JP6087542B2 - 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 voice data based on 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.

Japanese Patent Laying-Open No. 2005-091758

  However, in the above-described conventional technology, it is necessary to learn long-spoken speech in order to construct registered model data, and complicated calculations are required during model construction and speaker recognition using the model. Therefore, there is a problem that it is difficult to perform high-speed processing at low cost.

  In addition, in the conventional technology described above, a model for each person to be collated is individually constructed. However, in order to individually construct a model for each person to be collated, the voices of each person to be collated are used. Since it is necessary to collect them separately, there is a problem that registration of a person to be verified becomes complicated.

  In the above-described conventional technique, since a series of utterances is divided into N frames and the average value of the calculated feature parameters is used as an index, speaker recognition is performed after the series of utterances is completed. Thus, there is a problem that it takes time to recognize.

  For this reason, it has been important to improve the convenience of speaker recognition by realizing inexpensive and high-speed processing, simplifying registration processing, shortening the time until recognition, and the like. For example, considering the case where the security operation of the home security system is operated by the user's voice, inexpensive and high-speed processing is required to suppress the cost required for system introduction. Moreover, it is sufficient for the confirmation of the operation authority that it is sufficient to determine that the voice is one of a plurality of residents, and it is required that the voices of the plurality of residents can be registered easily in a lump. Furthermore, it is also required to determine the speaker at an earlier timing and to quickly control the guard operation.

  The present invention has been made to solve the above-described problems of the prior art, and provides a speaker recognition device, a speaker recognition method, and a speaker recognition program that improve the convenience of registration and recognition. With the goal.

In order to solve the above-described problems and achieve the object, the invention described in claim 1 is a speaker recognition device for recognizing a speaker of the voice data based on the voice data. Similarity calculating means for calculating the similarity between the spectral envelope of the partially registered speech data cut out from the included registered speech data and the spectral envelope of the partially input speech data cut out from the input speech data to be recognized; Recognition means for recognizing a speaker of the input voice data based on the degree of similarity calculated by the degree calculation means, and the registered voice data includes voices of a plurality of registration subjects, and the partial registration voice data Are cut out so as to include the voices of each registration target person, and the similarity calculation means performs the partial input voice for each spectrum envelope of the plurality of partial registration voice data. Calculating a similarity between the spectral envelope of over data respectively, said recognition means, speaker of the input voice data matches one of said plurality of registered person, or the input speech data speaker plurality characterized that you outputs the recognition result of the does not match any of the registered person.

According to a second aspect of the present invention, there is provided a speaker recognition device for recognizing a speaker of voice data based on the voice data, and the partially registered voice cut out from the registered voice data including the voice of the registration target person. Similarity calculation means for calculating the similarity between the spectral envelope of the data and the spectral envelope of the partial input voice data cut out from the input voice data to be recognized, and based on the similarity calculated by the similarity calculation means Recognizing means for recognizing a speaker of the input voice data, wherein the registered voice data includes voices of a plurality of registration subjects, and the similarity calculation means is cut out from the same registered voice data. For a plurality of partial registration voice data, a feature parameter indicating a spectral envelope of each partial registration voice data and a characteristic indicating a spectral envelope of the partial input voice data The distance from the parameter is calculated as the degree of similarity of the partial input voice data with respect to each partial registration voice data, and the minimum distance of the partial input voice data with respect to each partial registration voice data is calculated as the minimum value. The distance of the partial input voice data with respect to the registered voice data is used, and the recognition means uses the distance of the partial input voice data with respect to the registered voice data as the similarity, so that a speaker of the input voice data can register the plurality of registration voice data. It matches any of the subjects, or the speaker of the input speech data and outputting a recognition result of not match any of the plurality of registered persons.

The invention according to claim 3 is the invention according to claim 1 or 2, wherein the similarity calculation means calculates a similarity with respect to the input voice data for each of the plurality of registered voice data, and performs the recognition. The means estimates that the speaker of the input voice data matches one of a plurality of registration target persons included in the registered voice data having the highest similarity to the input voice data among the plurality of registered voice data. It is characterized by.

