JP6377921B2 - Speaker recognition device, speaker recognition method, and speaker recognition program - Google Patents

Description

  The present invention relates to a speaker recognition device, a speaker recognition method, and a speaker recognition program for recognizing a speaker of input voice data based on input voice data.

  Conventionally, a technique for recognizing a speaker of voice data based on the voice data is known. For example, Patent Document 1 generates and stores registered model data in advance from voice data of a person to be collated, and collates voice analysis data obtained by analyzing the input voice data with the registered model data. A speaker recognition system for determining whether or not a speaker is a verification target is disclosed.

  In such speaker recognition, the registration model data greatly affects the recognition accuracy. When the registration model data is constructed from a single voice data, the recognition accuracy is greatly reduced if the voice data is inappropriate. Therefore, the registration model data can be constructed from a plurality of voice data, or a plurality of voice data It is used as registration model data.

Japanese Patent Laying-Open No. 2005-091758

  However, in the conventional technique, there is a problem that the registration model data of the person to be collated is not always in an appropriate state, and the recognition accuracy may be lowered. Specifically, when inappropriate audio data is mixed in a plurality of audio data obtained at the time of registration, the inappropriate audio data affects the registered model data, leading to a reduction in recognition accuracy. Even if the state is appropriate at the time of registration, if the voice on the speaker side changes due to a change over time or the like, the recognition accuracy deteriorates.

  For these reasons, it is an important issue to improve the accuracy of speaker recognition by holding the voice data of the person to be verified in an appropriate state and comparing it with the input voice data.

  The present invention was made in order to solve the above-described problems of the prior art, and holds the speech data of the person to be verified in an appropriate state, thereby improving the speaker recognition accuracy, It is an object to provide a speaker recognition method and a speaker recognition program.

In order to solve the above-described problems and achieve the object, the invention according to claim 1 is a speaker recognition device for recognizing a speaker of input voice data based on the input voice data that has been input. Registered voice data receiving means for receiving a plurality of registered voice data for a speaker to be registered, and registered voice data selecting means for selecting a predetermined number of registered voice data from the plurality of registered voice data received by the registered voice data receiving means Storage means for storing a predetermined number of registered voice data selected by the registered voice data selection means in association with the speaker, the input voice data, and the registered voice data stored in the storage means. a similarity calculation means for calculating the degree of similarity, and a speaker identification means for identifying the speaker on the basis of the similarity calculated by the similarity calculation unit, the registration The voice data selection means calculates, for each of the plurality of registered voice data received by the registered voice data reception means, a variance of the distances of the plurality of registered voice data when the registered voice data is excluded from selection targets, By excluding the registered voice data so as to minimize the variance, a predetermined number of registered voice data is selected from the plurality of registered voice data received by the registered voice data receiving means .

Further, the invention according to claim 2 is the invention according to claim 1 , wherein the storage means sets registered voice data of a predetermined number of data of the same speaker as a registered voice set, and a plurality of the same speaker The registered voice set is stored in association with each other.
According to a third aspect of the present invention, there is provided a speaker recognition device for recognizing a speaker of the input voice data based on the input voice data, and a plurality of registered voice data for the speaker to be identified. Registered voice data receiving means, registered voice data selecting means for selecting a predetermined number of registered voice data from a plurality of registered voice data received by the registered voice data receiving means, and selected by the registered voice data selecting means Storage means for storing a predetermined number of registered voice data in association with the speaker; similarity calculation means for calculating a similarity between the input voice data and the registered voice data stored in the storage means; Speaker identification means for identifying the speaker based on the similarity calculated by the similarity calculation means, and the storage means stores a predetermined number of data of the same speaker. The audio data and the registered voice set, and to store in association with a plurality of registered voice sets for the same speaker.

Further, in the invention according to claim 4 , in the invention according to any one of claims 1 to 3 , when the identification by the speaker identification means is performed, the registered voice set contributing to the identification is stored. And storing the input voice data as registered voice data of an update preparation set, deleting a registered voice set whose contribution to identification is less indicated by the result data, and the update preparation set Is further provided with an update processing means for adding as a new registered voice set.

Further, the invention according to claim 5 is the invention according to any one of claims 1 to 4 , wherein the speaker identifying means is a plurality of similarities calculated by the similarity calculating means. When the speaker corresponding to the highest similarity is a speaker candidate of the input speech data, and the highest similarity among a plurality of similarities calculated by the similarity calculation means exceeds a predetermined matching threshold, The apparatus further comprises speaker verification means for determining that the speaker candidate and the speaker of the input voice data are the same person.

The invention according to claim 6 is the invention according to claim 5, wherein the monitoring means for performing a monitoring operation on the monitoring target, the word determination means for determining a word included in the input voice data, and the speaker When the collation unit obtains a collation result that the speaker of the input voice data is the speaker to be identified, the monitoring unit performs the monitoring operation based on the word determined by the word determination unit. And a control means for controlling.

In addition, the invention according to claim 7 is the invention according to any one of claims 1 to 6, wherein the similarity calculation means calculates a small distance between the registered voice data and the input voice data. It is calculated as the height of the similarity.

