WO2023189173A1 - 話者識別方法、話者識別装置及び話者識別プログラム - Google Patents

話者識別方法、話者識別装置及び話者識別プログラム Download PDF

Info

Publication number
WO2023189173A1
WO2023189173A1 PCT/JP2023/007820 JP2023007820W WO2023189173A1 WO 2023189173 A1 WO2023189173 A1 WO 2023189173A1 JP 2023007820 W JP2023007820 W JP 2023007820W WO 2023189173 A1 WO2023189173 A1 WO 2023189173A1
Authority
WO
WIPO (PCT)
Prior art keywords
speaker
voice data
registered
identification
identified
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Ceased
Application number
PCT/JP2023/007820
Other languages
English (en)
French (fr)
Japanese (ja)
Inventor
美沙貴 土井
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Panasonic Intellectual Property Corp of America
Original Assignee
Panasonic Intellectual Property Corp of America
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Panasonic Intellectual Property Corp of America filed Critical Panasonic Intellectual Property Corp of America
Priority to JP2024511542A priority Critical patent/JPWO2023189173A1/ja
Priority to CN202380030965.2A priority patent/CN118871985A/zh
Publication of WO2023189173A1 publication Critical patent/WO2023189173A1/ja
Priority to US18/898,006 priority patent/US20250022470A1/en
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G10—MUSICAL INSTRUMENTS; ACOUSTICS
    • G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
    • G10L17/00—Speaker identification or verification techniques
    • G10L17/04—Training, enrolment or model building
    • G—PHYSICS
    • G10—MUSICAL INSTRUMENTS; ACOUSTICS
    • G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
    • G10L17/00—Speaker identification or verification techniques
    • G—PHYSICS
    • G10—MUSICAL INSTRUMENTS; ACOUSTICS
    • G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
    • G10L17/00—Speaker identification or verification techniques
    • G10L17/02—Preprocessing operations, e.g. segment selection; Pattern representation or modelling, e.g. based on linear discriminant analysis [LDA] or principal components; Feature selection or extraction
    • G—PHYSICS
    • G10—MUSICAL INSTRUMENTS; ACOUSTICS
    • G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
    • G10L17/00—Speaker identification or verification techniques
    • G10L17/06—Decision making techniques; Pattern matching strategies
    • G—PHYSICS
    • G10—MUSICAL INSTRUMENTS; ACOUSTICS
    • G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
    • G10L17/00—Speaker identification or verification techniques
    • G10L17/22—Interactive procedures; Man-machine interfaces