According to a fourth aspect of the present invention, in the invention according to any one of the first to third aspects, the recognition means has a similarity of the registered voice data to the input voice data equal to or greater than a similarity threshold. In some cases, it is determined that a speaker of the input voice data matches one of a plurality of registration target persons included in the registered voice data.

The invention according to claim 5 is the invention according to any one of claims 1 to 4, wherein the plurality of partial registration voice data obtained from the registration voice data including the voices of the plurality of registration subjects. And classifying means for calculating a representative value of the feature parameter for each classification, and the similarity calculation means includes the partial input speech. A distance between a characteristic parameter indicating a spectral envelope of data and a representative value of each classification is calculated, and each partially registered speech data belonging to the classification having a minimum distance from the representative value is used for calculating the similarity. It is characterized by that.

The invention according to claim 6 is the invention according to any one of claims 1 to 5 , wherein the monitoring means for performing the monitoring operation on the monitoring target, and the word for determining the word included in the input voice data A discriminating unit; and a control unit that controls a monitoring operation of the monitoring unit based on a word discriminated by the word discriminating unit when a recognition result by the recognizing unit satisfies a predetermined condition. Features.

The invention according to claim 7 is a speaker recognition method for recognizing a speaker of the voice data based on the voice data, and each registration target person from the registered voice data including a plurality of registration target person voices. A part of the registered voice data that is extracted to include the voice of the registered voice data processing step for obtaining the spectral envelope of the extracted partial registered voice data, and the part of the input voice data that has been cut out from the input voice data to be recognized An input speech data processing step for obtaining a spectral envelope of the input speech data; and a similarity calculation step for calculating a similarity between the spectral envelopes of the partial input speech data for each of the plurality of partial registered speech data; Based on the similarity calculated in the similarity calculation step, a speaker of the input voice data Matches one of a plurality of registered person, or the speaker of the input voice data, characterized in that the containing and recognition step of outputting a recognition result of not match any of the plurality of registered persons.
The invention according to claim 8 is a speaker recognition method for recognizing a speaker of the voice data based on the voice data, and the partial registration voice data from the registered voice data including the voices of a plurality of registration subjects. A registered voice data processing step for obtaining a spectral envelope of the cut partial registered voice data, and an input for cutting the partial input voice data from the input voice data to be recognized and obtaining the spectral envelope of the cut partial input voice data A plurality of partially registered speech data extracted from the same registered speech data, a feature parameter indicating a spectral envelope of each partially registered speech data, and a feature parameter indicating a spectrum envelope of the partially input speech data The partial input voice data for each partial registered voice data Respectively, the minimum value of the distance of the partial input voice data with respect to each of the partial registered voice data is set as the distance of the partial input voice data with respect to the registered voice data, and the distance is set as the similarity. Based on the similarity calculation step to be output and the similarity calculated by the similarity calculation step, a speaker of the input voice data matches one of the plurality of registration target persons , or a story of the input voice data And a recognition step of outputting a recognition result that the person does not match any of the plurality of registration subjects .

