-tune this part in addition to speaker embedding for adaptation. …” page 1) (“In fine-tuning, we only adapt the parameters related to the conditional layer normalization. In this way, we can greatly reduce adaptation parameters and thus memory storage2 compared with fine-tuning the whole model, but maintain high-quality adaptation voice thanks to the flexibility of conditional layer normalization. …” page 2, 2nd paragraph from bottom page) (“We compare AdaSpeech with several settings: 1) GT, the ground-truth recordings; 2) GT mel + Vocoder, using ground-truth mel-spectrogram to synthesize waveform with MelGAN vocoder; 3) Baseline (spk emb), a baseline system based on FastSpeech2 which only fine-tunes the speaker embedding during adaptation, and can be regarded as our lower bound; 4) Baseline (decoder), another baseline system based on FastSpeech2 which fine-tunes the whole decoder during adaptation, and can be regarded as a strong comparable system since it uses more parameters during adaptation; 5) AdaSpeech, our proposed AdaSpeech system with utterance-/phoneme-level acoustic condition modeling and conditional layer normalization during adaptation6.” Page 6, Section 4.1, last paragraph) by Chen et al. (“ADASPEECH: ADAPTIVE TEXT TO SPEECH FOR CUSTOM VOICE” )
Mingjian Chen is considered to be analogous to the claimed invention because it relates to performing speech recognition.
Therefore, it would have been obvious for someone of ordinary skill in the art before the effective filing date of the claimed invention to modify Lee and Chen to incorporate the teachings of Mingjian Chen in order to include updating parameter feature.
One could have been motivated to do so because system will be more adaptable to different adaptation data..(“ …The key contribution in our acoustic condition modeling in this work is the novel perspective to model the diverse acoustic conditions in different granularities to make the source model more adaptable to different adaptation data.” Page 4, 3rd paragraph) by Mingjian Chen (“ADASPEECH: ADAPTIVE TEXT TO SPEECH FOR CUSTOM VOICE”)
Claim 14 is a system claim with a limitation similar to the limitation of method Claim 7 and is rejected under similar rationale.
Allowable Subject Matter
Claims 2, 6, 9, 13, 16, and 20 are objected to as being dependent upon a rejected base claim, but would be allowable if rewritten in independent form, e.g. contained within independent Claim 1, independent Claim 8, independent Claim 15 respectively, including all of the limitations of the base claim and any intervening claims. Such amendments should further consider having all independent base claims the same in scope, with the exception of the embodiments. The combination of the cited references does not reasonably teach or suggest the pending claim features.
Claims 6, is/are considered allowable, since certain key features of the claimed invention are not taught or fairly suggested by the prior art, particularly in ordered combination with the rest of the limitations recited within each of the independent claims.
The primary reason for considering Claim 4 is the combination of limitations recited in Claim 6, “… generating, via second linear layer, a bias vector based on the style vector, wherein the at least one conditional layer normalization layer modifies a mean of an input of the at least one conditional layer normalization layer based on the bias vector, and wherein the at least one conditional layer normalization layer modifies a variance of an input of the at least one conditional layer normalization layer based on the scale vector.”
No combination of arts teaches all of the cited limitations, in combination with the rest of the limitations recited in independent Claim 1 in a way that would have been obvious to one of ordinary skill in the art at the time the invention was effectively filed.
Claim 9 is a system claim with a limitation similar to the limitation of method Claim 2 and is objected under similar rationale.
Claim 13 is a system claim with a limitation similar to the limitation of method Claim 6 and is objected under similar rationale.
Claim 16 is a non-transitory machine-readable medium claim with a limitation similar to the limitation of method Claim 2 and is objected under similar rationale.
Claim 20 is a non-transitory machine-readable medium claim with a limitation similar to the limitation of method Claim 6 and is objected under similar rationale.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to FOUZIA HYE SOLAIMAN whose telephone number is (571)270-5656. The examiner can normally be reached M-F (8-5)AM.
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, Paras D. Shah can be reached at (571) 270-1650. 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.
/F.H.S./Examiner, Art Unit 2653
/Paras D Shah/Supervisory Patent Examiner, Art Unit 2653
04/28/2026
Prosecution Insights