DETAILED ACTION
This office action is in response to Applicant’s Amendment/Request for Reconsideration, received on 07/29/2026. Claims 1, 6, 11, and 16 have been amended. Claims 1-20 are pending and have been considered.
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 .
Response to Arguments
Applicant’s arguments, see pgs. 7-9, filed 07/29/2026, with respect to the rejection(s) of claim(s) 1, 5, 9, 10-11, 15, 19-20 under 35 U.S.C. 103 (Xu in view of Mao, further in view of Li) have been fully considered and are persuasive. Therefore, the rejection has been withdrawn. However, upon further consideration, a new ground(s) of rejection is made in view of Ye et al. (US-20180144742-A1), hereinafter Ye. Ye discloses “A method and an apparatus for processing voice data. A specific implementation of the method includes: receiving voice data sent by a user terminal; extracting a voiceprint characteristic vector in the voice data; matching the voiceprint characteristic vector with a registered voiceprint vector prestored by the user, and generating a matching degree between the voiceprint characteristic vector and the registered voiceprint vector; determining whether the matching degree is greater than or equal to a preset update threshold; and updating the registered voiceprint vector by using the voiceprint characteristic vector and the voice data in response to determining that the matching degree is greater than or equal to the preset update threshold” (abstract). Ye is being incorporate in replacement of Li. See updated rejections below.
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claim(s) 1, 5, 9, 10-11, 15, 19-20 is/are rejected under 35 U.S.C. 103 as being unpatentable over Xu et al. (US-20240071392-A1), hereinafter Xu, in view of Mao et al. (US-20210183358-A1), hereinafter Mao, further in view of Ye et al. (US-20180144742-A1), hereinafter Ye.
Regarding claim 1, Xu discloses: A computer-implemented method ([0143] an embodiment of this application further provides a computer program product including instructions) comprising:
at an enrollment ([0042] a registration procedure, [Registration and enrollment are synonymous terms in the context of speaker verification]):
Xu does not disclose:
obtaining a plurality of enrollment embeddings.
Mao discloses:
obtaining a plurality of enrollment embeddings ([0023] The user device 110a may process (132) the audio data using a first neural network to determine first embedding data representing characteristics of the utterance. This first embedding data may include a data vector that represents vocal characteristics of the voice of the user 10, [Consider “enrollment” audio data of [0045] of Mao]).
Xu and Mao are considered analogous art within audio signal conversion. Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the teachings of Xu to incorporate the teachings of Mao, because of the novel way to convert audio data into formats capable of being sent/handled by (Mao, [0001]).
Xu further discloses:
obtaining, by a computer ([0054] the electronic device includes, but is not limited to, a mobile phone, a tablet computer), a plurality of enrollment embeddings extracted using a plurality of enrollment signals for an enrolled user by applying a first embedding extractor comprising a first plurality of machine-learning layers trained to generate the plurality of enrollment embeddings for a first attribute-type corresponding to a first type of machine learning architecture ([Fig. 3, S302, “The electronic device processes the first registration voice by using a pre-stored first model, to obtain a first user feature template…”], [0042] a voiceprint recognition system acquires a registration voice entered by the user, extracts a voiceprint feature from the registration voice according to a pre-trained depth model (referred to as a “voiceprint feature extraction model” or “model” in this specification), and stores the voiceprint feature in an electronic device as a user feature template, [In view of the previously disclosed plurality of embeddings representing vocal characteristics of Mao, features extracted from a voice to form a voiceprint indicates the features of Xu to be representative of embeddings. Further, a pre-stored first model tracks to a first type of machine learning architecture with a first attribute-type corresponding to the first feature template. The first user feature template is the plurality of enrollment embeddings, see [0103] defining feature vectors of user template features. Processing using a model suggests the model to be comprised of layers]).
Xu in view of Mao does not disclose:
generating, by the computer, a plurality of converted embeddings corresponding to the plurality of enrollment embeddings by
applying, to each enrollment embedding of the plurality of enrollment embeddings, an embedding convertor comprising a second plurality of machine-learning layers trained to map the plurality of enrollment embeddings from a first feature vector space for the first type of machine-learning architecture to a second feature vector space for a second type of machine-learning architecture and
generate a converted embedding having a second attribute-type corresponding to the second type of machine learning architecture,
the converted embedding generated for an enrollment embedding having the first attribute-type; and
generating, by the computer, a converted enrolled voiceprint having the second attribute-type for the enrolled user based upon the plurality of converted embeddings.
