JP2014026222A - Data generation device and data generation method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To diversify emotional expression without increasing throughput required to extract data including the content of a specified emotion when emotional expression is performed by a synthesized voice in speech synthesis.SOLUTION: The data generation method comprises: executing clustering to all speech parameters PV (S130); estimating representative tag data TD on the basis of tag data TG included in classification clusters CL(S150); deriving a difference between an average parameter CPV_Ain the classification clusters CLand a reference parameter NPV for each classification cluster CLas a representative parameter DPVin the classification clusters CL(S190); and generating speech emotion data EThaving representative tag data TDand a representative parameter DPVassociated with each classification cluster CLto be stored (S200).

Description

  The present invention relates to a data generation apparatus and a data generation method for generating voice emotion data.

  2. Description of the Related Art Conventionally, a speech synthesizer that arbitrarily converts the voice quality of synthesized sound output by speech synthesis is known (see, for example, Patent Document 1). Among these types of speech synthesizers, as a speech synthesis method, there is one that uses formant synthesis that generates speech waveforms by adjusting speech parameters prepared in advance, and thus synthesized speech.

  When an emotion is given to a synthesized sound generated by such formant synthesis, the adjustment of the voice parameter needs to be performed based on at least one voice data. The voice data here refers to the content of the emotion when a person utters the voice and the voice parameters generated based on the voice waveform uttered by the emotion of the content for each person who uttered the voice. Is one piece of data previously associated with each other. This audio data is generally stored in a storage device, and a database is constructed.

  In such a speech synthesizer, it is necessary to increase the types of emotions constituting the speech data, and hence the number of speech data, in order to increase the types of emotions to be given to the synthesized speech.

JP 2004-38071 A

  As described above, if the number of voice data stored in the storage device is increased, in the conventional voice synthesizer, the types of emotions that can be added to the synthesized sound can be increased, that is, the emotional expression by the synthesized sound can be diversified. .

  However, conventional voice data is a one-to-one correspondence between emotion content and voice parameters for each person who utters voice. Therefore, in the conventional technique, in order to diversify the types of emotions added to the synthesized sound, it is necessary to prepare voice data separately for each person who utters the voice and store it in the storage device. The number of could be enormous.

  For this reason, when the number of speech data stored in the storage device becomes enormous, the speech synthesizer requires processing to extract speech data including emotional content designated by the user during speech synthesis. There is a problem that the amount increases and the time required to extract the target audio data becomes longer.

  Therefore, an object of the present invention is to suppress an increase in the amount of processing required to extract data including the contents of a specified emotion while diversifying emotion expressions by synthesized sounds in speech synthesis.

The present invention made to achieve the above object relates to a data generation apparatus.
The data generation apparatus of the present invention includes parameter acquisition means, classification means, tag acquisition means, representative estimation means, parameter determination means, and data generation means.

  The parameter acquisition means acquires a voice parameter included in each of the voice data from a first storage device that stores at least two voice data. The voice parameter referred to here is at least one feature amount representing a waveform of a sound emitted by a person. Furthermore, the voice data referred to here is data in which voice parameters are associated with tag data, which is information including the emotions of the person who produced the sound represented by the voice parameters, for each person.

  The classifying unit classifies the group of voice parameters acquired by the parameter acquiring unit into at least two groups based on the voice parameter distribution. Each of the groups classified by the classification means is set as a classification cluster, and the tag acquisition means sends each of the tag data associated with the voice parameter included in each classification cluster from the first storage device for each classification cluster. get.

  Based on the acquired tag data, the representative estimation means estimates at least one representative tag data, which is information including emotions representing each classification cluster, for each classification cluster. Further, the parameter determining means determines, for each classification cluster, a representative parameter that is a voice parameter representing the classification cluster based on the voice parameters included in each classification cluster classified by the classification means.

  The data generation means generates voice emotion data in which the determined representative parameter and the representative tag data estimated by the representative estimation means are associated with each corresponding classification cluster, and the second storage device To remember.

  According to the data generation device of the present invention, even if the number of types of emotions that can be imparted to the synthesized sound output by the speech synthesizer is the same as in the past, the voice emotion data stored in the second storage device The amount can be reduced compared to the prior art.

  As a result, if the second storage device that stores the voice emotion data generated by the data generation device of the present invention is used, the content of the emotion designated by the user is included at the time of speech synthesis in the speech synthesizer. The amount of processing required to extract the voice emotion data can be reduced, and consequently, the time length required to extract the voice emotion data can be shortened.