Definitions

  • the present disclosure relates to technology for identifying speakers.
  • Patent Document 1 discloses that acoustic features are extracted frame by frame from input speech, speech sections of the input speech are detected, noise sections for each type of noise are detected, and a noise suppression method is selected.
  • a noise suppressed speech recognition device has been disclosed that generates an acoustic feature by suppressing the acoustic feature of noise using a noise suppression method, and performs speech recognition using the generated acoustic feature.
  • the present disclosure has been made to solve the above problem, and improves the accuracy of identifying which of a plurality of pre-registered speakers the speaker to be identified is without increasing the amount of calculation.
  • the purpose is to provide technology that can be improved.
  • a speaker identification method is a speaker identification method in a computer, in which voice data to be identified is acquired, a plurality of registered voice data registered in advance is acquired, and the voice data to be identified and the plurality of registered voice data are acquired.
  • the degree of similarity with each of the registered voice data is calculated, and the registered speaker of the registered voice data corresponding to the highest degree of similarity among the plurality of calculated degrees of similarity is selected, and based on the plurality of degrees of similarity calculated.
  • it is determined whether the voice data to be identified is suitable for speaker identification, and when it is determined that the voice data to be identified is suitable for speaker identification, the voice data to be identified is selected based on the highest degree of similarity. It is determined whether or not the registered speaker is to be identified as an identification target speaker of the identification target audio data, and an identification result is output.
  • FIG. 1 is a diagram showing the configuration of a speaker identification system in Embodiment 1 of the present disclosure.
  • 3 is a first flowchart for explaining the operation of speaker identification processing by the speaker identification device in the first embodiment
  • FIG. 7 is a second flowchart for explaining the operation of speaker identification processing by the speaker identification device in the first embodiment.
  • FIG. 2 is a diagram showing the configuration of a speaker identification system in Embodiment 2 of the present disclosure.
  • 7 is a first flowchart for explaining the operation of speaker identification processing by the speaker identification device in the second embodiment
  • FIG. 7 is a second flowchart for explaining the operation of speaker identification processing by the speaker identification device in the second embodiment.
  • FIG. 3 is a diagram showing the configuration of a speaker identification system in Embodiment 3 of the present disclosure.
  • 12 is a first flowchart for explaining the operation of speaker identification processing by the speaker identification device in the third embodiment.
  • 12 is a second flowchart for explaining the operation of speaker identification processing by the speaker identification device in the third embodiment
  • input voice data of a speaker to be identified is acquired, and based on the acquired input voice data and a plurality of pre-registered voice data, the target speaker to be identified is identified by a plurality of pre-registered speakers.
  • Speaker identification is known. In conventional speaker identification, similarity scores are calculated between the feature amount of input voice data of a speaker to be identified and the feature amount of registered voice data of a plurality of registered speakers. Then, the registered speaker of the registered voice data corresponding to the highest similarity score among the plurality of calculated similarity scores is identified as the identification target speaker.
  • the speech section of the input speech is detected, and the noise in the speech section is suppressed to perform speech recognition.
  • a speaker identification method is a speaker identification method in a computer, which acquires voice data to be identified, acquires a plurality of registered voice data registered in advance, and acquires voice data to be identified.
  • the degree of similarity between the target voice data and each of the plurality of registered voice data is calculated, and the registered speaker of the registered voice data corresponding to the highest degree of similarity among the plurality of degrees of similarity calculated is selected, and the plurality of the calculated degrees of similarity are selected. It is determined whether the voice data to be identified is suitable for speaker identification based on the degree of similarity of Based on this, it is determined whether or not the selected registered speaker is to be identified as the identification target speaker of the identification target audio data, and an identification result is output.
  • the degree of similarity between the voice data to be identified and each of the plurality of registered voice data is calculated, and based on the plurality of degrees of similarity calculated, it is determined whether the voice data to be distinguished is suitable for speaker identification. is determined. If it is determined that the voice data to be identified is suitable for speaker identification, whether or not to identify the selected registered speaker as the target speaker of the voice data to be identified based on the highest degree of similarity. is determined.
  • the amount of calculation for processing to calculate multiple degrees of similarity is smaller than the amount of calculation for signal processing to suppress noise included in the audio data to be identified. Also, based on the calculated multiple similarities, it is determined whether the voice data to be identified is suitable for speaker identification, thereby suppressing noise that may distort the personal characteristics of the speaker. Signal processing is not performed on the audio data to be identified. Therefore, it is possible to improve the accuracy of identifying which of a plurality of pre-registered speakers the speaker to be identified is, without increasing the amount of calculation.
  • the highest similarity among the plurality of calculated similarities is determined. It may be determined whether or not the similarity is higher than a first threshold, and if it is determined that the highest similarity is higher than the first threshold, it may be determined that the voice data to be identified is suitable for the speaker identification.
  • the highest similarity among the plurality of calculated similarities is compared with the first threshold, thereby easily determining whether or not the speech data to be identified is suitable for speaker identification. be able to.
  • a variance value of the plurality of calculated similarities is calculated; It is determined whether the variance value obtained is higher than a first threshold value, and if it is determined that the variance value is higher than the first threshold value, it is determined that the voice data to be identified is suitable for the speaker identification. Good too.
  • the variance value of the multiple calculated similarities will be low. Therefore, by comparing the calculated variance values of the plurality of similarities with the first threshold value, it is possible to easily determine whether or not the voice data to be identified is suitable for speaker identification.
  • the plurality of numbers calculated in determining whether or not the selected registered speaker is to be identified as the identification target speaker of the identification target audio data It is determined whether the highest similarity among the degrees of similarity is higher than a second threshold that is higher than the first threshold, and if it is determined that the highest similarity is higher than the second threshold, the selected The registered speaker may be identified as the identification target speaker of the identification target audio data.
  • the highest similarity among the plurality of calculated similarities is compared with the second threshold, which is higher than the first threshold, so that the selected registered speaker can identify the voice data to be identified. It is possible to easily identify whether the speaker is the target speaker or not.
  • the plurality of registered voice data include a plurality of first registered voice data in which voices uttered by a plurality of registered speakers to be identified are registered in advance, and and a plurality of second registered voice data in which voices uttered by a plurality of other registered speakers other than the target registered speakers are registered in advance, and in calculating the similarity, the identification target voice data and the A first degree of similarity with each of the plurality of first registered voice data is calculated, a second degree of similarity is calculated between the voice data to be identified and each of the plurality of second registered voice data, and the registered speaker is selected.
  • the voice data to be identified is speaker-identifiable
  • the possibility that the voice data to be identified is similar to any of the plurality of registered voice data increases as the number of multiple pieces of registered voice data increases. Therefore, not only the plurality of first similarities calculated from the plurality of first registered voice data in which the voices uttered by the plurality of registered speakers to be identified are registered in advance, but also the A plurality of second registered voice data in which voices uttered by other registered speakers are registered in advance.By also using a plurality of calculated second similarities, it is determined whether the voice data to be identified is suitable for speaker identification. It can be determined reliably.