Ye discloses:
generating, by the computer ([In view of the previously disclosed computer of Xu]), a plurality of converted embeddings corresponding to the plurality of enrollment embeddings by
applying, to each enrollment embedding of the plurality of enrollment embeddings ([Fig. 3, S301 “Acquiring a preset amount of registered voice data”], [Wherein a preset amount which corresponds to multiple vectors (as seen below) indicates a plurality of registered, i.e. enrollment, data/embeddings]), an embedding convertor comprising a second plurality of machine-learning layers trained to map the plurality of enrollment embeddings from a first feature vector space for the first type of machine-learning architecture to a second feature vector space for a second type of machine-learning architecture ([Fig. 3, S302 “Importing the preset amount of registered voice data into a pre-trained universal background model for mapping to obtain a preset amount of equal-length intermediate vectors”, S303 “Perform a characteristic transform on the preset number of equal length intermediate vectors to obtain a preset number of registered voiceprint characteristic vectors”], [Fig. 4, S406 “Carrying out a data standardization operation on the each of voiceprint characteristic vectors prestored by the user…”], [0050] Characteristic transformation may be carried out on the intermediate vector by means of Joint Factor Analysis (JFA) to obtain a voiceprint characteristic vector… Alternatively, characteristic transformation may be carried out on the intermediate vector by means of Probabilistic Linear Discriminant Analysis (PLDA) to obtain the voiceprint characteristic vector. The PLDA also is a channel compensation algorithm and is a Linear Discriminant Analysis (LDA) in the form of probability. Furthermore, characteristic transformation may be carried out on the intermediate vector by means of an Identifying Vector (I-Vector) to obtain the voiceprint characteristic vector, [Wherein the intermediate vectors tracks to enrollment embeddings output by a “pre-trained universal background model”, indicating the intermediate vectors to have a first feature vector space and be output from a first-type of machine learning architecture (see [0068] disclose that the universal background model is a GMM). Further, wherein performing a characteristic transformation by means of an JFA and/or PDLA to compensate for channel factors suggests a second feature vector for a second type of machine-learning architecture corresponding to either of the transformation means (the standardization operation of Fig. 4 could also be an embedding conversion to a second vector space, wherein the compensation is performed based on PDLA/JFA, indicating a second vector space with a second type of machine learning architecture to account for the normalized vectors). It would be obvious to apply the voiceprint feature extraction neural network model of Xu and/or Mao (a second type of machine learning architecture) as the feature transformation of Ye because Xu and Mao discloses a PLDA model for voiceprint feature vector calculation (Xu, [0103], Mao, [0083]), indicating the neural network of Xu/Mao could be used in the PLDA feature transformation of Ye without a change in functionality to Ye. Further still, a plurality of intermediate vectors and/or registered voiceprint characteristic vectors indicates a plurality of enrollment embeddings]) and
generate a converted embedding having a second attribute-type corresponding to the second type of machine learning architecture ([As previously cited, the voiceprint characteristic vector is the converted embedding having the second attribute type corresponding to a second type of machine learning architecture after normalization, channel compensation, etc.]),
the converted embedding generated for an enrollment embedding having the first attribute-type ([As can be seen through the previously cited portions of Fig. 3, a converted embedding, i.e. registered voiceprint characteristic vector, is generated for an enrollment embedding having a first attribute type, i.e. equal-length intermediate vectors mapped based on registered, e.g. enrollment, voice data]); and
generating, by the computer ([In view of the previously disclosed computer of Xu]), a converted enrolled voiceprint having the second attribute-type for the enrolled user based upon the plurality of converted embeddings ([Fig. 3, S304 “Fusing the preset number of registered voiceprint characteristic vectors to generate the registered voiceprint vector”], [The examiner asserts that the fused, registered voiceprint vector is a converted enrolled voiceprint having a second-attribute type]).
Xu, Mao, and Ye are considered analogous art within speaker identity verification. Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the teachings of Xu in view of Mao to incorporate the teachings of Ye, because of the novel way to adaptively update voiceprint models of users with newly added voice samples, enhancing the functionalities of a voiceprint recognition system over time as one’s vocal features may change (Ye, [0003]-[0005]).
Xu further discloses:
at a deployment ([0042] In the verification procedure, [In the context of speech/speaker recognition, deployment and verification are synonymous terms]):
generating, by the computer ([In view of the previously disclosed computer of Xu]), an inbound voiceprint for an inbound user extracted using an inbound signal for an inbound user by applying a second embedding extractor for the second attribute-type ([0105] the electronic device processes the first verification voice by using the second model, to obtain a second voiceprint feature); and,
generating, by the computer ([In view of the previously disclosed computer of Xu]), a similarity score for the inbound signal based upon a distance between the converted enrolled voiceprint and the inbound voiceprint ([0103] For example, in cosine distance model scoring, a cosine value between a feature vector of a to-be-verified first voiceprint feature and a feature vector of the user template feature is calculated, and the cosine value is used as a similarity score in other words, a scoring result)), the similarity score indicating a likelihood that the inbound user is the enrolled user ([0050] obtain a verification result, [A verification result tracks to a decision as to whether or not the two voiceprints verified to be from the same speaker]).
Regarding claim 5, Xu in view of Mao, further in view of Ye discloses: The method according to claim 1.