The invention according to claim 9 is a speaker recognition program for recognizing a speaker of the voice data based on the voice data, and each registration target person from the registered voice data including voices of a plurality of registration target persons. A part of the registered speech data processing procedure for obtaining the spectral envelope of the segmented speech data that has been cut out so as to include the voice, and the portion that has been cut out from the input speech data to be recognized An input speech data processing procedure for obtaining a spectrum envelope of input speech data, and a similarity calculation procedure for calculating a similarity between each of the plurality of partial registered speech data and a spectrum envelope of the partial input speech data, on the basis of the similarity calculated by the similarity calculation procedure, Noboru speaker of the input voice data of the plurality It matches any of the subjects, or the speaker of the input voice data, characterized in that to perform the certified識手order and you outputs the recognition result of not match any of the plurality of registered person on the computer .
According to a tenth aspect of the present invention, there is provided a speaker recognition program for recognizing a speaker of the voice data based on the voice data, and the partial registration voice data from the registered voice data including voices of a plurality of registration subjects. A registered voice data processing procedure for obtaining a spectral envelope of the extracted partial registered voice data, and an input for extracting the partial input voice data from the input voice data to be recognized and obtaining the spectral envelope of the extracted partial input voice data For a plurality of partially registered speech data extracted from the same registered speech data, a feature parameter indicating a spectrum envelope of each partially registered speech data and a feature parameter indicating a spectrum envelope of the partially input speech data The partial input voice data for each partial registered voice data A similarity calculation procedure for calculating a minimum value of a distance to each of the partial registered voice data as a distance of the partial input voice data with respect to the registered voice data, and outputting the distance as a similarity And, based on the similarity calculated by the similarity calculation procedure, a speaker of the input speech data matches any of the plurality of registration subjects , or a speaker of the input speech data is registered in the plurality of registrations. It is characterized by causing a computer to execute a recognition procedure for outputting a recognition result indicating that it does not match any of the target persons .

  According to the present invention, the similarity between the spectral envelope of the partially registered speech data extracted from the registered speech data including the speech of the person to be registered and the spectral envelope of the partially input speech data extracted from the input speech data to be recognized. Since the degree is calculated and the speaker of the input voice data is recognized based on the degree of similarity, the convenience of registration and recognition can be improved.

FIG. 1 is a system configuration diagram illustrating a system configuration of the home security system according to the first 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 concept of distance calculation. FIG. 4 is an explanatory diagram for explaining the collation threshold used by the speaker collation unit. FIG. 5 is a flowchart showing a processing procedure of the speaker recognition unit in the registration mode. FIG. 6 is a flowchart showing a processing procedure of the speaker recognition unit in the recognition mode. FIG. 7 is an explanatory diagram for explaining an experiment result of speaker recognition according to the first embodiment. FIG. 8 is an internal configuration diagram illustrating an internal configuration of the speaker recognition unit according to the second embodiment. FIG. 9 is an explanatory diagram of the minimum distance search using clusters.

  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 first and second embodiments, 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 first 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.

  As described above, in the home security system according to the first embodiment, the on / off control of the monitoring operation can be performed by the 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 speech segment extraction unit 42, a feature parameter calculation unit 43, a switching unit 44, a storage unit 45, a minimum distance search unit 46, and a recognition processing unit 47. Have.

  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 extracts a plurality of partial speech signals from the speech signal output from the speech segment extraction unit 42 and calculates feature parameters indicating the characteristics of the spectral envelope of the speech signal. 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 stored as registration data in the storage unit 45. 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 minimum distance search unit 46 as input data.

The storage unit 45 is a storage device such as a hard disk device or a nonvolatile memory, and stores registration data. The registration data is generated every time registration processing is performed and stored as separate data. In FIG. 2, the storage unit 45 stores registration data R ¹ and registration data R ² . The feature parameter included in the registration data may include only a single speaker or may include a plurality of speakers.

となる。ここで、分析フレームとは、登録音声データから部分登録音声データを切り出すための、一定のフレーム長の切り出し範囲である。すなわち、Ｎ個の登録音声データのそれぞれから、Ｍ個の部分登録音声データが切り出され、部分登録音声データのそれぞれについて算出された次数Ｋの特徴パラメータが登録データＲである。 The minimum distance search unit 46 is a processing unit that calculates the small distance between the input data and each registered data as the high degree of similarity. Specifically, R which is a characteristic parameter of registration data is

It becomes. Here, the analysis frame is a cutout range having a certain frame length for cutting out the partially registered voice data from the registered voice data. That is, M pieces of partial registration voice data are cut out from each of the N pieces of registration voice data, and the feature parameter of the order K calculated for each of the partial registration voice data is the registration data R.

となる。すなわち、入力音声データからは、Ｌ個の部分入力音声データが切り出され、部分入力音声データのそれぞれについて算出された次数Ｋの特徴パラメータが入力データである。 Also, the feature parameter of the input data is

It becomes. That is, L pieces of partial input voice data are cut out from the input voice data, and the feature parameter of the order K calculated for each of the partial input voice data is the input data.