  • the plurality of second registered voice data may include only the voice uttered by the other registered speaker without including noise.
  • the second similarity between the identification target voice data and each of the plurality of second registered voice data is stabilized by using the plurality of second registered voice data that includes only clean voices that do not contain noise. It can be calculated by
  • the plurality of numbers calculated in determining whether or not the selected registered speaker is to be identified as the identification target speaker of the identification target audio data It is determined whether the highest first similarity among the first similarities is higher than a second threshold that is higher than the first threshold, and it is determined that the highest first similarity is higher than the second threshold.
  • the selected registered speaker may be identified as the identification target speaker of the identification target audio data.
  • the highest first similarity among the plurality of calculated first similarities is compared with the second threshold, which is higher than the first threshold, so that the selected registered speaker can be identified. It is possible to easily identify whether or not the speaker is the target speaker of the voice data.
  • the identification target audio data may be output that prompts the speaker to be identified to re-enter data.
  • the speaker to be identified when the voice data to be identified is not suitable for speaker identification, the speaker to be identified can be prompted to re-input the voice data to be identified, and the speaker can speak using the re-input voice data to be identified. User identification can be performed.
  • the voice data to be identified in the first cut out section does not include the voice of the speaker to be identified, it is determined that the voice data to be identified is not suitable for speaker identification. In that case, another voice data to be identified is obtained by cutting out a section different from the first section from the voice data. Therefore, when it is determined that the voice data to be identified is not suitable for speaker identification, it is possible to perform speaker identification using another voice data to be identified.
  • the present disclosure can not only be implemented as a speaker identification method that executes the above-described characteristic processing, but also includes a characteristic configuration corresponding to the characteristic method executed by the speaker identification method. It can also be realized as a speaker identification device or the like. Further, the present invention can also be implemented as a computer program that causes a computer to execute the characteristic processing included in such a speaker identification method. Therefore, the following other aspects can also achieve the same effects as the above speaker identification method.
  • a speaker identification device includes an identification target voice data acquisition unit that acquires voice data to be identified, and a registered voice data acquisition unit that acquires a plurality of registered voice data registered in advance. a calculation unit that calculates the degree of similarity between the voice data to be identified and each of the plurality of registered voice data; and a registered speaker of the registered voice data corresponding to the highest degree of similarity among the plurality of calculated degrees of similarity.
  • a selection unit that selects the voice data to be identified, a similarity determination unit that determines whether the voice data to be identified is suitable for speaker identification based on the plurality of calculated similarities; If it is determined that the speaker is suitable for speaker identification, speaker determination determines whether or not the selected registered speaker is to be identified as the identification target speaker of the identification target voice data based on the highest similarity. and an output unit that outputs the identification result.
  • a speaker identification program acquires voice data to be identified, acquires a plurality of registered voice data registered in advance, and combines the voice data to be identified and the plurality of registered voices.
  • the degree of similarity with each of the data is calculated, the registered speaker of the registered voice data corresponding to the highest degree of similarity among the plurality of calculated degrees of similarity is selected, and the said identification is performed based on the plurality of degrees of similarity calculated. It is determined whether the target audio data is suitable for speaker identification, and when it is determined that the target audio data is suitable for speaker identification, the registered speech selected based on the highest similarity.
  • the computer is operated to determine whether or not to identify the speaker as the speaker to be identified of the voice data to be identified, and to output an identification result.
  • a non-temporary computer-readable recording medium records a speaker identification program, and the speaker identification program acquires speech data to be identified and registers it in advance.
  • a plurality of registered voice data are acquired, a degree of similarity between the voice data to be identified and each of the plurality of registered voice data is calculated, and a registration corresponding to the highest degree of similarity among the plurality of degrees of similarity calculated is obtained.
  • a registered speaker of the voice data is selected, and based on the plurality of calculated similarities, it is determined whether the voice data to be identified is suitable for speaker identification, and the voice data to be identified is suitable for the speaker identification.
  • the selected registered speaker is suitable for the identification target speech data, it is determined based on the highest degree of similarity whether or not the selected registered speaker is to be identified as the identification target speaker of the identification target voice data, and an identification result is output. Make your computer function like this.
  • FIG. 1 is a diagram showing the configuration of a speaker identification system in Embodiment 1 of the present disclosure.
  • the speaker identification system shown in FIG. 1 includes a microphone 1 and a speaker identification device 2. Note that the speaker identification device 2 may or may not include the microphone 1.
  • the microphone 1 picks up the voice uttered by the speaker, converts it into voice data, and outputs it to the speaker identification device 2.
  • the microphone 1 When identifying a speaker, the microphone 1 outputs identification target audio data uttered by the speaker to the speaker identification device 2.
  • the microphone 1 may output the registration target audio data uttered by the speaker to the speaker identification device 2.
  • the microphone 1 may be fixed within the space where the speaker to be identified is present, or may be movable.
  • the speaker identification device 2 includes an identification target speech data acquisition section 21, a first feature calculation section 22, a registered speech data storage section 23, a registered speech data acquisition section 24, a second feature calculation section 25, and a similarity score calculation section. 26, a speaker selection section 27, a similarity score determination section 28, a speaker determination section 29, an identification result output section 30, and an error processing section 31.
  • the unit 28, the speaker determining unit 29, the identification result output unit 30, and the error processing unit 31 are realized by a processor.
  • the processor includes, for example, a central processing unit (CPU).
  • the registered voice data storage unit 23 is realized by a memory.
  • the memory includes, for example, a ROM (Read Only Memory) or an EEPROM (Electrically Erasable Programmable Read Only Memory).
  • the speaker identification device 2 may be, for example, a computer, a smartphone, a tablet computer, or a server.
  • the identification target audio data acquisition unit 21 acquires the identification target audio data output from the microphone 1.
  • the microphone 1 may be built into a terminal such as a smartphone used by the speaker to be identified.
  • the terminal may transmit the voice data to be identified to the speaker identification device 2.
  • the registered voice data acquisition unit 24 may be, for example, a communication unit, and may receive the identification target voice data transmitted by the terminal.
  • the first feature amount calculation unit 22 calculates the feature amount of the identification target audio data acquired by the identification target audio data acquisition unit 21.
  • the feature amount is, for example, an i-vector.
  • the i-vector is a low-dimensional vector feature calculated from audio data by using factor analysis on a GMM (Gaussian Mixture Model) supervector. Note that the method for calculating the i-vector is a conventional technique, so a detailed explanation will be omitted. Further, the feature amount is not limited to the i-vector, and may be other feature amount such as an x-vector.
  • the registered voice data storage unit 23 stores in advance a plurality of registered voice data associated with information regarding speakers.
  • the information regarding the speaker is, for example, a speaker ID for identifying the speaker or the speaker's name.
  • the speaker identification device 2 includes a registration unit that registers the registration target audio data output from the microphone 1 as registered audio data in the registered audio data storage unit 23, and an input unit that receives input of information regarding the speaker of the registered audio data.