Mao further discloses:
training a plurality of embedding convertors according to a plurality of attribute-types ([0031] As explained in greater detail with reference to FIG. 3, the feature conversion component 208 may be trained such that it outputs sufficiently different converted embedding data 210 given embedding data 206 that corresponds to different users [Outputting data corresponding to a plurality of users indicates those users have different embedding features, i.e. voice features, which track to attribute type in view of [0040] of the instant application, i.e. other attributes (specific voice/frequency features) of signals]).
Regarding claim 9, Xu in view of Mao, further in view of Ye discloses: the method according to claim 1,
Mao further discloses:
wherein generating the converted enrolled voiceprint having the second attribute-type includes algorithmically combining, by the computer, the converted enrollment embeddings having the second attribute-type ([0076] the feature extraction component 204 may accumulate or otherwise combine the audio data 202 as it comes in. That is, for a certain frame's worth of audio data 202 that comes in, the feature extraction component 204 may combine that frame's worth of data to the previous data received for the particular utterance).
Regarding claim 10, Xu in view of Mao, further in view of Ye discloses: the method according to claim 1.
Mao further discloses:
extracting, by the computer, a set of enrollment features from an enrollment signal ([Fig. 2A, 204], [0030] The feature extraction component 204 may be trained to process the audio data 202 and output corresponding embedding data 206, which may be a vector of values representing audio characteristics of the audio data 202).
extracting, by the computer, an enrollment embedding based upon the set of features extracted from the enrollment signal by applying the first embedding extractor for the first attribute-type ([Fig. 2A, 206], [0030] In other words, the feature extraction component 204 extracts features associated with the voice type associated with the audio data 202 but not necessarily the particular words represented in the audio data 202 [i.e. they are extracting embeddings of features for an input signal 202 having a first attribute-type]).
Regarding claim 11, Xu discloses:
a system ([0042] the voiceprint recognition system) comprising:
a non-transitory machine-readable memory configured to store machine-readable instructions for one or more neural networks ([0056] The mobile phone 100 includes a processor 110, an internal memory 121, [Internal memory on a physical device will be non-transitory]); and,
a computer comprising a processor ([0148] These computer program instructions may be provided for a general-purpose computer, a dedicated computer, an embedded processor) configured to:
at an enrollment ([0042] a registration procedure, [Registration and enrollment are synonymous terms in the context of speaker verification]):
Xu does not disclose:
obtain a plurality of enrollment embeddings.
Mao discloses:
obtain a plurality of enrollment embeddings ([0023] The user device 110a may process (132) the audio data using a first neural network to determine first embedding data representing characteristics of the utterance. This first embedding data may include a data vector that represents vocal characteristics of the voice of the user 10, [Consider “enrollment” audio data of [0045] of Mao]).
Xu and Mao are considered analogous art within audio signal conversion. Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the teachings of Xu to incorporate the teachings of Mao, because of the novel way to convert audio data into formats capable of being sent/handled by (Mao, [0001]).
Xu further discloses:
obtain a plurality of enrollment embeddings extracted using a plurality of enrollment signals for an enrolled user by applying a first embedding extractor comprising a first plurality of machine-learning layers trained to generate the plurality of enrollment embeddings for a first attribute-type corresponding to a first type of machine learning architecture ([Fig. 3, S302, “The electronic device processes the first registration voice by using a pre-stored first model, to obtain a first user feature template…”], [0042] a voiceprint recognition system acquires a registration voice entered by the user, extracts a voiceprint feature from the registration voice according to a pre-trained depth model (referred to as a “voiceprint feature extraction model” or “model” in this specification), and stores the voiceprint feature in an electronic device as a user feature template, [In view of the previously disclosed plurality of embeddings representing vocal characteristics of Mao, features extracted from a voice to form a voiceprint indicates the features of Xu to be representative of embeddings. Further, a pre-stored first model tracks to a first type of machine learning architecture with a first attribute-type corresponding to the first feature template. The first user feature template is the plurality of enrollment embeddings, see [0103] defining feature vectors of user template features. Processing using a model suggests the model to be comprised of layers]).
Xu in view of Mao does not disclose:
generate a plurality of converted embeddings corresponding to the plurality of enrollment embeddings by
applying, to each enrollment embedding of the plurality of enrollment embeddings, an embedding convertor comprising a second plurality of machine-learning layers trained to map the plurality of enrollment embeddings from a first feature vector space for the first type of machine-learning architecture to a second feature vector space for a second type of machine-learning architecture and
generate a converted embedding having a second attribute-type corresponding to the second type of machine learning architecture,
the converted embedding generated for an enrollment embedding having the first attribute-type; and
generate a converted enrolled voiceprint having the second attribute-type for the enrolled user based upon the plurality of converted embeddings.