により算出する。図３は、距離算出の概念を説明するための説明図である。図３に示すように、入力データの各フレームについて、登録データの全フレームに対する特徴パラメータとの距離を総当たりで算出し、入力フレーム毎の最小距離の平均値を、入力データと登録データとの距離とする。 The distance d between the input data and the registration data R is

Calculated by FIG. 3 is an explanatory diagram for explaining the concept of distance calculation. As shown in FIG. 3, for each frame of the input data, the distance from the feature parameter to all the frames of the registration data is calculated brute force, and the average value of the minimum distance for each input frame is calculated between the input data and the registration data. Distance.

により求められる。最小距離探索部４６は、入力データに対して最も距離が小さい登録データと、その距離を認識処理部４７に出力する。 The registered data I having the smallest distance to the input data and its distance d are

Is required. The minimum distance search unit 46 outputs the registered data having the smallest distance to the input data and the distance to the recognition processing unit 47.

  The recognition processing unit 47 illustrated in FIG. 2 includes a speaker identification unit 47a and a speaker verification unit 47b. The speaker identification unit 47a estimates that the speaker of the registered data having the shortest distance from the input data is the same as the speaker of the input voice data.

  The speaker verification unit 47b compares the distance of the registration data having the smallest distance with respect to the input data with a verification threshold, and if the distance is equal to or smaller than the verification threshold, the speaker of the registration data and the input data It is determined that the person matches the person. 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.

  Next, the verification threshold used by the speaker verification unit 47b will be described. FIG. 4 is an explanatory diagram for explaining a matching threshold used by the speaker matching unit 47b. When obtaining the collation threshold, the distance between registered data is calculated in advance, and as shown in FIG. 4A, a different story from the intra-speaker distance distribution, which is a distance distribution when the same speaker is used. And the inter-speaker distance distribution, which is the distance distribution for the speaker.

  From this intra-speaker distance distribution and inter-speaker distance distribution, as shown in FIG. 4 (b), an error rate when collating the speakers is obtained. If the collation threshold value is reduced, that is, the judgment is made stricter, the acceptance rate of others who mistakenly accept others decreases, but the rejection rate of falsely rejecting the principals increases. In view of this, it is preferable to set a value at which the acceptance rate of others and the rejection rate of the person coincide with each other as a collation threshold. If necessary, adjustments such as reducing the acceptance rate of others may be performed on the verification threshold.

  Next, the processing procedure of the speaker recognition unit 31 will be described. FIG. 5 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). Then, the calculated feature parameters are accumulated as registration data in the storage unit 45 (step S104), and the registration process is terminated.

  FIG. 6 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 S201). The voice segment extraction unit 42 extracts a voice segment from the acoustic signal acquired by the microphone 20 (step S202).

  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 S203).

  The minimum distance search unit 46 calculates the distance between the input data and each registration data, and searches for the registration data having the smallest distance with respect to the input data and the distance (step S204). The recognition processing unit 47 estimates that the speaker of the registered data with the shortest distance from the input data is the same as the speaker of the input voice data, and if the distance is equal to or smaller than the matching threshold, It is determined that the speaker and the speaker of the input data match (step S205), the estimation and determination results are output to the monitoring control unit 33 (step S206), and the recognition process is terminated.

  Next, an experiment result of speaker recognition according to the present embodiment will be described. FIG. 7 is an explanatory diagram for explaining an experiment result of speaker recognition according to the first embodiment. As shown in FIG. 7, four types of registration data (registration voice) are used in increments of 5 seconds from 5 seconds to 20 seconds, and the length of input data (input voice) is 0.1 to 1.5 seconds. Speaker identification and speaker verification experiments were performed using 15 types in increments of 0.1 seconds to seconds. LPC cepstrum coefficients (analysis frame length 32 ms, analysis frame shift 16 ms, order 32) are used for speech feature parameter calculation, and the verification threshold in speaker verification is the same as the rejection rate and the acceptance rate of others. The distance to be.