The invention according to claim 8 is a speaker recognition method for recognizing a speaker of the input voice data based on the input voice data that has been input, and a plurality of registered voice data for the speaker to be identified. Selected in the registered voice data selecting step, a registered voice data selecting step for selecting a predetermined number of registered voice data from the plurality of registered voice data received in the registered voice data receiving step, and the registered voice data selecting step A storage step for storing a predetermined number of registered voice data in the storage unit in association with the speaker, and a similarity calculation for calculating a similarity between the input voice data and the registered voice data stored in the storage unit a step, viewed contains a speaker identification step of identifying the speaker on the basis of the similarity calculated by the similarity calculation step, said Noboru The voice data selection step calculates a variance of the distances of the plurality of registered voice data when the registered voice data is excluded from selection targets for each of the plurality of registered voice data received in the registered voice data receiving step. By excluding the registered voice data so as to minimize the variance, a predetermined number of registered voice data is selected from the plurality of registered voice data received in the registered voice data receiving step .
The invention according to claim 9 is a speaker recognition method for recognizing a speaker of the input voice data based on the input voice data that has been input, and a plurality of registered voice data for the speaker to be identified. Selected in the registered voice data selecting step, a registered voice data selecting step for selecting a predetermined number of registered voice data from the plurality of registered voice data received in the registered voice data receiving step, and the registered voice data selecting step A storage step for storing a predetermined number of registered voice data in the storage unit in association with the speaker, and a similarity calculation for calculating a similarity between the input voice data and the registered voice data stored in the storage unit And a speaker identification step for identifying the speaker based on the similarity calculated by the similarity calculation step, Step, the same predetermined number of data registered voice data of the speaker to the registered voice set, and storing, in association with a plurality of registered voice sets for the same speaker.

According to a tenth aspect of the present invention, there is provided a speaker recognition program for recognizing a speaker of the input voice data based on the input voice data that has been input. Registered voice data receiving procedure, registered voice data selecting procedure for selecting a predetermined number of registered voice data from a plurality of registered voice data received in the registered voice data receiving procedure, and selected in the registered voice data selecting procedure A storage procedure for storing a predetermined number of registered voice data in the storage unit in association with the speaker, and similarity calculation for calculating a similarity between the input voice data and the registered voice data stored in the storage unit and procedures, speaker identification procedure and cause the computer to execute identifying the speaker on the basis of the similarity calculated by the similarity calculation procedures, the registered voice The data selection procedure calculates, for each of the plurality of registered voice data received in the registered voice data reception procedure, a variance of the distances of the plurality of registered voice data when the registered voice data is excluded from the selection target, by dispersion exclude registered voice data to the minimum, characterized by you to select a predetermined number of data registered voice data from a plurality of registered voice data received by the registered voice data reception process.
The invention according to claim 11 is a speaker recognition program for recognizing a speaker of the input voice data based on the input voice data, and a plurality of registered voice data for the speaker to be identified. Registered voice data receiving procedure, registered voice data selecting procedure for selecting a predetermined number of registered voice data from a plurality of registered voice data received in the registered voice data receiving procedure, and selected in the registered voice data selecting procedure A storage procedure for storing a predetermined number of registered voice data in the storage unit in association with the speaker, and similarity calculation for calculating a similarity between the input voice data and the registered voice data stored in the storage unit A storage step for causing the computer to execute a procedure and a speaker identification procedure for identifying the speaker based on the similarity calculated by the similarity calculation procedure; , The same predetermined number of data registered voice data of the speaker to the registered voice set, and storing, in association with a plurality of registered voice sets for the same speaker.

  According to the present invention, a plurality of registered voice data is received for a speaker to be identified, a predetermined number of registered voice data is selected from the plurality of registered voice data, and the selected predetermined number of registered voice data is sent to the speaker. Since the speaker is identified and stored based on the similarity between the input voice data and the registered voice data, the voice data of the person to be collated is held in an appropriate state, so that the accuracy of speaker recognition Can be improved.

FIG. 1 is a system configuration diagram illustrating a system configuration of the home security system according to the embodiment. FIG. 2 is an internal configuration diagram showing an internal configuration of the speaker recognition unit shown in FIG. FIG. 3 is an explanatory diagram for explaining the speaker registration data shown in FIG. FIG. 4 is an explanatory view for explaining the influence of inappropriate registered voice data. FIG. 5 is an explanatory diagram for describing selection of registered voice data. FIG. 6 is an explanatory diagram for explaining generation of an update candidate set. FIG. 7 is an explanatory diagram for explaining replacement of a registered voice set. FIG. 8 is a flowchart showing a processing procedure of the speaker recognition unit in the registration mode. FIG. 9 is a flowchart showing a processing procedure of the audio data selection processing shown in FIG. FIG. 10 is a flowchart illustrating a processing procedure of the speaker recognition unit in the recognition mode. FIG. 11 is a flowchart illustrating a processing procedure for the distance calculation processing illustrated in FIG. 10. FIG. 12 is a flowchart showing the processing procedure of the update process shown in step S308 of FIG. FIG. 13 is an explanatory diagram illustrating a specific example of inappropriate registered voice data. FIG. 14 is an explanatory diagram of the effect of updating the speaker registration data.

  Exemplary embodiments of a speaker recognition device, a speaker recognition method, and a speaker recognition program according to the present invention will be described below in detail with reference to the accompanying drawings. In the following embodiment, a case where the speaker recognition device, the speaker recognition method, and the speaker recognition program according to the present invention are applied to a home security system for a house will be described.

  FIG. 1 is a system configuration diagram illustrating a system configuration of the home security system according to the embodiment. The home security system shown in FIG. 1 has a configuration in which the door monitoring device 11, the window monitoring device 12, the fire detection device 13, and the speaker recognition device 30 are connected to the monitoring device 60, and the microphone 20 is connected to the speaker recognition device 30. Have.

  The door monitoring device 11 is a device that monitors attempts to illegally enter a house door. When the door monitoring device 11 detects an intrusion attempt such as picking, the door monitoring device 11 notifies the monitoring device 60.

  The window monitoring device 12 is a device that monitors unauthorized attempts to enter a residential window. The window monitoring device 12 notifies the monitoring device 60 when it detects an impact or the like on the window.