  • the computer may further include a reception section. The registration unit may then register the registered voice data in the registered voice data storage unit 23 in association with the information regarding the speaker received by the input reception unit.
  • the utterance content of the identification target voice data and the registered voice data may be anything.
  • the identification target voice data and the registered voice data may be specific words or phrases.
  • the registered voice data acquisition unit 24 acquires a plurality of registered voice data registered in advance in the registered voice data storage unit 23.
  • the registered voice data acquisition unit 24 reads out a plurality of registered voice data registered in advance from the registered voice data storage unit 23.
  • the second feature amount calculation unit 25 calculates the feature amount of the plurality of registered voice data acquired by the registered voice data acquisition unit 24.
  • the feature amount is, for example, an i-vector.
  • the similarity score calculation unit 26 calculates the similarity score between the feature amount of the voice data to be identified and each of the feature amounts of the plurality of registered voice data.
  • the similarity score is a numerical value representing the degree of similarity between the feature amount of the voice data to be identified and the feature amount of the registered voice data.
  • the similarity score indicates the degree of similarity between the feature amount of the voice data to be identified and the feature amount of the registered voice data.
  • the similarity score calculation unit 26 calculates a similarity score using probabilistic linear discriminant analysis (PLDA).
  • the similarity score is a log-likelihood ratio that indicates whether two utterances were generated by the same generative model (same speaker), assuming that the features of the utterances were generated from a probabilistic model. This is what I did.
  • the similarity score is calculated based on the following formula.
  • Similarity score log (likelihood of being uttered by the same speaker/likelihood of being uttered by different speakers)
  • the similarity score calculation unit 26 automatically selects a feature effective for speaker identification from the 400-dimensional i-vector feature and calculates a log-likelihood ratio as a similarity score.
  • the similarity score is calculated when the speaker of the voice data to be identified and the speaker of the registered voice data are the same, and the similarity score is calculated when the speaker of the voice data to be identified is different from the speaker of the registered voice data.
  • the similarity score will be higher than the similarity score.
  • the similarity score calculated from the target speech data that is not suitable for speaker identification and contains noise that is louder than a predetermined volume is calculated from the target speech data that is suitable for speaker identification and contains noise that is lower than a predetermined volume. lower than the calculated similarity score.
  • the similarity score calculation unit 26 may calculate the similarity score between the identification target audio data and each of the plurality of registered audio data.
  • the speaker selection unit 27 selects the registered speaker of the registered voice data corresponding to the highest similarity score among the plurality of similarity scores calculated by the similarity score calculation unit 26.
  • the similarity score determination unit 28 determines whether the identification target audio data is suitable for speaker identification based on the plurality of similarity scores calculated by the similarity score calculation unit 26.
  • the similarity score determination unit 28 determines whether the highest similarity score among the plurality of similarity scores calculated by the similarity score calculation unit 26 is higher than the first threshold value.
  • the similarity score determining unit 28 determines that the identification target speech data is suitable for speaker identification.
  • the similarity score determination unit 28 determines that the highest similarity score is less than or equal to the first threshold, it determines that the identification target audio data is not suitable for speaker identification.
  • the speaker determination unit 29 selects the voice data selected by the speaker selection unit 27 based on the highest similarity score. It is determined whether the registered speaker is to be identified as the identification target speaker of the identification target audio data.
  • the speaker determination unit 29 determines whether the highest similarity score among the plurality of similarity scores calculated by the similarity score calculation unit 26 is higher than a second threshold that is higher than the first threshold. do.
  • the speaker determination unit 29 determines to identify the registered speaker selected by the speaker selection unit 27 as the identification target speaker of the identification target audio data. .
  • the speaker determination unit 29 determines that the highest similarity score is less than or equal to the second threshold, the speaker determination unit 29 identifies the registered speaker selected by the speaker selection unit 27 as the identification target speaker of the identification target audio data. It is determined that it does not.
  • the speaker determining unit 29 selects the voice data selected by the speaker selecting unit 27.
  • the registered speaker may be identified as the speaker to be identified of the voice data to be identified.
  • the speaker determination unit 29 does not determine whether the highest similarity score among the plurality of similarity scores calculated by the similarity score calculation unit 26 is higher than the second threshold.
  • the registered speaker selected by the selection unit 27 may be identified as the speaker to be identified of the voice data to be identified.
  • the identification result output unit 30 outputs the identification result by the speaker determination unit 29.
  • the identification result output unit 30 When the selected registered speaker is identified as the identification target speaker of the identification target audio data, the identification result output unit 30 outputs an identification result including the name or speaker ID of the selected registered speaker. Further, the identification result may include a similarity score.
  • the identification result output unit 30 outputs a plurality of registered speakers in which the identification target speaker of the identification target audio data is registered in advance. Outputs an identification result indicating that it was not identified by any of the speakers.
  • the identification result output unit 30 is, for example, a display or a speaker, and when the selected registered speaker is identified as the identification target speaker of the identification target audio data, the identification result output unit 30 is a display or a speaker. A message indicating that the speaker is a registered speaker is output from the display or speaker. On the other hand, if the selected registered speaker is not identified as the identification target speaker of the identification target audio data, the identification result output unit 30 outputs a plurality of registered speakers in which the identification target speaker of the identification target audio data is registered in advance. A message indicating that the speaker is not one of the speakers is output from the display or speaker.
  • the identification result output unit 30 may output the identification result by the speaker determination unit 29 to a device other than the speaker identification device 2.
  • the identification result output unit 30 may include, for example, a communication unit, and may transmit the identification result to a terminal such as a smartphone used by the speaker to be identified.
  • the terminal may include a display or a speaker. The display or speaker of the terminal may output the received identification result.
  • the error processing unit 31 If the similarity score determining unit 28 determines that the target voice data is not suitable for speaker identification, the error processing unit 31 outputs an error message urging the target speaker to re-enter the target voice data. . For example, the error processing unit 31 outputs an error message saying, "Please move closer to the microphone or speak in a quiet place.”
  • the error processing unit 31 is, for example, a display or a speaker, and when the similarity score determining unit 28 determines that the target voice data is not suitable for speaker identification, the error processing unit 31 requests re-input of the target voice data to the target speaker. Outputs an error message from the display or speaker to prompt you.
  • the error processing unit 31 may output an error message to a device other than the speaker identification device 2 to prompt the speaker to be identified to re-input the voice data to be identified.
  • the error processing unit 31 may include, for example, a communication unit, and may send an error message to a terminal such as a smartphone used by the speaker to be identified.
  • the terminal may include a display or a speaker. The terminal's display or speaker may output the received error message.
  • FIG. 2 is a first flowchart for explaining the operation of the speaker identification process of the speaker identification device 2 in the first embodiment
  • FIG. 12 is a second flowchart for explaining the operation of speaker identification processing.
  • step S1 the identification target audio data acquisition unit 21 acquires the identification target audio data output from the microphone 1.