  FIG. 7A shows the experimental results of speaker identification. As shown in FIG. 7A, if the utterance time length of the input voice is 0.1 second, the average speaker identification rate is 82% when the utterance time length of the registered voice is 5 seconds, The average speaker identification rate is 86% when the utterance time length is 10 seconds, the average speaker identification rate is 89% when the utterance time length of the registered speech is 15 seconds, and the utterance time length of the registered speech is 20 seconds. In this case, the average speaker identification rate is 91%.

  These average speaker identification rates can be improved by increasing the utterance time length of the input speech. If the utterance time length of the input speech is 0.7 seconds or more, the utterance time length of the registered speech can be any value. Even if it exists, an average speaker identification rate will be 99% or more.

  FIG. 7B shows an experimental result of speaker verification. As shown in FIG. 7B, if the utterance time length of the input speech is 0.1 second, the average speaker verification rate is 93.5% when the utterance time length of the registered speech is 5 seconds, When the speech duration is 10 seconds, the average speaker verification rate is 94%, and when the registered speech duration is 15 seconds, the average speaker verification rate is 95% and the registration speech duration is In the case of 20 seconds, the average speaker verification rate is 95%.

  These average speaker verification rates are improved by increasing the utterance time length of the input speech. If the utterance time length of the input speech is 0.7 seconds or more, the utterance time length of the registered speech is any value. Even if it exists, an average speaker collation rate will be 98% or more.

  As described above, when the utterance time length of the input voice is 0.7 seconds or more, high accuracy recognition is possible in both speaker identification and speaker verification. Further, even if the utterance time length of the input voice is a short utterance of 0.1 to 0.7 seconds, sufficient recognition accuracy can be obtained.

  As described above, in the first embodiment, the speaker recognizing unit 31 stores the feature parameter calculated in units of frames from the registered speech, and determines the minimum distance from the feature parameter of the input speech as the similarity. Therefore, it is not necessary to build a statistical model or the like in advance. As a result, it is possible to perform speaker recognition with a simple calculation.

  In addition, even if the registered voice data includes the voices of a plurality of registered speakers, the distance from the frame closest to the input voice among the plurality of frames cut out from the registered voice Since the distance is adopted, the distance to the single registered speaker is obtained between the frames. Therefore, it is possible to recognize that the speaker of the input voice is one of a plurality of persons included in the registered voice.

  In the first embodiment, since the minimum distance between one frame of the input sound and all the frames of the registered sound is obtained and averaged every time the input sound frame increases, the number of frames of the input sound is small. That is, speaker recognition is possible even when the input speech is short. If the number of frames of input speech increases, speaker recognition can be performed with higher accuracy.

  In the first embodiment, the case where all the analysis frames of registration data are used has been described. However, if the number of analysis frames of registration data is sufficient, the processing speed is increased by using only a part of the analysis frames. be able to. Therefore, in the second embodiment, speaker recognition that performs efficient processing by selecting an analysis frame to be used will be described.

  FIG. 8 is an internal configuration diagram illustrating an internal configuration of the speaker recognition unit 131 according to the second embodiment. The storage unit 45 shown in FIG. 8 stores registration data that may include a plurality of speaker's voice feature parameters as a group registration data group, and stores registration data that includes only a single speaker's voice. Store as personal registration data group.

  The speaker recognition unit 131 further includes a registration processing unit 52 and a cluster setting unit 53, and the operation of the minimum distance search unit 51 is different from the minimum distance search unit 46 shown in the first embodiment. Since other configurations and operations are the same as those in the first embodiment, the same components are denoted by the same reference numerals and description thereof is omitted.

  When the speaker verification unit 47b of the recognition processing unit 47 determines that the speaker of the input data matches the speaker of the registration data, the registration processing unit 52 sets the input data as registration data belonging to the personal registration data group. A processing unit to be registered.