  The fire detection device 13 is a device that is provided in a living room of a house and detects the occurrence of a fire. The fire detection device 13 notifies the monitoring device 60 when the occurrence of a fire is detected.

  The microphone 20 is a device that is installed at an entrance such as an entrance, acquires an acoustic signal, and outputs it to the speaker recognition device 30. The microphone 20 always operates and acquires and outputs an acoustic signal. Note that acoustic signal acquisition may be switched on and off using a human sensor or the like. The speaker recognition device 30 can be installed at an arbitrary location. Further, the microphone 20 may be provided in the housing of the speaker recognition device 30.

  The speaker recognition device 30 performs speaker recognition using the acoustic signal acquired by the microphone 20 and outputs it to the monitoring device 60 that manages the operation of the home security system. The speaker recognition device 30 includes a speaker recognition unit 31 and a text determination unit 32, and the monitoring device 60 includes a monitoring control unit 33 and a monitoring unit 34. The speaker recognizing unit 31 cuts out a sound from the acoustic signal acquired by the microphone 20, recognizes whether the sound is a resident's sound, and outputs the recognition result to the monitoring control unit 33 of the monitoring device 60. In addition, the text determination unit 32 cuts out sound from the acoustic signal acquired by the microphone 20 and outputs a word in the sound to the monitoring control unit 33 of the monitoring device 60 as text information.

  The monitoring control unit 33 is a processing unit that controls the operation of the monitoring unit 34 based on the text information output from the text determination unit 32 when the speaker recognition unit 31 recognizes that the speaker is a resident. It is. Specifically, when text information such as “security on” or “coming” is included, the monitoring operation by the monitoring unit 34 is started, and when text information such as “security off” or “just now” is included. Terminates the monitoring operation by the monitoring unit 34.

  The monitoring unit 34 is a processing unit that monitors the dwelling using the outputs of the door monitoring device 11, the window monitoring device 12, and the fire detection device 13. Specifically, the monitoring unit 34 starts a monitoring operation when receiving a start instruction from the monitoring control unit 33, and an abnormality occurs from the door monitoring device 11, the window monitoring device 12, or the fire detection device 13 during the monitoring operation. When the notification is received, an alarm operation is performed and an abnormality occurrence is notified to the center. This monitoring operation ends when an end instruction is received from the monitoring control unit 33.

  Thus, in the home security system according to the present embodiment, the on / off control of the monitoring operation can be performed by voice operation by recognizing the voice of the resident.

  Next, the internal configuration of the speaker recognition unit 31 shown in FIG. 1 will be described. FIG. 2 is an internal configuration diagram showing an internal configuration of the speaker recognition unit 31 shown in FIG. As shown in FIG. 2, the speaker recognition unit 31 includes an AD conversion unit 41, a voice segment extraction unit 42, a feature parameter calculation unit 43, a switching unit 44, a registration processing unit 45, a storage unit 46, a distance calculation unit 47, a recognition unit. A processing unit 48 and an update processing unit 49 are included.

  The AD conversion unit 41 is a processing unit that performs a process of converting the acoustic signal acquired by the microphone 20 from an analog signal to a digital signal and outputting the converted signal to the speech segment extraction unit 42.

  The voice segment extraction unit 42 is a processing unit that extracts a voice segment from the acoustic signal converted into a digital signal by the AD conversion unit 41. The extraction of the voice section can be performed based on the signal power of the acoustic signal, the number of zero crosses, and the like.

  The feature parameter calculation unit 43 is a processing unit that calculates a feature parameter indicating the characteristics of the spectral envelope of the speech signal output from the speech segment extraction unit 42. As a feature parameter calculation method, any method such as an LPC (Linear Predictive Coding) cepstrum coefficient or an MFCC (Mel-Frequency Cepstrum Coefficient) can be used.

  The switching unit 44 is a processing unit that switches the operation mode of the speaker recognition unit 31. The operation modes of the speaker recognition unit 31 include a registration mode and a recognition mode. When the registration mode is set by the switching unit 44, the feature parameter calculated by the feature parameter calculation unit 43 is output to the registration processing unit 45 as registered voice data. On the other hand, when the switching unit 44 sets the recognition mode, the feature parameter calculated by the feature parameter calculation unit 43 is output to the distance calculation unit 47 as input voice data.

  The registration processing unit 45 associates the registered resident's registered voice data with the resident who is to be registered, generates speaker registration data, and stores it in the storage unit 46. At this time, the registration processing unit 45 constructs a registered voice set including a predetermined number of registered voice data of the same speaker and associates it with the resident.

  In addition, the registration processing unit 45 includes an audio data selection processing unit 45a. The voice data selection processing unit 45a is a processing unit that selects registered voice data to be stored in the registered voice set from a plurality of registered voice data received at the time of registration. The specific operation of the audio data selection processing unit 45a will be described later.

  The storage unit 46 is a storage device such as a hard disk device or a nonvolatile memory, and stores speaker registration data. In the speaker registration data, information on the resident who is the speaker to be identified, a registered voice set storing a predetermined number of registered voice data of the resident, and registered voice data for updating the registered voice set are stored. Update preparation set. In FIG. 2, the storage unit 46 stores speaker registration data R1 and speaker registration data R2.

  The distance calculation unit 47 is a processing unit that calculates the small distance between the input voice data and the registered voice data as the high degree of similarity. Specifically, the distance calculation unit 47 cuts out a plurality of partial registration voice data from the same registered voice data, cuts out a plurality of partial input voice data from the input voice data, and performs partial registration voice data, partial input voice data, The distance is calculated for each of the combinations, and the minimum distance among the calculated distances is set as the distance between the registered voice data and the input voice data. Note that the average of the calculated plurality of distances may be used as the distance to the registered voice data.