  In other words, according to the present invention, in speech synthesis, it is possible to suppress an increase in the amount of processing required to extract data including the contents of a designated emotion, while diversifying emotion expressions by synthesized sounds.

  In the parameter determining means in the present invention, the averaging means derives an average parameter, which is an average value of the speech parameters included in each of the classification clusters, for each classification cluster, and the difference deriving means calculates the average derived by the averaging means. The difference between the parameter and the reference parameter representing the speech parameter at the specified reference value may be derived for each classification cluster as a representative parameter.

According to such a data generation device, the representative parameter can be a difference between the average parameter and the reference parameter.
When speech synthesis is performed by adjusting speech parameters using such speech emotion data, a synthesized sound to which emotions are added can be generated by speech synthesis even in a situation where only a reference parameter exists as a speech parameter.

  Further, in the difference deriving unit in the present invention, the extracting unit extracts each of the speech parameters associated with the tag data representing that the emotion is a natural body from the first storage device, and the reference deriving unit includes the extracting unit. The average of the speech parameters extracted in step 1 may be derived as a reference parameter.

  According to such a data generation device, tag data associated with an audio parameter used for derivation of a reference parameter can be tag data representing that an emotion is a natural body.

  If this emotion is natural, for example, if it is an emotion when there is no expression like an expression in a news program or an emotion in a normal conversation, the voice parameters associated with the tag data can be easily And thus the reference parameters can be easily derived.

By the way, this invention may be made | formed as a data generation method.
When the present invention is used as a data generation method, in the data generation method of the present invention, a parameter acquisition procedure for acquiring audio parameters included in each audio data from the first storage device, and a group of the acquired audio parameters It is necessary to have a classification procedure for classifying into at least two groups based on the distribution of speech parameters. Furthermore, in the data generation method, based on the tag acquisition procedure for acquiring each tag data associated with the speech parameter included in each classification cluster from the first storage device for each classification cluster, and the acquired tag data Thus, it is necessary to have a representative estimation procedure for estimating at least one representative tag data, which is information including emotion representing each classification cluster, for each classification cluster.

  Further, in the data generation method, a parameter determination procedure for determining, for each classification cluster, a representative parameter, which is a voice parameter representing the classification cluster, based on the voice parameter included in each classified classification cluster. Data for generating voice emotion data in which the determined representative parameter and the representative tag data estimated in the representative estimation procedure are associated with each other corresponding to each classification cluster, and stored in the second storage device Production procedure.

  According to such a data generation method, the same effect as that of the data generation apparatus according to claim 1 can be obtained.

It is a block diagram which shows schematic structure of a speech synthesis system. It is a flowchart which shows the process sequence of an audio | voice emotion data generation process. It is explanatory drawing which shows the process outline | summary of an audio | voice emotion data generation process. It is explanatory drawing which shows the process outline | summary of an audio | voice emotion data generation process. It is a flowchart which shows the process sequence of a speech synthesis process.

Embodiments of the present invention will be described below with reference to the drawings.
<Speech synthesis system>
As shown in FIG. 1, the speech synthesis system 1 is configured to synthesize speech based on speech parameters PV registered in advance so that speech having the content specified by the user of the speech synthesis system 1 is output (that is, , Synthesized sound). In the speech synthesis by the speech synthesis system 1, a sound property including an emotion designated by the user is added to the synthesized sound based on speech emotion data ET described later in detail.

  To realize this, the speech synthesis system 1 includes at least one speech input device 10, at least one speech storage server 25, at least one information processing device 30, at least one data storage server 50, One audio output terminal 60 is provided.

  The voice input device 10 is a device to which voice is input. The voice storage server 25 stores voice data SD in which the voice parameter PV generated based on the voice input by the voice input device 10 is associated with the tag data TG representing the nature of the voice.

  The information processing apparatus 30 generates at least two or more voice emotion data ET based on the voice data SD group stored in the voice input device 10. The data storage server 50 stores the voice emotion data ET generated by the information processing apparatus 30.

The voice output terminal 60 outputs a synthesized sound obtained by voice synthesis based on the voice parameter PV stored in the voice storage server 25 and the voice emotion data ET stored in the data storage server 50.
<Voice input device>
The voice input device 10 includes a communication unit 11, an input receiving unit 12, a display unit 13, a voice input unit 14, a voice output unit 15, a storage unit 17, and a control unit 20. The voice input device 10 may be configured as, for example, a well-known karaoke device, or may be configured as another device.