  When the speaker of the input data matches the speaker of the registration data belonging to the group registration data group, the registration processing unit 52 registers the input data as new registration data. The reason for the new registration data is that the registration data belonging to the group registration data group may include the voices of a plurality of speakers and cannot be discriminated although it matches the voices of any of the speakers. . On the other hand, since the input data used for the recognition process can be estimated as a single speaker, the input data is registered as registration data belonging to the personal registration data group.

  When the speaker of the input data matches the speaker of the registration data belonging to the personal registration data group, the registration processing unit 52 adds the input data to the matching registration data and registers it. This is because the registration data belonging to the personal registration data group is composed of the voice of a single speaker. Thus, the number of registration data belonging to the personal registration data group and the number of analysis frames of registration data belonging to the personal registration data group are increased by the recognition process, thereby enabling more accurate recognition.

  The cluster setting unit 53 is a processing unit that performs clustering on registration data. Specifically, for registered data in which a sufficient number of analysis frames are accumulated, the analysis frames are classified into a plurality of clusters based on the similarity of the feature parameters. The number of clusters can be arbitrarily set based on the amount of registered data. For each cluster, a representative value of the characteristic parameter of the analysis frame belonging to the cluster is calculated. An arbitrary value such as an average value can be used as the representative value.

  The cluster setting unit 53 associates each analysis frame of the registered data with the cluster to which the analysis frame belongs, and associates the representative value of each cluster with the registered data.

  The processing by the cluster setting unit 53 can be performed at an arbitrary timing. For example, when the registration data is changed, it is preferable to perform processing on the changed registration data.

  The minimum distance search unit 51 calculates the distance between the input data and each registered data as the similarity. The registration data to be used is all the registration data regardless of whether it belongs to the group registration data group or the personal registration data group.

  When calculating the distance between the registration data and the input data, the minimum distance search unit 51 first calculates the distance between the frame of the input data and the representative value of each cluster. Then, the distance to each analysis frame belonging to the cluster having the smallest distance is calculated as a brute force, and the smallest distance is set as the minimum distance for the registered data.

  In this way, when calculating the distance to the registered data, it is possible to speed up the search for the minimum distance by performing processing while limiting the clusters. The limitation of the cluster can be arbitrarily selected by using only the cluster having the smallest distance or excluding the cluster having the maximum distance.

  FIG. 9 is an explanatory diagram of the minimum distance search using clusters. The order of the characteristic parameters of the analysis frame is actually 32 dimensions, but in FIG. 9, it is assumed to be two dimensions for the sake of simplicity.

  In FIG. 9, the values of the characteristic parameters (X, Y) for the analysis frame of the registered data are plotted on the XY plane. The XY plane is classified into three clusters A1 to A3. Further, the representative values of the analysis frames belonging to the clusters A1 to A3 are obtained.

  When obtaining the distance to the input data, the distance between the characteristic parameter for the analysis frame of the input data and the representative values of the clusters A1 to A3 is obtained. In FIG. 9, the distance from the cluster A3 is minimized. For this reason, the distance to the feature parameter for each analysis frame belonging to the cluster A3 is calculated as a brute force, and the smallest distance is the minimum distance for the registered data.

  As described above, in the second embodiment, it is possible to perform efficient processing by classifying registration data into clusters and limiting the analysis frames used for speaker recognition using the clusters. In addition, since the speaker generates single registration data using the result of the recognition process, the latest voice features of the speaker can be retained and the accuracy of speaker recognition can be improved.

  In the above embodiment, the feature parameter is calculated from the voice data at the time of registration and the feature parameter is stored in the storage unit 45. However, the voice data itself is stored in the storage unit 45, and the feature parameter is appropriately set at the time of recognition. You may comprise so that it may calculate.

  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 has been described. However, the voice of a specific speaker is registered as a blacklist, and the speaker registered in the blacklist It may be configured to reject the operation.

  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. In particular, when speaker recognition is performed on a terminal having limited calculation capability, the present invention that can obtain high recognition accuracy with a small processing load is useful.

  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. Further, if the function units of the speaker recognition units 31 and 131 are realized by software and executed by a computer, a speaker recognition program for operating the computer as a speaker recognition device can be obtained.