Ye discloses:
generate a plurality of converted embeddings corresponding to the plurality of enrollment embeddings by
applying, to each enrollment embedding of the plurality of enrollment embeddings ([Fig. 3, S301 “Acquiring a preset amount of registered voice data”], [Wherein a preset amount which corresponds to multiple vectors (as seen below) indicates a plurality of registered, i.e. enrollment, data/embeddings]), an embedding convertor comprising a second plurality of machine-learning layers trained to map the plurality of enrollment embeddings from a first feature vector space for the first type of machine-learning architecture to a second feature vector space for a second type of machine-learning architecture ([Fig. 3, S302 “Importing the preset amount of registered voice data into a pre-trained universal background model for mapping to obtain a preset amount of equal-length intermediate vectors”, S303 “Perform a characteristic transform on the preset number of equal length intermediate vectors to obtain a preset number of registered voiceprint characteristic vectors”], [Fig. 4, S406 “Carrying out a data standardization operation on the each of voiceprint characteristic vectors prestored by the user…”], [0050] Characteristic transformation may be carried out on the intermediate vector by means of Joint Factor Analysis (JFA) to obtain a voiceprint characteristic vector… Alternatively, characteristic transformation may be carried out on the intermediate vector by means of Probabilistic Linear Discriminant Analysis (PLDA) to obtain the voiceprint characteristic vector. The PLDA also is a channel compensation algorithm and is a Linear Discriminant Analysis (LDA) in the form of probability. Furthermore, characteristic transformation may be carried out on the intermediate vector by means of an Identifying Vector (I-Vector) to obtain the voiceprint characteristic vector, [Wherein the intermediate vectors tracks to enrollment embeddings output by a “pre-trained universal background model”, indicating the intermediate vectors to have a first feature vector space and be output from a first-type of machine learning architecture (see [0068] disclose that the universal background model is a GMM). Further, wherein performing a characteristic transformation by means of an JFA and/or PDLA to compensate for channel factors suggests a second feature vector for a second type of machine-learning architecture corresponding to either of the transformation means (the standardization operation of Fig. 4 could also be an embedding conversion to a second vector space, wherein the compensation is performed based on PDLA/JFA, indicating a second vector space with a second type of machine learning architecture to account for the normalized vectors). It would be obvious to apply the voiceprint feature extraction neural network model of Xu and/or Mao (a second type of machine learning architecture) as the feature transformation of Ye because Xu and Mao discloses a PLDA model for voiceprint feature vector calculation (Xu, [0103], Mao, [0083]), indicating the neural network of Xu/Mao could be used in the PLDA feature transformation of Ye without a change in functionality to Ye. Further still, a plurality of intermediate vectors and/or registered voiceprint characteristic vectors indicates a plurality of enrollment embeddings]) and
generate a converted embedding having a second attribute-type corresponding to the second type of machine learning architecture ([As previously cited, the voiceprint characteristic vector is the converted embedding having the second attribute type corresponding to a second type of machine learning architecture after normalization, channel compensation, etc.]),
the converted embedding generated for an enrollment embedding having the first attribute-type ([As can be seen through the previously cited portions of Fig. 3, a converted embedding, i.e. registered voiceprint characteristic vector, is generated for an enrollment embedding having a first attribute type, i.e. equal-length intermediate vectors mapped based on registered, e.g. enrollment, voice data]); and
generate a converted enrolled voiceprint having the second attribute-type for the enrolled user based upon the plurality of converted embeddings ([Fig. 3, S304 “Fusing the preset number of registered voiceprint characteristic vectors to generate the registered voiceprint vector”], [The examiner asserts that the fused, registered voiceprint vector is a converted enrolled voiceprint having a second-attribute type]).
Xu, Mao, and Ye are considered analogous art within speaker identity verification. Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the teachings of Xu in view of Mao to incorporate the teachings of Ye, because of the novel way to adaptively update voiceprint models of users with newly added voice samples, enhancing the functionalities of a voiceprint recognition system over time as one’s vocal features may change (Ye, [0003]-[0005]).
Xu further discloses:
at a deployment ([0042] In the verification procedure, [In the context of speech/speaker recognition, deployment and verification are synonymous terms]):
generate an inbound voiceprint for an inbound user extracted using an inbound signal for an inbound user by applying a second embedding extractor for the second attribute-type ([0105] the electronic device processes the first verification voice by using the second model, to obtain a second voiceprint feature); and,
generate a similarity score for the inbound signal based upon a distance between the converted enrolled voiceprint and the inbound voiceprint ([0103] For example, in cosine distance model scoring, a cosine value between a feature vector of a to-be-verified first voiceprint feature and a feature vector of the user template feature is calculated, and the cosine value is used as a similarity score in other words, a scoring result)), the similarity score indicating a likelihood that the inbound user is the enrolled user ([0050] obtain a verification result, [A verification result tracks to a decision as to whether or not the two voiceprints verified to be from the same speaker]).
Regarding claim 15, Xu in view of Mao, further in view of Ye discloses the system according to claim 11.
Mao further discloses:
train a plurality of embedding convertors according to a plurality of attribute-types ([0031] As explained in greater detail with reference to FIG. 3, the feature conversion component 208 may be trained such that it outputs sufficiently different converted embedding data 210 given embedding data 206 that corresponds to different users [Outputting data corresponding to a plurality of users indicates those users have different embedding features, i.e. voice features, which track to attribute types in view of [0040] of the instant application, i.e. other attributes (specific voice/frequency features) of signals]).