In the communication unit 11, the voice input device 10 communicates with the outside through a communication network. The communication network mentioned here includes, for example, a public wireless communication network and a network line.
The input receiving unit 12 is an input device that receives input of information and commands in accordance with external operations. The input device includes, for example, a key, a switch, and a remote control receiving unit.

  The display unit 13 is a display device that displays at least an image including information indicated by a character code. Examples of the display device include a liquid crystal display and a CRT. The voice input unit 14 is a device that converts sound into an electric signal and inputs the electric signal to the control unit 20, that is, a so-called microphone.

  The audio output unit 15 is a device that converts an electrical signal from the control unit 20 into sound and outputs the sound. The audio output unit 15 may be configured as a sound source module that outputs output sound that simulates sound from a sound source based on data defined by the MIDI (Musical Instrument Digital Interface) standard. This tone generator module includes, for example, a MIDI tone generator.

The storage unit 17 is a non-volatile storage device configured to be able to read and write stored contents.
The control unit 20 is stored in the ROM 21 that stores processing programs and data that need to retain stored contents even when the power is turned off, the RAM 22 that temporarily stores processing programs and data, and the ROM 21 and RAM 22. It is mainly configured by a known computer having at least a CPU 23 that executes each process (various operations) according to the processing program.
<Audio storage server>
The voice storage server 25 is a device that is mainly configured of a non-volatile storage device configured to be able to read and write stored contents. The voice storage server 25 is connected to the voice input device 10, the information processing device 30, and the data storage server 50 through a communication network.

The voice storage server 25 stores at least two pieces of voice data SD. The voice data SD is data in which voice parameters PV _i and tag data TG _i are associated with each speaker. That is, data in which an identification number (ID) for identifying a speaker is added to the audio parameter PV _i and the tag data TG _i is generated as the audio data SD.

  The voice parameter PV is prepared for each waveform of sound produced by a person, and is at least one feature amount representing the voice waveform i. This feature amount is a feature amount of speech used for so-called formant synthesis, and is prepared for each speaker. As a feature amount in the speech parameter PV, at least a fundamental frequency F0, a mel frequency cepstrum (MFCC), a phoneme length, a power, and a time difference thereof in each phoneme in the uttered speech are provided. The feature amount in the voice parameter PV is prepared for each phoneme.

  Since the fundamental frequency F0, MFCC, and the power deriving method are well known, detailed description thereof is omitted here. For example, in the case of the fundamental frequency F0, the time axis of the speech unit for each phoneme is aligned. What is necessary is just to derive | lead-out using methods, such as autocorrelation, the autocorrelation of the frequency spectrum of the speech segment for every phoneme, or a cepstrum method. In the case of MFCC, the time analysis window is applied to the speech element for each phoneme, and the result of frequency analysis (for example, FFT) for each time analysis window is logarithmized for each frequency. The result may be derived by further frequency analysis. The power may be derived by integrating the result of squaring the amplitude by applying a time analysis window to the speech element for each phoneme in the time direction.

  The tag data TG is data representing the nature of the sound represented by the speech parameter PV, and includes speaker feature information representing the features of the speaker, emotion information representing the speaker's emotion when the speech is spoken, and At least. The speaker characteristic information includes, for example, the gender and age of the speaker. In addition to information representing the emotion itself, the emotion information may include information representing a scene, emotion, atmosphere, and the like when speaking, and information necessary for estimating the emotion of the speaker.

  The generation of the voice data SD may be performed by manually analyzing a voice input via the voice input device 10 or an information processing device such as the voice input device 10 executes a program. You can go there.

  When generating the voice data SD by executing a program in the voice input device 10, for example, if the voice input device 10 is a well-known karaoke device, a karaoke that is prepared in advance for karaoke and represents a musical score of music. Using data (that is, MIDI data), the following may be executed. By collating karaoke data against a speech waveform, speech segments for each syllable or phoneme are extracted, and speech parameters PV are derived from each speech segment.

Further, when the voice input device 10 is assumed to be a well-known karaoke device, as an example of a method for associating the voice parameter PV _i and the tag data TG _i for each speaker, it is used when logging into the voice input device 10. The ID input from the user and associated with the song when the song is reserved may be associated with the voice parameter PV _i and the tag data TG _i as the speaker identification number.
<Information processing device>
The information processing apparatus 30 includes a communication unit 31, an input reception unit 32, a display unit 33, a storage unit 34, and a control unit 40.