  • the speaker to be identified speaks into the microphone 1.
  • the microphone 1 collects the voice uttered by the speaker to be identified, and outputs the voice data to be identified.
  • step S2 the first feature amount calculation unit 22 calculates the feature amount of the identification target audio data acquired by the identification target audio data acquisition unit 21.
  • step S3 the registered voice data acquisition unit 24 acquires registered voice data from the registered voice data storage unit 23.
  • the registered voice data acquisition section 24 acquires one registered voice data from among the plurality of registered voice data registered in the registered voice data storage section 23.
  • step S4 the second feature amount calculation unit 25 calculates the feature amount of the registered voice data acquired by the registered voice data acquisition unit 24.
  • step S5 the similarity score calculation unit 26 calculates the similarity score between the feature amount of the identification target audio data and the feature amount of the registered audio data.
  • step S6 the similarity score calculation unit 26 calculates the similarity score between the feature amount of the voice data to be identified and the feature amount of all the registered voice data stored in the registered voice data storage unit 23. Determine whether or not it has been done.
  • the process returns to step S3.
  • the registered voice data acquisition unit 24 acquires registered voice data for which a similarity score has not been calculated from among the plurality of registered voice data stored in the registered voice data storage unit 23.
  • step S7 the speaker selection unit 27 The registered speaker of the registered voice data corresponding to the highest similarity score among the plurality of similarity scores calculated by the degree score calculation unit 26 is selected.
  • step S8 the similarity score determination unit 28 determines whether the highest similarity score is higher than the first threshold.
  • step S9 the error processing unit 31 requests the identification target speaker to re-input the identification target speech data. Outputs a prompting error message.
  • step S10 the speaker determination unit 29 selects the plurality of similarities calculated by the similarity score calculation unit 26. It is determined whether the highest similarity score among the scores is higher than a second threshold that is higher than the first threshold.
  • the speaker determination unit 29 selects the registered speaker selected by the speaker selection unit 27 in step S11. Identifies as the target speaker of the target voice data.
  • step S12 the speaker determination unit 29 determines that the registered speaker selected by the speaker selection unit 27 is It is determined that the speaker is not the speaker to be identified in the voice data to be identified.
  • the identification result output unit 30 outputs the identification result by the speaker determination unit 29.
  • the identification result output unit 30 determines that the identification target speaker of the identification target audio data is the selected registered speaker. Outputs a message indicating.
  • the identification result output unit 30 outputs a plurality of pre-registered identification target speakers of the identification target audio data. Outputs a message indicating that the speaker is not one of the registered speakers.
  • similarity scores between the voice data to be identified and each of the plurality of registered voice data are calculated, and based on the plurality of calculated similarity scores, it is determined whether the voice data to be distinguished is suitable for speaker identification. is determined. If it is determined that the voice data to be identified is suitable for speaker identification, the selected registered speaker is identified as the target speaker of the voice data to be identified based on the highest similarity score. It is determined whether
  • the amount of calculation for the process of calculating multiple similarity scores is smaller than the amount of calculation for signal processing for suppressing noise included in the audio data to be identified. Also, based on the multiple similarity scores calculated, it is determined whether the voice data to be identified is suitable for speaker identification, thereby suppressing noise that may distort the personal characteristics of the speaker. signal processing is not performed on the audio data to be identified. Therefore, it is possible to improve the accuracy of identifying which of a plurality of pre-registered speakers the speaker to be identified is, without increasing the amount of calculation.
  • the error processing unit 31 outputs an error message prompting the identification target speaker to re-input the identification target speech data, but the present disclosure is not particularly limited to this.
  • the identification target audio data acquisition unit 21 may acquire identification target audio data obtained by cutting out a predetermined section from the audio data uttered by the identification target speaker. At this time, there is a possibility that the voice data to be identified that has been extracted from a predetermined section does not include the voice of the speaker to be identified. In this case, the similarity score determining unit 28 determines that the voice data to be identified is not suitable for speaker identification.
  • the error processing unit 31 creates another identification target by cutting out a section different from the predetermined section from the speech data. Audio data may also be acquired. Then, the process returns to step S2, and the first feature amount calculation section 22 may calculate the feature amount of another identification target audio data acquired by the error processing section 31. After that, the processes after step S3 may be performed.
  • the voice data to be identified in the first cut out section does not include the voice of the speaker to be identified, it is determined that the voice data to be identified is not suitable for speaker identification.
  • another voice data to be identified is obtained by cutting out a section different from the first section from the voice data. Therefore, when it is determined that the voice data to be identified is not suitable for speaker identification, it is possible to perform speaker identification using another voice data to be identified.
  • Emodiment 2 In the first embodiment described above, it is determined whether the highest similarity score among the plurality of calculated similarity scores is higher than the first threshold, and it is determined that the highest similarity score is higher than the first threshold. If so, it is determined that the voice data to be identified is suitable for speaker identification.
  • the variance value of the plurality of calculated similarity scores is calculated, it is determined whether the calculated variance value is higher than the first threshold value, and the variance value is higher than the first threshold value. If it is determined that the voice data to be identified is high, it is determined that the voice data to be identified is suitable for speaker identification.
  • FIG. 4 is a diagram showing the configuration of a speaker identification system in Embodiment 2 of the present disclosure.
  • the speaker identification system shown in FIG. 4 includes a microphone 1 and a speaker identification device 2A. Note that the speaker identification device 2A may or may not include the microphone 1.
  • Embodiment 2 the same components as in Embodiment 1 are given the same reference numerals, and description thereof will be omitted.
  • the speaker identification device 2A includes an identification target speech data acquisition section 21, a first feature amount calculation section 22, a registered speech data storage section 23, a registered speech data acquisition section 24, a second feature amount calculation section 25, and a similarity score calculation section. 26, a speaker selection section 27, a similarity score determination section 28A, a speaker determination section 29, an identification result output section 30, and an error processing section 31.
  • the similarity score determination unit 28A determines whether the identification target audio data is suitable for speaker identification based on the plurality of similarity scores calculated by the similarity score calculation unit 26.
  • the similarity score determination unit 28A calculates a variance value of the plurality of similarity scores calculated by the similarity score calculation unit 26.
  • the similarity score determination unit 28A determines whether the calculated variance value is higher than the first threshold. If the similarity score determination unit 28A determines that the variance value is higher than the first threshold, it determines that the voice data to be identified is suitable for speaker identification. On the other hand, if the similarity score determination unit 28A determines that the variance value is less than or equal to the first threshold, it determines that the identification target speech data is not suitable for speaker identification.
  • the similarity scores between the voice data to be identified and the plurality of registered voice data will all be low values. . Therefore, if the variance value of the plurality of similarity scores is low, it is possible to determine that the speech data to be identified is not suitable for speaker identification.
  • FIG. 5 is a first flowchart for explaining the operation of the speaker identification process of the speaker identification device 2A in the second embodiment
  • FIG. 12 is a second flowchart for explaining the operation of speaker identification processing.
  • step S21 to step S27 is the same as the processing from step S1 to step S7 in FIG. 2, so a description thereof will be omitted.