  As described above, the speaker recognition device, the speaker recognition method, and the speaker recognition program are suitable for improving the convenience 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, 131 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 Characteristic parameter calculation Unit 44 switching unit 45 storage unit 46, 51 minimum distance search unit 47 recognition processing unit 47a speaker identification unit 47b speaker verification unit 52 registration processing unit 53 cluster setting unit 60 monitoring device

Claims (10)

A speaker recognition device for recognizing a speaker of voice data based on voice data,
Similarity that calculates the degree of similarity between the spectral envelope of the partially registered speech data extracted from the registered speech data including the speech of the person to be registered and the spectral envelope of the partially input speech data extracted from the input speech data to be recognized A calculation means;
Recognizing means for recognizing a speaker of the input voice data based on the similarity calculated by the similarity calculating means;
The registered voice data includes voices of a plurality of registration subjects ,
The partial registration voice data is cut out in a plurality so as to include the voice of each registration target person,
The similarity calculation means calculates a similarity with the spectrum envelope of the partial input speech data for each spectrum envelope of the plurality of partial registration speech data,
The recognition means obtains a recognition result that a speaker of the input voice data matches any of the plurality of registration target persons , or a speaker of the input voice data does not match any of the plurality of registration target persons. A speaker recognition device characterized by outputting. A speaker recognition device for recognizing a speaker of voice data based on voice data,
Similarity that calculates the degree of similarity between the spectral envelope of the partially registered speech data extracted from the registered speech data including the speech of the person to be registered and the spectral envelope of the partially input speech data extracted from the input speech data to be recognized A calculation means;
Recognizing means for recognizing a speaker of the input voice data based on the similarity calculated by the similarity calculating means;
The registered voice data includes voices of a plurality of registration subjects ,
The similarity calculation means includes a feature parameter indicating a spectral envelope of each partial registered voice data and a spectral envelope of the partial input voice data for a plurality of partial registered voice data cut out from the same registered voice data. A small distance from the parameter is calculated as a height of similarity of the partial input voice data with respect to each partial registration voice data, and a minimum value of the distance of the partial input voice data with respect to each partial registration voice data is calculated. The distance between the partial input voice data and the registered voice data
The recognizing means uses a distance of the partial input voice data to the registered voice data as the similarity, and a speaker of the input voice data matches any of the plurality of registration subjects , or the input voice data A speaker recognition apparatus, wherein a recognition result indicating that the speaker does not match any of the plurality of registration subjects is output. The similarity calculation means calculates a similarity to the input voice data for each of the plurality of registered voice data,
The recognition means presumes that the speaker of the input voice data matches one of a plurality of registration target persons included in the registered voice data having the highest similarity to the input voice data among the plurality of registered voice data The speaker recognition device according to claim 1, wherein the speaker recognition device is a speaker recognition device. When the similarity of the registered voice data with respect to the input voice data is greater than or equal to a similarity threshold, the recognizing unit determines whether the speaker of the input voice data is one of a plurality of registration subjects included in the registered voice data. The speaker recognition device according to claim 1, wherein the speaker recognition devices are determined to match. The spectral envelopes of a plurality of partially registered voice data obtained from the registered voice data including the voices of the plurality of registration subjects are classified based on similarity of feature parameters indicating the spectrum envelope, and the feature parameters for each classification A classification means for calculating a representative value of
The similarity calculation means calculates a distance between a characteristic parameter indicating a spectrum envelope of the partial input speech data and a representative value of each classification, and registers each partial belonging to the classification with a minimum distance to the representative value The speaker recognition apparatus according to claim 1, wherein voice data is used for calculating the similarity. Monitoring means for performing a monitoring operation on the monitoring target;
Word discrimination means for discriminating words included in the input voice data;
And a control unit configured to control a monitoring operation of the monitoring unit based on a word determined by the word determination unit when a recognition result by the recognition unit satisfies a predetermined condition. Item 6. The speaker recognition device according to any one of Items 1 to 5. A speaker recognition method for recognizing a speaker of voice data based on voice data,