  The communication unit 31 communicates with the outside via a communication network. The input receiving unit 32 is an input device that receives input of information and commands in accordance with external operations. The input device includes, for example, a keyboard and a pointing device.

  The display unit 33 is a display device that displays an image. Examples of the display device include a liquid crystal display and a CRT. The storage unit 34 is a non-volatile storage device configured to be able to read and write stored contents. The storage device includes, for example, a hard disk device and a flash memory.

The control unit 40 is configured around a known computer having at least a ROM 41, a RAM 42, and a CPU 43.
In the ROM 41 of the information processing apparatus 30, the control unit 40 performs voice emotion data generation processing for generating at least two or more voice emotion data ET based on the voice data SD group stored in the voice storage server 25. A processing program for execution is stored.

That is, the information processing apparatus 30 functions as the data generation apparatus of the present invention by executing the voice emotion data generation process.
The data storage server 50 is a device that is mainly configured by a non-volatile storage device that can read and write stored contents, and is connected to at least the information processing device 30 via a communication network.
<Voice emotion data generation processing>
As shown in FIG. 2, when the voice emotion data generation process is started, all voice parameters PV stored in the voice storage server 25 are acquired (S110).

  Subsequently, the cluster number k, which is a numerical value of at least 2 or more, is acquired (S120). Clustering for classifying all voice parameters PV acquired in S110 into clusters of cluster number k is executed according to the acquired cluster number k (S130). The clustering executed in S130 may be executed by a known method such as k-means. Hereinafter, each cluster classified by clustering is referred to as a classification cluster CL.

That is, as shown in FIG. 3, the classification cluster CL is a group (set) of speech parameters PV that can be regarded as similar in the spatial plane.
In S130, for each classification cluster CL, each speech parameter PV included in the classification cluster CL is stored with an identification code N _{k, j} . However, code j is the index of the speech parameter PV included in the classification cluster CL _k.

  In S120, a numerical value input by the user may be acquired as the cluster number k, or a result estimated from the spatial distribution of each voice parameter PV in the voice parameter PV group may be acquired as the cluster number k.

In the latter case, as an example of a method for estimating the number of clusters k, the distribution of the distance ED _i from the origin in the space plane to each sound parameter PV _i , and the reference vector RV in the space plane and each sound parameter PV _i are formed. It is possible to estimate based on the distribution of the angle ANG _i . More specifically, the cluster number k may be estimated as a value obtained by multiplying the number of peaks in the distribution of the distance ED _i by the number of peaks in the distribution of the angle ANG _i .

The distance ED _i may be obtained by the following equation (1), and the angle ANG _i may be obtained by the following equation (2).

However, t () in the equation (1) means transposition of a vector, that is, a transpose matrix. The reference vector RV is a reference vector having an arbitrary direction.
Furthermore, the peak mentioned here is a maximum value in each distribution. The number of peaks may be obtained by counting the number of zero crossings according to the result of differentiating the curve represented by the distribution of distance ED _i or angle ANG _i .

Subsequently, the voice emotion data generation processing, classification for each cluster CL _k, speech parameters PV included in each classification cluster CL _k (N _{k, j)} collecting tag data TG associated respectively (S140).

The collected at S140, based on the tag data TG group of each classification cluster CL _k, estimating the tag data TG representing each classified cluster CL _k as a representative tag data TD (S150). Estimation of the representative tag data TD is a histogram of the tag data TG constituting the tag data TG group by category cluster CL _k, in the histogram for each classification cluster CL _k, the tag data TG frequency is maximum, it may be the representative tag data TD in the classification cluster CL _k.

Furthermore, the classification cluster for each CL _k, derives the cluster representative value is a representative value of the speech parameter PV in the classification cluster CL _k (S160). The cluster representative value, an average parameter CPV_A _k is the mean value of speech parameters PV included in each classification cluster CL _k, and a dispersion parameter CPV_V _k is the variance of the speech parameter PV included in each classification cluster CL _k included.

Note that the derivation of the mean parameters CPV_A _k in each category cluster CL _k is performed according to the following equation (3). Also, derivation of dispersion parameter CPV_V _k in each category cluster CL _k is performed according to the following equation (4).