  • step S28 the similarity score determination unit 28A calculates the variance value of the plurality of similarity scores calculated by the similarity score calculation unit 26.
  • step S29 the similarity score determination unit 28A determines whether the calculated variance value is higher than the first threshold value.
  • step S30 the error processing unit 31 sends an error message to the speaker to be identified to prompt the speaker to re-enter the voice data to be identified. Output.
  • step S31 the speaker determining unit 29 selects one of the plurality of similarity scores calculated by the similarity score calculating unit 26. It is determined whether the highest similarity score of is higher than a second threshold that is higher than the first threshold.
  • step S31 to step S34 is the same as the processing from step S9 to step S12 in FIG. 3, so a description thereof will be omitted.
  • the variance value of the plurality of calculated similarity scores will be low. Therefore, by comparing the variance values of the plurality of calculated similarity scores with the first threshold value, it is possible to easily determine whether or not the identification target speech data is suitable for speaker identification.
  • the similarity score calculation unit 26 calculates the similarity score between the feature amount of the voice data to be identified and each of the feature amounts of the plurality of registered voice data.
  • the present disclosure is not particularly limited thereto.
  • the similarity score calculation unit 26 may calculate similarity scores between the identification target audio data and each of the plurality of registered audio data. In this case, it becomes unnecessary to calculate the feature amount of the voice data to be identified and the feature amount of the plurality of registered voice data.
  • the first similarity score between the voice data to be identified and each of the plurality of first registered voice data in which voices uttered by the plurality of registered speakers to be identified are registered in advance is calculated. Based on the plurality of first similarity scores, it is determined whether the identification target speech data is suitable for speaker identification. In contrast, in Embodiment 3, the identification target voice data and a plurality of second registered voice data in which voices uttered by a plurality of registered speakers other than the plurality of registered speakers to be identified are registered in advance. A second similarity score with each is calculated, and it is determined whether the identification target voice data is suitable for speaker identification based on the plurality of first similarity scores and the plurality of second similarity scores. be done.
  • FIG. 7 is a diagram showing the configuration of a speaker identification system in Embodiment 3 of the present disclosure.
  • the speaker identification system shown in FIG. 7 includes a microphone 1 and a speaker identification device 2B. Note that the speaker identification device 2B may or may not include the microphone 1.
  • Embodiment 3 the same components as in Embodiment 1 are given the same reference numerals, and the description thereof will be omitted.
  • the speaker identification device 2B includes an identification target voice data acquisition section 21, a first feature amount calculation section 22, a first registered voice data storage section 23B, a first registered voice data acquisition section 24B, a second feature amount calculation section 25B, and a similar degree score calculation unit 26B, speaker selection unit 27B, similarity score determination unit 28B, speaker determination unit 29B, identification result output unit 30, error processing unit 31, second registered voice data storage unit 32, second registered voice data It includes an acquisition section 33 and a third feature amount calculation section 34.
  • the score determination section 28B, the speaker determination section 29B, the identification result output section 30, the error processing section 31, the second registered voice data acquisition section 33, and the third feature amount calculation section 34 are realized by a processor.
  • the first registered voice data storage section 23B and the second registered voice data storage section 32 are realized by memory.
  • the first registered voice data storage unit 23B stores in advance a plurality of first registered voice data associated with information about speakers.
  • the plurality of first registered voice data indicate voices uttered by the plurality of registered speakers to be identified.
  • the plurality of first registered voice data are the same as the plurality of registered voice data in the first embodiment.
  • the first registered voice data acquisition unit 24B acquires a plurality of first registered voice data registered in advance in the first registered voice data storage unit 23B.
  • the second feature amount calculation unit 25B calculates the feature amount of the plurality of first registered voice data acquired by the first registered voice data acquisition unit 24B.
  • the feature amount is, for example, an i-vector.
  • the second registered voice data storage unit 32 stores a plurality of second registered voice data in advance.
  • the plurality of second registered voice data indicate voices uttered by a plurality of other registered speakers other than the plurality of registered speakers to be identified.
  • the plurality of second registered audio data does not include noise and only includes audio.
  • the second registered voice data acquisition unit 33 acquires a plurality of second registered voice data registered in advance in the second registered voice data storage unit 32.
  • the third feature amount calculation unit 34 calculates the feature amount of the plurality of second registered voice data acquired by the second registered voice data acquisition unit 33.
  • the feature amount is, for example, an i-vector.
  • the similarity score calculation unit 26B calculates a first similarity score between the feature amount of the identification target audio data and each of the plurality of first registered audio data, and also calculates a first similarity score between the feature amount of the identification target audio data and each of the plurality of first registered audio data. A second similarity score with each of the feature amounts of the two registered voice data is calculated.
  • the speaker selection unit 27B selects the registered speaker of the first registered voice data corresponding to the highest first similarity score among the plurality of first similarity scores calculated by the similarity score calculation unit 26B.
  • the similarity score determination unit 28B determines whether or not the voice data to be identified is suitable for speaker identification based on the plurality of first similarity scores and the plurality of second similarity scores calculated by the similarity score calculation unit 26B. Determine whether Here, the similarity score determination section 28B determines whether the highest first similarity score or the second similarity score is the highest among the plurality of first similarity scores and the plurality of second similarity scores calculated by the similarity score calculation section 26B. It is determined whether the degree score is higher than a first threshold value. When determining that the highest first similarity score or second similarity score is higher than the first threshold, the similarity score determination unit 28B determines that the identification target audio data is suitable for speaker identification.
  • the similarity score determination unit 28B determines that the highest first similarity score or second similarity score is less than or equal to the first threshold, the similarity score determination unit 28B determines that the voice data to be identified is not suitable for speaker identification. .
  • the second registered voice data storage unit 32 in the third embodiment stores a plurality of clean voices that do not contain noise and are uttered by a plurality of registered speakers other than the plurality of registered speakers to be identified.
  • the second registered audio data is stored in advance.
  • the number of other registered speakers is, for example, 100
  • the number of second registered voice data is, for example, 100. If there is second registered voice data similar to the identification target voice data among the plurality of second registered voice data, it can be determined that the speaker of the identification target voice data can be identified.
  • the speaker determining unit 29B selects the voice data by the speaker selecting unit 27B based on the highest first similarity score. It is determined whether or not the registered speaker is to be identified as the identification target speaker of the identification target audio data.
  • the speaker determination unit 29B determines whether the highest first similarity score among the plurality of first similarity scores calculated by the similarity score calculation unit 26B is higher than a second threshold that is higher than the first threshold. Determine whether or not.
  • the speaker determination unit 29B determines that the highest first similarity score is higher than the second threshold, the speaker determination unit 29B identifies the registered speaker selected by the speaker selection unit 27B as the identification target speaker of the identification target audio data. judge.
  • the speaker determination unit 29B determines that the highest first similarity score is equal to or less than the second threshold, the speaker determination unit 29B selects the registered speaker selected by the speaker selection unit 27B as the identification target speaker of the identification target audio data. It is determined that it is not identified as