  The distance calculation unit 47 outputs the distance between the input voice data and the registered voice data to the recognition processing unit 48. The distance calculation unit 47 calculates and outputs the distance for each of a plurality of registered voice data stored in the storage unit 46.

  The recognition processing unit 48 includes a speaker identification unit 48a and a speaker verification unit 48b. The speaker identifying unit 48 a selects registered voice data that minimizes the distance calculated by the distance calculating unit 47. The speaker of the registered voice data becomes a speaker candidate of the input voice data.

  The speaker verification unit 48b compares the distance between the registered voice data selected by the speaker identification unit 48a and the input voice data with a verification threshold value. If this distance is smaller than the verification threshold, the speaker verification unit 48b determines that the speaker of the registered voice data matches the speaker of the input voice data. Since the small distance corresponds to the high degree of similarity, the distance being equal to or smaller than the matching threshold means that the similarity is equal to or larger than the predetermined similarity threshold. The speaker verification unit 48 b outputs the determination result to the monitoring device 60 and the update processing unit 49.

  The update processing unit 49 is a processing unit that updates the speaker registration data based on the recognition result in the recognition mode. Specifically, when the speaker verification unit 48b of the recognition processing unit 48 determines that the speaker of the input voice data matches the speaker of the registered voice data, the update processing unit 49 converts the input voice data into the relevant speech. Is stored in the update preparation set as registered voice data for updating the registered voice set of the user. Then, if the registered voice data is sufficiently accumulated in the update preparation set, a predetermined number of registered voice data is selected by the voice data selection processing unit 49a and set as an update candidate set.

  The update processing unit 49 that has generated the update candidate set additionally registers the update candidate set as a registered voice set if the registered voice set in the speaker registration data has not reached the upper limit. If the registered voice set in the speaker registration data has reached the upper limit, the update processing unit 49 selects the voice set by the voice set selection processing unit 49b.

  The voice set selection processing unit 49b determines whether or not there is a registered voice set that has no record of contribution to identification in the existing registered voice set. If there is a registered voice set that does not contribute to identification, the voice set selection processing unit 49b deletes the registered voice set and additionally registers the update candidate set as a registered voice set. If any of the existing registered voice sets has a record of contribution to identification, the voice set selection processing unit 49b resets the update candidate set and the record of contribution to identification.

  FIG. 3 is an explanatory diagram for explaining the speaker registration data shown in FIG. As shown in FIG. 3, the speaker registration data R1 is data in which speaker data, a registered voice set, and an update preparation set are associated with each other.

  The speaker data is information regarding the speaker in the speaker registration data R1, and indicates name, sex, age, and the like. The registered voice set is a set of registered voice data including a predetermined number of registered voice data of the speaker. In FIG. 3, m registered voice sets of the registered voice sets G1 to Gm are included in the speaker registration data R1. The registered voice sets G1 to Gm include (Nn) pieces of registered voice data of the registered voice data v1 to v (Nn), respectively.

  When the speaker registration data R1 is first registered, at least one registered voice set is generated in the registration mode. Thereafter, based on the recognition result in the recognition mode, addition and replacement of the registered voice set are performed.

  The update preparation set G0 is a set of registered voice data that stores registered voice data for adding or replacing a registered voice set. When it is determined that the speaker of the input voice data matches the speaker of the registered voice data in the recognition mode, the input voice data is added to the update preparation set G0 as registered voice data.

  At this time, verification result data indicating the registered voice set to which the registered voice data selected by the speaker identifying unit 48a belongs is generated and associated with the registered voice data added to the update preparation set G0. That is, the verification result data indicates which registered voice set is used to identify the input voice data as registered voice data.

  In FIG. 3, the update preparation set G0 includes N pieces of registered voice data of the registered voice data v1 to vN. The registered voice data v1 to vN are associated with the matching record data d1 to dN, respectively.

  If N pieces of registered voice data are accumulated in the update preparation set G0, a predetermined number (N−n) of registered voice data is selected by the voice data selection processing unit 49a, and an update candidate set is generated.

  Thereafter, if the number of registered voice sets has not reached the upper limit m, the update candidate set is additionally registered as a registered voice set. If the number of registered voice sets has reached the upper limit m, the registered voice set is replaced or the update preparation set is reset based on the matching record data d1 to d (N−n).

  Next, selection of registered voice data will be described. FIG. 4 is an explanatory view for explaining the influence of inappropriate registered voice data. FIG. 4 shows the distance between the registered voice data and the input voice data and the frequency distribution. When the registered voice data speakers and the input voice data speakers are the same, the distance distribution is the “personal distribution”, and the registered voice data speakers and the input voice data speakers are different. The distribution of distance is “others distribution”.

  In order to avoid “rejection of the person” that is recognized as a different person even though they are the same person, it is necessary to increase the verification threshold for speaker verification so that the distribution of the individual is below the verification threshold. It is valid. On the other hand, in order to avoid “acceptance of others” that recognizes the same person even though they are others, it is effective to reduce the collation threshold so that the distribution of others is equal to or greater than the collation threshold.

  As shown in FIG. 4A, if appropriate registered voice data is used and inappropriate registered voice data is not included, the distribution of the principal distribution becomes small and does not overlap with the other person distribution. Therefore, if the distance that distinguishes the person distribution from the other person distribution is used as a collation threshold, both the person rejection rate at which the person rejection occurs and the person acceptance rate at which the other person acceptance occurs can be sufficiently reduced.