  Subsequently, all voice parameters PV satisfying the specific condition are acquired from the voice storage server 25 as neutral parameters PV (S170). According to the following equation (5), the average value of the acquired neutral parameters PV is derived as the reference parameter NPV (S180).

  Specifically, the specific condition in S170 represents that the emotion in the tag data TG is a natural body. As a result, the reference parameter NPV becomes a voice parameter indicating that the emotion is a natural body.

  The specific condition is not limited to representing that the emotion in the tag data TG is a natural body, but may be that the tag data TG is not associated. Furthermore, the specific condition may be all voice parameters PV stored in the voice storage server 25.

Subsequently, according to the following equation (6), for each classification cluster CL _k, the difference between the average parameter CPV_A _k and the reference parameter NPV in the classification cluster CL _k, derives as a representative parameter DPV _k in the classification cluster CL _k ( S190).

As shown in FIG. 4, the difference between the average parameter CPV_A _k and the reference parameter NPV in the classification cluster CL _k is from a coordinate represented by the reference parameter NPV to a coordinate represented by each average parameter CPV_A _k in the spatial plane. It becomes the distance.

Furthermore, a representative tag data TD _k, the representative parameter DPV _k, and a dispersion parameter CPV_V _k, voice emotion data ET _k that associates each classification cluster CL _k corresponding respectively to generate for each classification cluster CL _k And stored in the data storage server 50 (S200).

Thereafter, the voice emotion data generation process ends.
<Audio output terminal>
As shown in FIG. 1, the audio output terminal 60 includes an information receiving unit 61, a display unit 62, a sound output unit 63, a communication unit 64, a storage unit 65, and a control unit 67. The audio output terminal 60 may be configured as, for example, a known portable terminal or a known information processing apparatus. The mobile terminals referred to here include mobile phones and mobile information terminals. The information processing apparatus includes a so-called personal computer.

  Among these, the information reception part 61 receives the information input via the input device (not shown). The display unit 62 displays an image based on a command from the control unit 67. The sound output unit 63 is a known device that outputs sound, and includes, for example, a PCM sound source and a speaker.

  The communication unit 64 is for the voice output terminal 60 to perform information communication with the outside via a communication network. The storage unit 65 is a non-volatile storage device configured to be able to read and write stored contents, and stores various processing programs and various data.

The control unit 67 is mainly configured by a known computer having at least a ROM, a RAM, and a CPU.
<Speech synthesis processing>
Next, speech synthesis processing executed by the control unit 67 of the speech output terminal 60 will be described.

This voice synthesis process is started when a start command is input via the information receiving unit 61 of the voice output terminal 60.
As shown in FIG. 5, when the speech synthesis process is started, first, information input via the information receiving unit 61 (hereinafter referred to as input information) is acquired (S510). The input information acquired in S510 includes, for example, output text indicating the content (word) of the sound output as synthesized sound, and output property information indicating the nature of the sound output as synthesized sound. Note that the output property information is information corresponding to the tag data TG, and includes speaker feature information and emotion information.

  Subsequently, the representative tag data TD corresponding to each phoneme necessary for outputting the output word acquired in S510 as a synthesized sound and most similar to the speaker characteristic information in the output property information acquired in S510. Is extracted from the voice storage server 25 (S520).

  Further, the voice emotion data ET including the representative tag data TD most similar to the emotion information in the output property information acquired in S510 is extracted from the data storage server 50 (S530).

  Then, the voice parameter PV extracted in S520 is adjusted based on the voice emotion data ET extracted in S530 so that a synthesized sound in accordance with the input information acquired in S510 is output (S540). Subsequently, speech synthesis is performed based on the speech parameter PV adjusted in S540 (S550). The voice synthesis in S550 uses a well-known voice synthesis technique using formant synthesis.

Furthermore, the synthesized sound generated by the voice synthesis in S550 is output from the sound output unit 63 (S560).
Thereafter, the speech synthesis process ends.
[Effect of the embodiment]
As described above, according to the information processing apparatus 30, the facial expression space region (cluster CL) is estimated by statistically processing a certain number of audio parameters PV. For this reason, according to the speech synthesis system 1, it is not necessary to prepare all emotional speech parameters PV for each speaker.

  Therefore, according to the information processing apparatus 30, even if the number of types of emotions that can be imparted to the synthesized sound output from the voice output terminal is the same as the conventional type, the data of the voice emotion data ET stored in the data storage server 50 The amount can be reduced compared to the prior art.