Regarding claim 19, Xu in view of Mao, further in view of Ye discloses: the system according to claim 11.
Mao further discloses:
wherein when generating the converted enrolled voiceprint having the second attribute-type the computer is further configured to store the converted enrolled voiceprint into a user profile database ([0076] the feature extraction component 204 may accumulate or otherwise combine the audio data 202 as it comes in. That is, for a certain frame's worth of audio data 202 that comes in, the feature extraction component 204 may combine that frame's worth of data to the previous data received for the particular utterance).
Regarding claim 20, Xu in view of Mao, further in view of Ye discloses: the system according to claim 11.
Mao further discloses:
extract a set of enrollment features from an enrollment signal ([Fig. 2A, 204], [0030] The feature extraction component 204 may be trained to process the audio data 202 and output corresponding embedding data 206, which may be a vector of values representing audio characteristics of the audio data 202).
extract an enrollment embedding based upon the set of features extracted from the enrollment signal by applying the first embedding extractor for the first attribute-type ([Fig. 2A, 206], [0030] In other words, the feature extraction component 204 extracts features associated with the voice type associated with the audio data 202 but not necessarily the particular words represented in the audio data 202 [i.e. they are extracting embeddings of features for an input signal 202 having a first attribute-type]).
Claim(s) 2, 4, 8, 12, 14, 18 is/are rejected under 35 U.S.C. 103 as being unpatentable over Xu in view of Mao, further in view of Ye, further in view of Thomson et al. (US-20200175961-A1), hereinafter Thomson.
Regarding claim 2, Xu in view of Mao, further in view of Ye discloses the method according to claim 1.
Xu in view of Mao, further in view of Ye does not disclose:
obtaining, by the computer, a plurality of training embeddings extracted using a plurality of training signals by applying the first embedding extractor for the first attribute-type; and,
training, by the computer, the embedding convertor by applying the second plurality of machine-learning layers of the embedding convertor on the plurality of training embeddings.
Thomson discloses:
obtaining, by the computer, a plurality of training embeddings extracted using a plurality of training signals by applying the first embedding extractor for the first attribute-type ([1023] The training set 5810 may be provided to feature extractors 5840).
training, by the computer, the embedding convertor by applying the second plurality of machine-learning layers of the embedding convertor on the plurality of training embeddings ([1023] The feature extractors 5840 may be configured to determine features such as n-grams or word embeddings of the training set 5810 that may be provided to the model trainer).
Xu, Mao are considered analogous art within speaker embedding conversion. Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the teachings of Xu in view of Mao, further in view of Ye to incorporate the teachings of Thomson, because of the novel way to incorporate a plurality of training voice profiles using live audio for improved automatic speech recognition system accuracy (Thomson, [0096]).
Regarding claim 4, Xu in view of Mao, further in view of Ye, further in view of Thomson discloses the method according to claim 2.
Thomson further discloses:
training the embedding extractor includes executing, by the computer, one or more data augmentation operations on at least of a training audio signal and an enrollment signal ([0766] Additionally or alternatively, audio in a first format may be converted to a second format and presented to an ASR system 3720 configured for the second format. For example, wideband audio may be downsampled to 8 kHz and processed by an ASR system 3720 configured to recognize 8 kHz speech. [Down-sampling is listed as one of a non-limiting example of a augmentation operation according to the spec, [0042]]).
Regarding claim 8, Xu in view of Mao, further in view of Ye discloses the method according to claim 1.
Xu in view of Mao, further in view of Ye does not disclose:
wherein generating the converted enrolled voiceprint having the second attribute-type includes storing, by the computer, the converted enrolled voiceprint into a user profile database.
Thomson discloses:
wherein generating the converted enrolled voiceprint having the second attribute-type includes storing, by the computer, the converted enrolled voiceprint into a user profile database ([0163] In some embodiments, the configuration service may include a business server, a user profile system, and a subscription management system. The configuration service may store information on the individual devices or on a server [Storing user profiles containing voice information ([0163]) in view of the voice prints of Mao, indicates that a converted enrolled voiceprint having the second attribute type is stored in a user profile]).
Xu, Mao are considered analogous art within speaker embedding conversion. Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the teachings of Xu in view of Mao, further in view of Ye to incorporate the teachings of Thomson, because of the novel way to incorporate a plurality of training voice profiles using live audio for improved automatic speech recognition system accuracy (Thomson, [0096]).
Regarding claim 12, Xu in view of Mao, further in view of Ye discloses: the system according to claim 11.
Xu in view of Mao, further in view of Ye does not disclose:
obtain, by the computer, a plurality of training embeddings extracted using a plurality of training signals by applying the first embedding extractor for the first attribute-type; and,
train, by the computer, the embedding convertor by applying the second plurality of machine-learning layers of the embedding convertor on the plurality of training embeddings.