  • the speaker determining section 29B selects the voice data selected by the speaker selecting section 27B.
  • the registered speaker may be identified as the speaker to be identified of the voice data to be identified.
  • the speaker determination unit 29 does not determine whether the highest first similarity score among the plurality of first similarity scores calculated by the similarity score calculation unit 26B is higher than the second threshold.
  • the registered speaker selected by the speaker selection unit 27B may be identified as the speaker to be identified of the voice data to be identified.
  • FIG. 8 is a first flowchart for explaining the operation of the speaker identification process of the speaker identification device 2B in the third embodiment
  • FIG. 12 is a second flowchart for explaining the operation of speaker identification processing.
  • step S41 and step S42 is the same as the processing in step S1 and step S2 in FIG. 2, so the explanation will be omitted.
  • step S43 the first registered voice data acquisition unit 24B acquires the first registered voice data from the first registered voice data storage unit 23B. At this time, the first registered voice data acquisition section 24B acquires one first registered voice data from among the plurality of first registered voice data registered in the first registered voice data storage section 23B.
  • step S44 the second feature amount calculation unit 25B calculates the feature amount of the first registered audio data acquired by the first registered audio data acquisition unit 24B.
  • step S45 the similarity score calculation unit 26B calculates a first similarity score between the feature amount of the identification target audio data and the feature amount of the first registered audio data.
  • step S46 the similarity score calculation unit 26B compares the feature amount of the voice data to be identified with the feature amount of all the first registered voice data stored in the first registered voice data storage unit 23B. 1. Determine whether or not a similarity score has been calculated. Here, if it is determined that the first similarity score between the feature amount of the voice data to be identified and the feature amount of all the first registered voice data has not been calculated (NO in step S46), the process proceeds to step S43. return. Then, the first registered voice data acquisition unit 24B selects the first registered voices for which the first similarity score has not been calculated from among the plurality of first registered voice data stored in the first registered voice data storage unit 23B. Get data.
  • step S47 the second registered voice The data acquisition section 33 acquires the second registered voice data from the second registered voice data storage section 32 . At this time, the second registered voice data acquisition section 33 acquires one second registered voice data from among the plurality of second registered voice data registered in the second registered voice data storage section 32.
  • step S48 the third feature amount calculation unit 34 calculates the feature amount of the second registered audio data acquired by the second registered audio data acquisition unit 33.
  • step S49 the similarity score calculation unit 26B calculates a second similarity score between the feature amount of the identification target audio data and the feature amount of the second registered audio data.
  • step S50 the similarity score calculation unit 26B compares the feature quantity of the identification target speech data with the feature quantity of all the second registered speech data stored in the second registered speech data storage section 32. 2. Determine whether or not a similarity score has been calculated. Here, if it is determined that the second similarity score between the feature amount of the voice data to be identified and the feature amount of all the second registered voice data has not been calculated (NO in step S50), the process proceeds to step S47. return. Then, the second registered voice data acquisition unit 33 selects second registered voice data for which the second similarity score has not been calculated from among the plurality of second registered voice data stored in the second registered voice data storage unit 32. Get data.
  • step S51 the speaker selection unit 27B selects the registered speaker of the first registered voice data corresponding to the highest first similarity score among the plurality of first similarity scores calculated by the similarity score calculation unit 26B.
  • step S52 the similarity score determination unit 28B determines whether the highest first similarity score or second similarity score is higher than the first threshold.
  • step S53 the error processing unit 31 Outputs an error message prompting the identified speaker to re-enter the input.
  • step S54 the speaker determination unit 29B selects the similarity score calculation unit 26B. It is determined whether the highest first similarity score among the plurality of first similarity scores calculated by is higher than a second threshold that is higher than the first threshold.
  • step S55 the speaker determining unit 29B selects the registered talk selected by the speaker selecting unit 27B.
  • the speaker is identified as the speaker to be identified in the voice data to be identified.
  • step S56 the speaker determining unit 29B selects the registered talk selected by the speaker selecting unit 27B. It is determined that the speaker is not the speaker to be identified of the voice data to be identified.
  • step S57 is the same as the process in step S12 in FIG. 3, so a description thereof will be omitted.
  • the voice data to be identified is speaker-identifiable, the possibility that the voice data to be identified is similar to any of the plurality of registered voice data increases as the number of multiple pieces of registered voice data increases. Therefore, in addition to the plurality of first similarity scores calculated from the plurality of first registered voice data in which the voices uttered by the plurality of registered speakers to be identified are registered in advance, the A plurality of second registered voice data in which voices uttered by a plurality of other registered speakers are registered in advance. By also using a plurality of calculated second similarity scores, it is possible to determine whether the voice data to be identified is suitable for speaker identification. It is possible to reliably determine whether
  • the similarity score determination unit 28B determines whether the audio data to be identified is based on the plurality of first similarity scores and the plurality of second similarity scores calculated by the similarity score calculation unit 26B.
  • the similarity score determination unit 28B may determine whether the identification target audio data is suitable for speaker identification based on the plurality of second similarity scores calculated by the similarity score calculation unit 26B.
  • the similarity score determination unit 28B determines whether the highest second similarity score among the plurality of second similarity scores calculated by the similarity score calculation unit 26B is higher than the first threshold. It's okay.
  • the similarity score determination unit 28B determines that the highest second similarity score is higher than the first threshold, it may determine that the identification target audio data is suitable for speaker identification. On the other hand, if the similarity score determination unit 28B determines that the highest second similarity score is less than or equal to the first threshold, it may determine that the identification target audio data is not suitable for speaker identification.
  • the similarity score calculation unit 26B calculates a first similarity score between the feature amount of the identification target audio data and each of the feature amounts of the plurality of first registered audio data, and also Although the second similarity score between the feature amount of the data and each feature amount of the plurality of second registered voice data is calculated, the present disclosure is not particularly limited thereto.
  • the similarity score calculation unit 26B calculates a first similarity score between the identification target audio data and each of the plurality of first registered audio data, and a second similarity score between the identification target audio data and each of the plurality of second registered audio data.
  • a similarity score may also be calculated. In this case, it is not necessary to calculate the feature amount of the identification target audio data, the feature amount of the multiple first registered audio data, and the feature amount of the multiple second registered audio data.
  • each component may be configured with dedicated hardware, or may be realized by executing a software program suitable for each component.
  • Each component may be realized by a program execution unit such as a CPU or a processor reading and executing a software program recorded on a recording medium such as a hard disk or a semiconductor memory.
  • the program may be executed by another independent computer system by recording the program on a recording medium and transferring it, or by transferring the program via a network.
  • LSI Large Scale Integration
  • circuit integration is not limited to LSI, and may be realized using a dedicated circuit or a general-purpose processor.
  • An FPGA Field Programmable Gate Array
  • reconfigurable processor that can reconfigure the connections and settings of circuit cells inside the LSI may be used.
  • a processor such as a CPU executing a program.
  • the technology according to the present disclosure can improve the accuracy of identifying which of a plurality of pre-registered speakers the speaker to be identified is without increasing the amount of calculation. It is useful as an identification technique.