  However, as shown in FIG. 4B, if inappropriate registered voice data is included, the distribution of the identity distribution becomes wide and overlaps with the distribution of others, so an appropriate collation threshold value can be set. become unable. As shown in FIG. 4B, when the collation threshold is set so as to sufficiently reduce the identity rejection rate, the acceptance rate of others is increased. Similarly, if you try to reduce the acceptance rate of others, the rejection rate will increase.

  Here, the inappropriate registered voice data is generated, for example, when noise or another person's speaking voice is superimposed on the voice of the person himself / herself. Therefore, when noise or the like is superimposed at the time of registration and inappropriate registration voice data is mixed in the speaker registration data, all subsequent recognitions are affected.

  Therefore, the audio data selection processing unit 45a of the registration processing unit 45 excludes inappropriate registration audio data from the plurality of registration audio data received at the time of registration, and selects registration audio data to be stored in the registration audio set.

  FIG. 5 is an explanatory diagram for describing selection of registered voice data. First, as shown in FIG. 5A, the voice data selection processing unit 45a uses registered voice data v1 to vN obtained by calculating respective feature amounts from voice uttered N times by a speaker to be registered. accumulate.

  Next, the audio data selection processing unit 45a calculates a distance for each combination of the registered audio data v1 to vN as shown in FIG. This distance calculation is the same as the distance calculation by the distance calculation unit 47. The voice data selection processing unit 45a calculates the average of the distances between the registered voice data and other registered voice data. For example, for the registered voice data v1, the average of the distance between the registered voice data v1 and the registered voice data v2 to vN is calculated, and for the registered voice data v2, the distance between the registered voice data v2 and the registered voice data v1, v3 to vN. The average is calculated. Then, those having the maximum distance are excluded from the selection targets. In FIG. 5B, the registered voice data v5 is excluded from the selection targets.

  The voice data selection processing unit 45a repeats such processing, and when the number of selection targets reaches a predetermined number (N−n), selects the remaining registered voice data as registered voice data to be stored in the registered voice set.

  Although the average distance is used here, the total distance may be used. Further, a threshold may be set for the average or total distance, and registered voice data exceeding the threshold may be excluded from selection targets.

  Further, the dispersion of distances may be used when selecting the registered voice data. In this case, first, the registered voice data v1 to vN are accumulated, a part thereof is excluded, and distances are obtained for all combinations of other registered voice data, and their variances are calculated. For other data as well, variance is calculated as an exclusion candidate. As a result, as shown in FIG. 5C, the dispersion of distance increases if inappropriate registered speech data remains, and the dispersion of distance decreases when inappropriate registered speech data is an exclusion candidate. . Therefore, the candidate for exclusion when the variance value is the smallest is excluded, and the remaining registered voice data is selected as registered voice data to be stored in the registered voice set. The number of repeated utterances N for registration is 3 or more, the number n of removal candidate data is 1 or more, and N−n is 2 or more.

  Next, update of the registered voice set will be described. Even if appropriate registered voice data can be obtained at the time of registration, if the speaker's voice changes due to changes over time or the like, the recognition accuracy will deteriorate. Therefore, the update processing unit 49 updates the registered voice set based on the recognition result in the recognition mode.

  Specifically, the update processing unit 49 updates the input voice data as the registered voice data of the speaker when it is determined that the speaker of the input voice data matches the speaker of the registered voice data in the recognition mode. Add to preparation set G0.

  Further, the registered voice set to which the registered voice data selected by the speaker identifying unit 48a belongs, that is, verification result data indicating the registered voice set that contributes to the identification is generated and associated with the registered voice data added to the update preparation set G0.

  As shown in FIG. 6, when N pieces of registered voice data are accumulated in the update preparation set G0, the voice data selection processing unit 49a selects a predetermined number (N−n) of registered voice data and updates candidates. Generate a set.

  Thereafter, if the number of registered voice sets has not reached the upper limit m, the update candidate set is additionally registered as a registered voice set. If the number of registered voice sets has reached the upper limit m, the voice set selection processing unit 49b selects a voice set.

  The voice set selection processing unit 49b refers to the matching record data d1 to d (N−n) and determines whether or not there is a registered voice set that has no record of contribution to identification. If there is a registered voice set that does not contribute to identification, the voice set selection processing unit 49b deletes the registered voice set and additionally registers the update candidate set as a registered voice set. If any of the existing registered voice sets has a record of contribution to identification, the voice set selection processing unit 49b resets the update candidate set and the record of contribution to identification.

  In the example illustrated in FIG. 7, the matching record data d1 to d4 indicate any of the registered voice sets G1, G2, and G3, and the registered voice set G2 has no identification record. Therefore, the registered voice set G2 is deleted, and the registration candidate set is stored as a new registered voice set G2.

  Next, the processing procedure of the speaker recognition unit 31 will be described. FIG. 8 is a flowchart showing a processing procedure of the speaker recognition unit 31 in the registration mode. Note that the processing procedure shown in this flowchart is executed in a state in which the switching unit 44 sets the registration mode.

  First, the microphone 20 acquires an acoustic signal (step S101). The voice segment extraction unit 42 extracts a voice segment from the acoustic signal acquired by the microphone 20 (step S102).

  The feature parameter calculation unit 43 cuts out a plurality of partial speech signals from the speech signal in the speech section, and calculates feature parameters indicating the characteristics of the spectral envelope of the speech signal (step S103). The registration processing unit 45 accumulates the calculated feature parameters as registered voice data (step S104), and determines whether the number of registered voice data has become N (step S105). If the number of registered voice data is less than N (step S105; No), the process proceeds to step S101, and an acoustic signal of the next utterance is acquired.

  If the number of registered voice data is N (step S105; Yes), the voice data selection processing unit 45a selects voice data for selecting (N−n) registered voice data from the N registered voice data. A selection process is performed (step S106). Thereafter, the selected (N−n) pieces of registered voice data are stored in the registered voice set G1 (step S107), and the process ends.