Thomson discloses:
obtain, by the computer, a plurality of training embeddings extracted using a plurality of training signals by applying the first embedding extractor for the first attribute-type ([1023] The training set 5810 may be provided to feature extractors 5840).
train, by the computer, the embedding convertor by applying the second plurality of machine-learning layers of the embedding convertor on the plurality of training embeddings ([1023] The feature extractors 5840 may be configured to determine features such as n-grams or word embeddings of the training set 5810 that may be provided to the model trainer).
Xu, Mao are considered analogous art within speaker embedding conversion. Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the teachings of Xu in view of Mao, further in view of Ye to incorporate the teachings of Thomson, because of the novel way to incorporate a plurality of training voice profiles using live audio for improved automatic speech recognition system accuracy (Thomson, [0096]).
Regarding claim 14, Xu in view of Mao, further in view of Ye, further in view of Thomson discloses: the system according to claim 12.
Thomson further discloses:
training the embedding extractor includes executing, by the computer, one or more data augmentation operations on at least of a training audio signal and an enrollment signal ([0766] Additionally or alternatively, audio in a first format may be converted to a second format and presented to an ASR system 3720 configured for the second format. For example, wideband audio may be downsampled to 8 kHz and processed by an ASR system 3720 configured to recognize 8 kHz speech. [Down-sampling is listed as one of a non-limiting example of a augmentation operation according to the spec, [0042]]).
Regarding claim 18, Xu in view of Mao, further in view of Ye discloses: the system according to claim 11.
Xu in view of Mao, further in view of Ye does not disclose:
wherein when generating the converted enrolled voiceprint having the second attribute-type the computer is further configured to store the converted enrolled voiceprint into a user profile database.
Thomson discloses:
wherein when generating the converted enrolled voiceprint having the second attribute-type the computer is further configured to store the converted enrolled voiceprint into a user profile database ([0163] In some embodiments, the configuration service may include a business server, a user profile system, and a subscription management system. The configuration service may store information on the individual devices or on a server [Storing user profiles containing voice information ([0163]) in view of the voice prints of Mao, indicates that a converted enrolled voiceprint having the second attribute type is stored in a user profile]).
Xu, Mao are considered analogous art within speaker embedding conversion. Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the teachings of Xu in view of Mao, further in view of Ye to incorporate the teachings of Thomson, because of the novel way to incorporate a plurality of training voice profiles using live audio for improved automatic speech recognition system accuracy (Thomson, [0096]).
Claim(s) 3 and 13 is/are rejected under 35 U.S.C. 103 as being unpatentable over Xu in view of Mao, further in view of Ye, further in view of Thomson, further in view of Khoury et al. (US-20190392842-A1), hereinafter Khoury.
Regarding claim 3, Xu in view of Mao, further in view of Ye, further in view of Thomson discloses: the method according to claim 2.
Xu in view of Mao, further in view of Ye, further in view of Thomson does not disclose:
performing, by the computer, a loss function of the embedding extractor according to a predicted converted embedding outputted by the embedding extractor for a training audio signal, the loss function instructing the computer to update one or more hyper-parameters of one or more layers of the embedding extractor.
Khoury teaches:
performing, by the computer, a loss function of the embedding extractor according to a predicted converted embedding outputted by the embedding extractor for a training audio signal, the loss function instructing the computer to update one or more hyper-parameters of one or more layers of the embedding extractor ([0034] After a given batch is processed, these embedding vectors are used to compute a loss, and the loss is used to modify connection weights in the three feed-forward neural networks 212, 222, 232 according to a back-propagation technique. Claim 15: …the loss function to modify one or more connection weights [Batch tracks to sample(s) and connection weights track to hyper-parameters]).
Xu, Mao, Ye, Thomson, and Khoury are considered analogous art because they are both in the language processing art, so 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 Xu in view of Mao, further in view of Ye, further in view of Thomson to incorporate a loss function. A loss function may facilitate a more accurate speech-processing system during training and therefore a more accurate implementation (Khoury, [0045]).
Regarding claim 13, Xu in view of Mao, further in view of Ye, further in view of Thomson discloses: the system according to claim 12,
Xu in view of Mao, further in view of Ye, further in view of Thomson does not disclose:
perform, by the computer, a loss function of the embedding extractor according to a predicted converted embedding outputted by the embedding extractor for a training audio signal, the loss function instructing the computer to update one or more hyper-parameters of one or more layers of the embedding extractor.
Khoury teaches:
perform, by the computer, a loss function of the embedding extractor according to a predicted converted embedding outputted by the embedding extractor for a training audio signal, the loss function instructing the computer to update one or more hyper-parameters of one or more layers of the embedding extractor ([0034] After a given batch is processed, these embedding vectors are used to compute a loss, and the loss is used to modify connection weights in the three feed-forward neural networks 212, 222, 232 according to a back-propagation technique. Claim 15: …the loss function to modify one or more connection weights [Batch tracks to sample(s) and connection weights track to hyper-parameters]).