A registration voice data processing step for cutting out the partial registration voice data so as to include the voice of each registration target person from the registration voice data including the voices of a plurality of registration target persons , and obtaining a spectrum envelope of the cut out partial registration voice data;
An input voice data processing step of cutting out the partial input voice data from the input voice data to be recognized and obtaining a spectrum envelope of the cut out partial input voice data;
A degree of similarity calculation step for calculating a degree of similarity with the spectrum envelope of the partial input voice data for each of the spectrum envelopes of the plurality of partial registration voice data,
Based on the similarity calculated in the similarity calculation step, a speaker of the input voice data matches any of the plurality of registration target persons , or a speaker of the input voice data is the plurality of registration target persons. A speaker recognition method, comprising: a recognition step of outputting a recognition result that does not match any of the above. A speaker recognition method for recognizing a speaker of voice data based on voice data,
A registration voice data processing step for cutting out the partial registration voice data from the registration voice data including the voices of a plurality of registration subjects and obtaining a spectrum envelope of the cut out partial registration voice data;
An input voice data processing step of cutting out the partial input voice data from the input voice data to be recognized and obtaining a spectrum envelope of the cut out partial input voice data;
For a plurality of partially registered speech data cut out from the same registered speech data, the distance between the feature parameter indicating the spectrum envelope of each partially registered speech data and the feature parameter indicating the spectrum envelope of the partially input speech data is small Is calculated as the degree of similarity of the partial input voice data with respect to each partial registered voice data, and the minimum distance of the partial input voice data with respect to each partial registered voice data is calculated as the partial input with respect to the registered voice data A similarity calculation step for outputting the distance as the distance of the audio data and outputting the distance as a similarity;
Based on the similarity calculated in the similarity calculation step, a speaker of the input voice data matches any of the plurality of registration target persons , or a speaker of the input voice data is the plurality of registration target persons. A speaker recognition method, comprising: a recognition step of outputting a recognition result that does not match any of the above. A speaker recognition program for recognizing a speaker of voice data based on voice data,
A registration voice data processing procedure for cutting out partial registration voice data so as to include each registration target person's voice from registration voice data containing a plurality of registration target person 's voices, and obtaining a spectrum envelope of the cut out partial registration voice data;
Input audio data processing procedure for extracting partial input audio data from input audio data to be recognized and obtaining a spectral envelope of the extracted partial input audio data;
For each spectrum envelope of a plurality of partial registration voice data, a similarity calculation procedure for calculating a similarity with the spectrum envelope of the partial input voice data,
Based on the similarity calculated by the similarity calculation procedure, a speaker of the input speech data matches any of the plurality of registration subjects , or a speaker of the input speech data is the plurality of registration subjects. speaker recognition program for sure識手order you outputs the recognition result of the does not match one of the characterized by causing a computer to execute the. A speaker recognition program for recognizing a speaker of voice data based on voice data,
A registration voice data processing procedure for extracting partial registration voice data from registration voice data including voices of a plurality of registration target persons and obtaining a spectrum envelope of the cut out partial registration voice data;
Input audio data processing procedure for extracting partial input audio data from input audio data to be recognized and obtaining a spectral envelope of the extracted partial input audio data;
For a plurality of partially registered speech data cut out from the same registered speech data, the distance between the feature parameter indicating the spectrum envelope of each partially registered speech data and the feature parameter indicating the spectrum envelope of the partially input speech data is small Is calculated as the degree of similarity of the partial input voice data with respect to each partial registration voice data, the minimum value of the distance to each partial registration voice data as the distance of the partial input voice data with respect to the registration voice data, A similarity calculation procedure for outputting the distance as a similarity;
Based on the similarity calculated by the similarity calculation procedure, a speaker of the input speech data matches any of the plurality of registration subjects , or a speaker of the input speech data is the plurality of registration subjects. A speaker recognition program for causing a computer to execute a recognition procedure for outputting a recognition result that does not match any of the above.

Images