  Next, the audio data selection process shown in FIG. 8 will be described. FIG. 9 is a flowchart showing a processing procedure of the audio data selection processing shown in FIG. The voice data selection processing unit 45a first calculates a distance for each combination of the registered voice data v1 to vN (step S201). Then, the voice data selection processing unit 45a calculates the average distance of each registered voice data from the other registered voice data (step S202), and excludes the registered voice data having the maximum distance from the selection target. (Step S203).

  After step S203, the voice data selection processing unit 45a determines whether or not the number of registered voice data removed is n (step S204). If the number of registered voice data removal has not reached n (step S204; No), the process proceeds to step S201, and the distance is calculated for each combination of the remaining registered voice data. If the number of removed registered voice data is n (step S204; Yes), the remaining (N−n) registered voice data are selected (step S205), and the voice data selection process is terminated.

  FIG. 10 is a flowchart showing a processing procedure of the speaker recognition unit 31 in the recognition mode. Note that the processing procedure shown in this flowchart is executed in a state where the switching unit 44 sets the recognition mode.

  First, the microphone 20 acquires an acoustic signal (step S301). The voice segment extraction unit 42 extracts a voice segment from the acoustic signal acquired by the microphone 20 (step S302).

  The feature parameter calculation unit 43 calculates a feature parameter indicating the characteristics of the spectral envelope of the speech segment (step S303).

  The distance calculation unit 47 calculates the distance from the input voice data for all the speaker registration data (step S304). The speaker identification unit 48a identifies the speaker registration data with the shortest distance calculated by the distance calculation unit 47 (step S305).

  The speaker verification unit 48b compares the distance between the speaker registration data specified by the speaker identification unit 48a and the input voice data, that is, the minimum value of the distance and the verification threshold value (step S306). If the minimum value of the distance is smaller than the verification threshold (step S306; Yes), the speaker verification unit 48b determines that the speaker of the speaker registration data matches the speaker of the input data, and the speaker The speaker data of the registered data is output as a verification result to the monitoring device 60 and the update processing unit 49 (step S307). After step S307, the update processing unit 49 performs an update process (step S308) and ends the process.

  On the other hand, if the minimum value of the distance is equal to or greater than the collation threshold value (step S306; No), the speaker collation unit 48b outputs to the monitoring device 60 that the input voice data does not correspond to the collation result and ends the process. (Step S309).

  Next, the distance calculation process shown in step S304 of FIG. 10 will be described. FIG. 11 is a flowchart illustrating a processing procedure for the distance calculation processing illustrated in FIG. 10. The distance calculation unit 47 first selects speaker registration data (step S401). Then, registered voice data is selected from the selected speaker registration data (step S402), and a distance between the selected registered voice data and input voice data is calculated (step S403).

  After step S403, the distance calculation unit 47 determines whether all registered voice data in the speaker registration data selected in step S401 has been selected (step S404). As a result, if unselected registered voice data remains (step S404; No), the process proceeds to step S402, and the registered voice data is selected.

  If all the registered voice data within the same speaker registration data have been selected (step S404; Yes). The distance calculation unit 47 sets the minimum distance among the distances calculated for all the registered voice data in the same speaker registration data as the distance of the speaker registration data (step S405).

  After step S405, the distance calculation unit 47 determines whether all the speaker registration data has been selected (step S406). As a result, if unselected speaker registration data remains (step S406). Step S406; No), the process proceeds to step S401, and speaker registration data is selected. If all the speaker registration data has been selected (step S406; Yes), the distance calculation process ends.

  Next, the update process shown in step S308 of FIG. 10 will be described. FIG. 12 is a flowchart showing a processing procedure of the update process shown in FIG. When it is determined in the recognition mode that the speaker of the input voice data matches the speaker of the registered voice data, the update processing unit 49 adds the input voice data to the update preparation set G0 as the registered voice data of the speaker. And accumulates (step S501).

  Then, the registered voice set to which the registered voice data selected by the speaker identifying unit 48a belongs, that is, the verification result data indicating the registered voice set contributing to the identification, is generated and associated with the registered voice data added to the update preparation set G0 ( Step S502).

  The update processing unit 49 determines whether or not the number of registered voice data stored in the update preparation set G0 is N (step S503). If the accumulated number of registered voice data has not reached N (step S503; No), the update process is terminated as it is.

  If the stored number of registered voice data is N (step S503; Yes), the voice data selection processing unit 49a performs a voice data selection process to select a predetermined number (N−n) of registered voice data ( Step S504). This audio data selection process is the same as the audio data selection process shown in FIG.

  The update processing unit 49 sets the (N−n) registered voice data selected in step S504 as an update candidate set (step S505), and determines whether the number of registered voice sets is m (step S506).

  If the number of registered voice sets is m (step S506; Yes), the voice set selection processing unit 49b selects a voice set. Specifically, the voice set selection processing unit 49b refers to the matching record data d1 to d (N−n) and determines whether or not there is a registered voice set that does not contribute to the identification (step) S509). When there is a registered voice set that does not contribute to identification (step S509; Yes), the voice set selection processing unit 49b deletes the registered voice set (step S510).

  After step S510, or when the number of registered voice sets has not reached m (step S506; No), the update processing unit 49 adds an update candidate set as a registered voice set (step S507).

  After step 507 or when there is no registered voice set with no record of contribution to identification (step S509; No), the update preparation set G0 is reset (step S508), and the update process is terminated. When the update preparation set G0 is reset, the registration candidate set and verification result data are also reset.