Xu, Mao, Ye, Thomson, and Khoury are considered analogous art because they are both in the language processing art, so 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 Xu in view of Mao, further in view of Ye, further in view of Thomson to incorporate a loss function. A loss function may facilitate a more accurate speech-processing system during training and therefore a more accurate implementation (Khoury, [0045]).
Claim(s) 6 and 16 is/are rejected under 35 U.S.C. 103 as being unpatentable over Xu in view of Mao, further in view of Ye, further in view of He et al (CN-102270451-A) with reference to the English machine translation provided, hereinafter He.
Regarding claim 6, Xu in view of Mao, further in view of Ye discloses: the method according to claim 5.
Xu in view of Mao, further in view of Ye fails to disclose:
wherein the computer generates a plurality of converted enrolled voiceprints by applying the plurality of embedding convertors corresponding to the plurality of attribute-types on the plurality of enrollment signals.
He discloses:
wherein the computer generates a plurality of converted enrolled voiceprints by applying the plurality of embedding convertors corresponding to the plurality of attribute-types on the plurality of enrollment signals ([0130] In the embodiment of the invention, the feature extraction unit 402 extracting each voice print characteristic sequence are corresponding to the one specific channel, correspondingly, the model training unit 403 may firstly respectively training to each voice print characteristic sequence to obtain the corresponding voiceprint model; a plurality of voiceprint model then, corresponding to the different channels obtained. [0071] each of the specific channel in the embodiment of the invention corresponds to a group of voice print characteristic sequence, so it can firstly respectively training the voice print characteristic sequence at each channel to obtain the corresponding voiceprint model).
Xu, Mao, Ye, and He are all considered analogous art within language processing. Therefore, it would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention to have modified Xu in view of Mao, further in view of Ye to incorporate generation of more than one voiceprint responding to the plurality of converted embeddings generated as disclosed in He. He specifically mentions compatibility as a reason for generating multiple voiceprints. (He, [0129]).
Regarding claim 16, Xu in view of Mao, further in view of Ye discloses: the system according to claim 15.
Xu in view of Mao, further in view of Ye fails to disclose:
wherein the computer generates a plurality of converted enrolled voiceprints by applying the plurality of embedding convertors corresponding to the plurality of attribute-types on the plurality of enrollment signals.
He discloses:
wherein the computer generates a plurality of converted enrolled voiceprints by applying the plurality of embedding convertors corresponding to the plurality of attribute-types on the plurality of enrollment signals ([0130] In the embodiment of the invention, the feature extraction unit 402 extracting each voice print characteristic sequence are corresponding to the one specific channel, correspondingly, the model training unit 403 may firstly respectively training to each voice print characteristic sequence to obtain the corresponding voiceprint model; a plurality of voiceprint model then, corresponding to the different channels obtained. [0071] each of the specific channel in the embodiment of the invention corresponds to a group of voice print characteristic sequence, so it can firstly respectively training the voice print characteristic sequence at each channel to obtain the corresponding voiceprint model).
Xu, Mao, Ye, and He are all considered analogous art within language processing. Therefore, it would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention to have modified Xu in view of Mao, further in view of Ye to incorporate generation of more than one voiceprint responding to the plurality of converted embeddings generated as disclosed in He. He specifically mentions compatibility as a reason for generating multiple voiceprints. (He, [0129]).
Claim(s) 7 and 17 is/are rejected under 35 U.S.C. 103 as being unpatentable over Xu in view of Mao, further in view of Ye, further in view of Thomson, further in view of Carbune et al. (US-20220157300-A1), hereinafter Carbune.
Regarding claim 7, Xu in view of Mao, further in view of Ye discloses: the method according to claim 5.
Xu in view of Mao, further in view of Ye does not disclose:
identifying, by the computer, the second attribute-type of the inbound embedding.
Thomson discloses:
identifying, by the computer, the second attribute-type of the inbound embedding ([0766] Additionally or alternatively, a first one of the ASR systems 3720 may be configured for audio sampled and encoded in a first format and a second one of the ASR systems 3720 may be configured for audio sampled and encoded in a second format. Additionally or alternatively, audio in a first format may be converted to a second format and presented to an ASR system 3720 configured for the second format. For example, wideband audio may be downsampled to 8 kHz and processed by an ASR system 3720 configured to recognize 8 kHz speech [Identifying the need to downsample indicates identification of type of embeddings]).
Xu are considered analogous art within speaker embedding conversion. Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the teachings of Xu in view of Mao, further in view of Ye to incorporate the teachings of Thomson, because of the novel way to incorporate a plurality of training voice profiles using live audio for improved automatic speech recognition system accuracy (Thomson, [0096]).
Xu in view of Mao, further in view of Ye, further in view of Thomson does not disclose:
selecting, by the computer, the converted enrolled voiceprint from a plurality of converted enrolled voiceprints according to the second attribute-type.