Landscapes

  • Engineering & Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • Audiology, Speech & Language Pathology (AREA)
  • Human Computer Interaction (AREA)
  • Physics & Mathematics (AREA)
  • Acoustics & Sound (AREA)
  • Multimedia (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Business, Economics & Management (AREA)
  • Game Theory and Decision Science (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
  • Telephone Function (AREA)
PCT/JP2023/007820 2022-03-29 2023-03-02 話者識別方法、話者識別装置及び話者識別プログラム Ceased WO2023189173A1 (ja)

Priority Applications (3)

Application Number Priority Date Filing Date Title
JP2024511542A JPWO2023189173A1 (cg-RX-API-DMAC7.html) 2022-03-29 2023-03-02
CN202380030965.2A CN118871985A (zh) 2022-03-29 2023-03-02 说话者识别方法、说话者识别装置以及说话者识别程序
US18/898,006 US20250022470A1 (en) 2022-03-29 2024-09-26 Speaker identification method, speaker identification device, and non-transitory computer readable recording medium storing speaker identification program

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
JP2022-053033 2022-03-29
JP2022053033 2022-03-29

Related Child Applications (1)

Application Number Title Priority Date Filing Date
US18/898,006 Continuation US20250022470A1 (en) 2022-03-29 2024-09-26 Speaker identification method, speaker identification device, and non-transitory computer readable recording medium storing speaker identification program

Publications (1)

Publication Number Publication Date
WO2023189173A1 true WO2023189173A1 (ja) 2023-10-05

Family

ID=88201204

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/JP2023/007820 Ceased WO2023189173A1 (ja) 2022-03-29 2023-03-02 話者識別方法、話者識別装置及び話者識別プログラム

Country Status (4)

Country Link
US (1) US20250022470A1 (cg-RX-API-DMAC7.html)
JP (1) JPWO2023189173A1 (cg-RX-API-DMAC7.html)
CN (1) CN118871985A (cg-RX-API-DMAC7.html)
WO (1) WO2023189173A1 (cg-RX-API-DMAC7.html)

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPS59185269A (ja) * 1983-04-05 1984-10-20 三菱電機株式会社 入出門管理システム
JP2000215316A (ja) * 1999-01-27 2000-08-04 Toshiba Corp 生体情報認識装置およびその方法
JP2005128628A (ja) * 2003-10-21 2005-05-19 Toshiba Corp パターン識別における照合に用いられるテンプレートの生成ならびに同テンプレートを用いたパターン識別のための方法、装置、およびプログラム
JP2006078855A (ja) * 2004-09-10 2006-03-23 Toshiba Tec Corp 個人認証装置
JP2016166927A (ja) * 2015-03-09 2016-09-15 日本電信電話株式会社 パラメータ学習装置、話者認識装置、パラメータ学習方法、話者認識方法、およびプログラム
JP2020154061A (ja) * 2019-03-19 2020-09-24 株式会社フュートレック 話者識別装置、話者識別方法およびプログラム
JP2020187479A (ja) * 2019-05-13 2020-11-19 株式会社マイクロネット 顔認識システム、顔認識方法及び顔認識プログラム

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPS59185269A (ja) * 1983-04-05 1984-10-20 三菱電機株式会社 入出門管理システム
JP2000215316A (ja) * 1999-01-27 2000-08-04 Toshiba Corp 生体情報認識装置およびその方法
JP2005128628A (ja) * 2003-10-21 2005-05-19 Toshiba Corp パターン識別における照合に用いられるテンプレートの生成ならびに同テンプレートを用いたパターン識別のための方法、装置、およびプログラム
JP2006078855A (ja) * 2004-09-10 2006-03-23 Toshiba Tec Corp 個人認証装置
JP2016166927A (ja) * 2015-03-09 2016-09-15 日本電信電話株式会社 パラメータ学習装置、話者認識装置、パラメータ学習方法、話者認識方法、およびプログラム
JP2020154061A (ja) * 2019-03-19 2020-09-24 株式会社フュートレック 話者識別装置、話者識別方法およびプログラム
JP2020187479A (ja) * 2019-05-13 2020-11-19 株式会社マイクロネット 顔認識システム、顔認識方法及び顔認識プログラム

Also Published As

Publication number Publication date
CN118871985A (zh) 2024-10-29
JPWO2023189173A1 (cg-RX-API-DMAC7.html) 2023-10-05
US20250022470A1 (en) 2025-01-16

Similar Documents

Publication Publication Date Title
CN111009248B (zh) 说话者识别装置、说话者识别方法及记录介质
JP6812843B2 (ja) 音声認識用コンピュータプログラム、音声認識装置及び音声認識方法
US8977547B2 (en) Voice recognition system for registration of stable utterances
JP6464005B2 (ja) 雑音抑圧音声認識装置およびそのプログラム
CN112420021B (zh) 学习方法、说话者识别方法以及记录介质
CN111108551B (zh) 一种声纹鉴定方法和相关装置
CN1639768B (zh) 自动语音识别方法及装置
US9786295B2 (en) Voice processing apparatus and voice processing method
WO2018051945A1 (ja) 音声処理装置、音声処理方法、および記録媒体
JPWO2020003413A1 (ja) 情報処理装置、制御方法、及びプログラム
WO2020195924A1 (ja) 信号処理装置および方法、並びにプログラム
WO2010128560A1 (ja) 音声認識装置、音声認識方法、及び音声認識プログラム
US20220383880A1 (en) Speaker identification apparatus, speaker identification method, and recording medium
CN112992184B (zh) 一种发音评测方法、装置、电子设备和存储介质
JP4074543B2 (ja) 音声処理装置、音声処理方法、音声処理プログラム、および、プログラム記録媒体
US11580989B2 (en) Training method of a speaker identification model based on a first language and a second language
JP2012168296A (ja) 音声による抑圧状態検出装置およびプログラム
CN109155128B (zh) 声学模型学习装置、声学模型学习方法、语音识别装置和语音识别方法
CN109065026B (zh) 一种录音控制方法及装置
CN111640423B (zh) 一种词边界估计方法、装置及电子设备
WO2023189173A1 (ja) 話者識別方法、話者識別装置及び話者識別プログラム
JP2013235050A (ja) 情報処理装置及び方法、並びにプログラム
JP7781783B2 (ja) 識別装置、識別方法、および、プログラム
CN115240650B (zh) 一种语音数据的处理方法、装置、电子设备及存储介质
JP6721676B2 (ja) 対話装置、対話装置の制御方法、および制御プログラム

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 23779202

Country of ref document: EP

Kind code of ref document: A1

ENP Entry into the national phase

Ref document number: 2024511542

Country of ref document: JP

Kind code of ref document: A

WWE Wipo information: entry into national phase

Ref document number: 202380030965.2

Country of ref document: CN

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 23779202

Country of ref document: EP

Kind code of ref document: A1