  Next, a specific example of inappropriate registered voice data will be described. FIG. 13 is an explanatory diagram illustrating a specific example of inappropriate registered voice data. FIG. 13 shows the average distance from each registered voice data to the other registered voice data, with the number of utterances being four (N = 4).

  The data example H0 shown in FIG. 13 shows a case where all of the registered voice data v1 to v4 are appropriate registered voice data and no registered voice data on which noise is superimposed is included. The data example H1 shows a case where noise is superimposed on the registered voice data v1, the data example H2 shows a case where noise is superimposed on the registered voice data v2, and the data example H3 is a case where noise is superimposed on the registered voice data v3. The data example H4 shows a case where noise is superimposed on the registered voice data v4.

  As shown in the data examples H0 to H4, the registered voice data on which noise is superimposed has a larger distance than the others. For this reason, it is possible to improve the accuracy during identification by identifying and removing in advance the inappropriate registered voice data based on the distance.

  Next, the effect of updating the speaker registration data will be described. FIG. 14 is an explanatory diagram of the effect of updating the speaker registration data. Specifically, a case where an adult male utters the same content seven times in the morning, noon, and evening three times a day for 52 weeks (one year) is shown.

  First, a registered voice set is generated with registered voice data for five times out of the first seven times. Using only this registered voice set, the verification rate for one year was 79.7%. Next, in the case where up to four registered voice sets were added using the recognition result in the recognition mode and no subsequent replacement was performed, the verification rate for one year was 97.6%. And in the case where the replacement by the verification results was performed after the four registered voice sets were generated, the verification rate for one year improved to 99.1%.

  As described above, in this embodiment, the speaker recognition unit 31 selects a predetermined number of registered voice data from a plurality of registered voice data to construct a registered voice set, and handles the registered voice set for the speaker. The speaker is identified based on the similarity between the input voice data and the registered voice data in the registered voice set. For this reason, inappropriate data can be removed from the registered voice data, thereby improving the accuracy of speaker recognition.

  Further, by adding or updating the registered voice set based on the recognition result in the recognition mode, it is possible to suppress a reduction in recognition accuracy even when the voice feature changes due to a change over time or the like.

  Furthermore, by accumulating verification record data indicating the registered voice set that contributed to the identification, by deleting the registered voice set indicated by the verification result data that the contribution to the identification is small, and adding a new registered voice set It is possible to keep the recognition accuracy high.

  The registered voice set may be deleted when it has not been used for verification for a predetermined period of time. Further, a distance may be calculated between registered voice sets, and it may be determined whether or not to delete depending on the magnitude of the distance.

  In the above embodiment, the case where the home security operation mode is switched by voice operation has been described. However, the speaker recognition according to the present invention is not limited to the operation mode switching, and various operations can be performed by text discrimination. It is applicable to.

  In the above embodiment, the configuration in which the security operation mode is switched on the condition that the speaker verification is successful is shown. However, the voice of a specific speaker is registered as a black list, and the speaker who is registered in the black list is registered. You may comprise so that operation by may be refused.

  The present invention is not limited to home security, and can be applied to speaker recognition of an arbitrary device such as speaker recognition using a mobile phone terminal. Also, it can be applied to “transfer fraud countermeasures” by speaker recognition via a telephone line, “identification by voice through intercom”, and the like.

  Each illustrated configuration is schematic in function, and does not necessarily need to be physically configured as illustrated. In other words, the form of distribution / integration of each device is not limited to the one shown in the figure, and all or a part thereof may be functionally / physically distributed / integrated in arbitrary units according to various loads and usage conditions. Can be configured. Moreover, if the function part of the speaker recognition part 31 is implement | achieved by software and it makes a computer perform, the speaker recognition program which makes a computer operate | move as a speaker recognition apparatus can be obtained.

  As described above, the speaker recognition device, the speaker recognition method, and the speaker recognition program are suitable for maintaining the voice data of the verification target person in an appropriate state and improving the accuracy of speaker recognition.

DESCRIPTION OF SYMBOLS 11 Door monitoring apparatus 12 Window monitoring apparatus 13 Fire detection apparatus 20 Microphone 30 Speaker recognition apparatus 31 Speaker recognition part 32 Text discrimination | determination part 33 Monitoring control part 34 Monitoring part 41 AD conversion part 42 Voice area extraction part 43 Feature parameter calculation part 44 Switching unit 45 registration processing unit 45a, 49a voice data selection processing unit 46 storage unit 47 distance calculation unit 48 recognition processing unit 48a speaker identification unit 48b speaker verification unit 49 update processing unit 49b voice set selection processing unit
60 Monitoring device

Claims (11)