Carbune discloses:
selecting, by the computer, the converted enrolled voiceprint from a plurality of converted enrolled voiceprints according to the second attribute-type ([0063] Filtering based on format or modality may allow selection of digital component objects that have a format that is compatible or optimized for computing device 140 [Format or modality tracks to attribute-type, digital component object tracks to a voiceprint]).
Xu, Mao, Ye, Thomson, and Carbune are all analogous art in the speech processing art. Therefore, it would have been obvious prima facie to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified Xu in view of Mao, further in view of Ye, further in view of Thomson to incorporate the teachings of Carbune, using a filter within a data processing system to determine/match the correct voiceprint based on a series of different voiceprints available with different attribute types (Carbune, [0063]).
Regarding claim 17, Xu in view of Mao, further in view of Ye discloses: the system according to claim 15.
Xu in view of Mao, further in view of Ye does not disclose:
identify, by the computer, the second attribute-type of the inbound embedding.
Thomson discloses:
identify, by the computer, the second attribute-type of the inbound embedding ([0766] Additionally or alternatively, a first one of the ASR systems 3720 may be configured for audio sampled and encoded in a first format and a second one of the ASR systems 3720 may be configured for audio sampled and encoded in a second format. Additionally or alternatively, audio in a first format may be converted to a second format and presented to an ASR system 3720 configured for the second format. For example, wideband audio may be downsampled to 8 kHz and processed by an ASR system 3720 configured to recognize 8 kHz speech [Identifying the need to downsample indicates identification of type of embeddings]).
Xu are considered analogous art within speaker embedding conversion. Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the teachings of Xu in view of Mao, further in view of Ye to incorporate the teachings of Thomson, because of the novel way to incorporate a plurality of training voice profiles using live audio for improved automatic speech recognition system accuracy (Thomson, [0096]).
Xu in view of Mao, further in view of Ye, further in view of Thomson does not disclose:
select, by the computer, the converted enrolled voiceprint from a plurality of converted enrolled voiceprints according to the second attribute-type.
Carbune discloses:
select, by the computer, the converted enrolled voiceprint from a plurality of converted enrolled voiceprints according to the second attribute-type ([0063] Filtering based on format or modality may allow selection of digital component objects that have a format that is compatible or optimized for computing device 140 [Format or modality tracks to attribute-type, digital component objects are listed as audio content items and therefore track to voiceprints]).
Xu, Mao, Ye, Thomson, and Carbune are all analogous art in the speech processing art. Therefore, it would have been obvious prima facie to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified Xu in view of Mao, further in view of Ye, further in view of Thomson to incorporate the teachings of Carbune, using a filter within a data processing system to determine/match the correct voiceprint based on a series of different voiceprints available with different attribute types (Carbune, [0063]).
Conclusion
THIS ACTION IS MADE FINAL. Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
Wang et al. (US-20150278496-A1) discloses “Provided is a method for identity verification that includes receiving an access request, generating a corresponding verification code and returning the verification code. The method acquires voiceprint information and semantic information corresponding to the access request. Then executes an identity verification according to the voiceprint information and semantic information.” (abstract). See entire document.
Johnson et al. (US-10013984-B1) discloses “Various embodiments of the technology described herein alleviate the need to specifically request enrollment information from a user to enroll the user in a voice biometric authentication program. For example, after receiving a call from a user, the system can identify the user and analyze the user's biometric information when the user speaks a command or request. The system can use the user's spoken command or request as enrollment information for the particular command or request or for all spoken requests. After enrollment into the voice biometric authentication program, the system can authenticate the user using biometric information before fulfilling requests or commands” (abstract). See entire document.
Zhao et al. (US-20220157324-A1) discloses “Embodiments may include determination, for each of a plurality of speech frames associated with an acoustic feature, of a phonetic feature based on the associated acoustic feature, generation of one or more two-dimensional feature maps based on the plurality of phonetic features, input of the one or more two-dimensional feature maps to a trained neural network to generate a plurality of speaker embeddings, and aggregation of the plurality of speaker embeddings into a speaker embedding based on respective weights determined for each of the plurality of speaker embeddings, wherein the speaker embedding is associated with an identity of the speaker” (abstract). See entire document.
Bhattacharya et al. (“Deeply Fused Embeddings for Text-Independent Speaker Verification”) discloses “deeply fused speaker embeddings - speaker representations that combine neural speaker embeddings with i-vectors. We show that by combining the two speaker representations we are able to learn robust speaker embeddings in a computationally efficient manner. We compare several different fusion strategies and find that the resulting speaker embeddings show significantly different verification performance. To this end we propose a novel fusion approach that uses an attention model to combine i-vectors with neural speaker embeddings” (abstract). See entire document.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to THEODORE JOHN WITHEY whose telephone number is (703)756-1754. The examiner can normally be reached Monday - Friday, 8am-5pm.
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, Andrew Flanders can be reached on (571) 272-7516. 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.
/THEODORE WITHEY/Examiner, Art Unit 2655
/ANDREW C FLANDERS/Supervisory Patent Examiner, Art Unit 2655
Prosecution Insights