A speaker recognition device for recognizing a speaker of input voice data based on input voice data,
A registered voice data receiving means for receiving a plurality of registered voice data for a speaker to be identified;
Registered voice data selecting means for selecting a predetermined number of registered voice data from a plurality of registered voice data received by the registered voice data receiving means;
Storage means for storing a predetermined number of registered voice data selected by the registered voice data selection means in association with the speaker;
Similarity calculation means for calculating the similarity between the input voice data and the registered voice data stored in the storage means;
Speaker identification means for identifying the speaker based on the similarity calculated by the similarity calculation means,
The registered voice data selection unit calculates, for each of the plurality of registered voice data received by the registered voice data reception unit, a variance of the distances of the plurality of registered voice data when the registered voice data is excluded from selection targets. And a predetermined number of registered voice data is selected from the plurality of registered voice data received by the registered voice data receiving means by excluding the registered voice data so as to minimize the variance. apparatus.   2. The storage unit according to claim 1, wherein the storage unit stores registered voice data of a predetermined number of data of the same speaker as a registered voice set, and stores a plurality of registered voice sets in association with the same speaker. Speaker recognition device. A speaker recognition device for recognizing a speaker of input voice data based on input voice data,
A registered voice data receiving means for receiving a plurality of registered voice data for a speaker to be identified;
Registered voice data selecting means for selecting a predetermined number of registered voice data from a plurality of registered voice data received by the registered voice data receiving means;
Storage means for storing a predetermined number of registered voice data selected by the registered voice data selection means in association with the speaker;
Similarity calculation means for calculating the similarity between the input voice data and the registered voice data stored in the storage means;
Speaker identification means for identifying the speaker based on the similarity calculated by the similarity calculation means,
The said memory | storage means makes registration voice data of the predetermined data number of the same speaker into a registration voice set, and associates and memorize | stores several registration voice sets about the same speaker, The speaker recognition apparatus characterized by the above-mentioned.   When identification by the speaker identification unit is performed, the record data indicating the registered voice set that contributes to the identification is accumulated, and the input voice data is stored as registered voice data of the update preparation set, and the record data The update processing means for deleting the registered voice set indicated as having a small contribution to the identification and adding the update preparation set as a new registered voice set according to claim 1 further comprising: The speaker recognition device according to any one of the above. The speaker identifying means sets a speaker corresponding to the highest similarity among a plurality of similarities calculated by the similarity calculating means as a speaker candidate of the input voice data,
When the highest similarity among a plurality of similarities calculated by the similarity calculating means exceeds a predetermined collation threshold, it is determined that the speaker candidate and the speaker of the input voice data are the same person The speaker recognition apparatus according to claim 1, further comprising speaker verification means for Monitoring means for performing a monitoring operation on the monitoring target;
Word determination means for determining a word included in the input voice data;
When the collation result that the speaker of the input voice data is the speaker to be identified is obtained by the speaker collating unit, the monitoring unit is based on the word determined by the word determining unit. The speaker recognition apparatus according to claim 5 , further comprising: a control unit that controls the monitoring operation. The said similarity calculation means calculates the small distance of the said registration audio | voice data and the said input audio | speech data as the height of the said similarity, The one of Claims 1-6 characterized by the above-mentioned. Speaker recognition device. A speaker recognition method for recognizing a speaker of input voice data based on input voice data,
A registered voice data receiving step for receiving a plurality of registered voice data for a speaker to be identified;
A registered voice data selecting step for selecting a predetermined number of registered voice data from the plurality of registered voice data received in the registered voice data receiving step;
A storing step of storing a predetermined number of registered voice data selected in the registered voice data selection step in the storage unit in association with the speaker;
A similarity calculation step of calculating a similarity between the input voice data and the registered voice data stored in the storage unit;
A speaker identification step for identifying the speaker based on the similarity calculated by the similarity calculation step,
In the registered voice data selection step, for each of the plurality of registered voice data received in the registered voice data reception step, a variance of the distances of the plurality of registered voice data when the registered voice data is excluded from selection targets is calculated. And a predetermined number of registered voice data is selected from the plurality of registered voice data received in the registered voice data receiving step by excluding the registered voice data so as to minimize the variance. Method. A speaker recognition method for recognizing a speaker of input voice data based on input voice data,
A registered voice data receiving step for receiving a plurality of registered voice data for a speaker to be identified;
A registered voice data selecting step for selecting a predetermined number of registered voice data from the plurality of registered voice data received in the registered voice data receiving step;
A storing step of storing a predetermined number of registered voice data selected in the registered voice data selection step in the storage unit in association with the speaker;
A similarity calculation step of calculating a similarity between the input voice data and the registered voice data stored in the storage unit;
A speaker identification step for identifying the speaker based on the similarity calculated by the similarity calculation step,
In the speaker recognition method, the storing step includes storing registered voice data of a predetermined number of data of the same speaker as a registered voice set, and storing a plurality of registered voice sets in association with the same speaker. A speaker recognition program for recognizing a speaker of input voice data based on input voice data,
Registered voice data reception procedure for receiving a plurality of registered voice data for a speaker to be identified;
A registered voice data selection procedure for selecting a predetermined number of registered voice data from a plurality of registered voice data received in the registered voice data reception procedure;
A storing procedure for storing a predetermined number of registered voice data selected in the registered voice data selection procedure in the storage unit in association with the speaker;
A similarity calculation procedure for calculating a similarity between the input voice data and the registered voice data stored in the storage unit;
Causing the computer to execute a speaker identification procedure for identifying the speaker based on the similarity calculated by the similarity calculation procedure,
In the registered voice data selection procedure, for each of a plurality of registered voice data received in the registered voice data reception procedure, a variance of distances of the plurality of registered voice data when the registered voice data is excluded from selection targets is calculated. And a predetermined number of registered voice data is selected from the plurality of registered voice data received in the registered voice data receiving procedure by excluding the registered voice data so as to minimize the variance. program. A speaker recognition program for recognizing a speaker of input voice data based on input voice data,
Registered voice data reception procedure for receiving a plurality of registered voice data for a speaker to be identified;
A registered voice data selection procedure for selecting a predetermined number of registered voice data from a plurality of registered voice data received in the registered voice data reception procedure;
A storing procedure for storing a predetermined number of registered voice data selected in the registered voice data selection procedure in the storage unit in association with the speaker;
A similarity calculation procedure for calculating a similarity between the input voice data and the registered voice data stored in the storage unit;
Causing the computer to execute a speaker identification procedure for identifying the speaker based on the similarity calculated by the similarity calculation procedure,
The storage procedure is a speaker recognition program characterized in that registered voice data of a predetermined number of data of the same speaker is used as a registered voice set, and a plurality of registered voice sets are stored in association with each other for the same speaker.

Images