  As a result, if the data storage server 50 generated by the information processing apparatus 30 and storing the voice emotion data ET is used, the voice emotion data ET including the emotion content designated by the user is obtained at the time of voice synthesis. The amount of processing required until extraction can be reduced, and consequently the time length required to extract the voice emotion data ET can be reduced.

In addition, the speech synthesis system 1 can reflect natural changes in human speech for each facial expression.
In other words, according to the information processing apparatus 30 of the above-described embodiment, the amount of processing required to extract data including the content of the specified emotion while diversifying the emotion expression by the synthesized sound at the time of speech synthesis. Can be suppressed.

  Moreover, in the voice emotion data generation process, the difference between the average parameter CPV_A and the reference parameter NPV is generated as the representative parameter DPV, and the data associated with the representative parameter DPV and the representative tag data TD is used as the voice emotion data ET. .

  If speech synthesis is performed by adjusting the speech parameter PV using such speech emotion data ET, even if only the reference parameter NPV exists as the speech parameter PV, the synthesized speech to which emotion is added by speech synthesis. Can be generated.

Further, in the voice emotion data generation process, the variance parameter CPV_V is included in the voice emotion data ET.
For this reason, according to the speech synthesis process of the above embodiment, the adjustment amount for adjusting the speech parameter PV at the time of speech synthesis can be finely adjusted.

  In the voice emotion data generation process, the tag data TG associated with the voice parameter used for deriving the reference parameter is the tag data TG indicating that the emotion is a natural body.

  If this emotion is a natural body, for example, if it is an emotion when there is no expression like an expression in a news program or an emotion in a normal conversation, the voice parameter PV associated with the tag data TG is It is easy to collect and thus the reference parameters can be easily derived.

As described above, according to the speech synthesis system 1, by adjusting the speech parameter PV according to the speech emotion data ET and performing speech synthesis, the speech emotion data can be used regardless of whether or not the speaker is the original speaker. It is possible to generate a synthesized sound to which emotions in ET are added.
[Other Embodiments]
As mentioned above, although embodiment of this invention was described, this invention is not limited to the said embodiment, In the range which does not deviate from the summary of this invention, it is possible to implement in various aspects.

  For example, in the above embodiment, the voice storage server 25 and the data storage server 50 are configured separately, but the configurations of the voice storage server 25 and the data storage server 50 are not limited to this. That is, the voice storage server 25 and the data storage server 50 may be configured as a common server.

  In the above embodiment, the reference vector RV is a reference vector having an arbitrary direction. However, the reference vector RV is not limited to this, and a vector from the origin in the space plane toward the neutral parameter PV is used as the reference vector. It may be a vector RV.

Furthermore, in S130 in the voice emotion data generation process of the above embodiment, clustering may be executed after screening is performed using the tag data TG.
Furthermore, the clustering executed in S130 is not limited to k-means, and other known clustering methods may be used.

The voice emotion data ET in the above embodiment includes the distributed parameter CPV_V, but the voice emotion data ET may not include the distributed parameter CPV_V. In other words, at least the representative tag data TD _k and the representative parameter DPV _k need only be associated with each classification cluster CL _k in the voice emotion data in the present invention.
[Correspondence between Embodiment and Claims]
Finally, the relationship between the description of the above embodiment and the description of the scope of claims will be described.

  S110 in the voice emotion data generation process of the above embodiment corresponds to the parameter acquisition means in the present invention, and S120 and S130 in the voice emotion data generation process correspond to the classification means. S140 in the voice emotion data generation process corresponds to a tag acquisition unit in the present invention, and S150 corresponds to a representative estimation unit. Further, S160 to S190 in the voice emotion data generation process correspond to the parameter determination means in the present invention, and S200 corresponds to the data generation means.

Further, the voice storage server 25 in the above embodiment corresponds to the first storage device of the present invention, and the data storage server 50 corresponds to the second storage device of the present invention.
Note that S160 in the voice emotion data generation process of the above embodiment corresponds to the averaging means in the present invention, and S190 corresponds to the difference deriving means in the present invention. Further, S170 in the voice emotion data generation process corresponds to an extraction unit, and S180 corresponds to a reference derivation unit.

  DESCRIPTION OF SYMBOLS 1 ... Speech synthesis system 10 ... Voice input device 25 ... Voice storage server 30 ... Information processing device 31 ... Communication part 32 ... Input reception part 33 ... Display part 34 ... Memory | storage part 40 ... Control part 41 ... ROM 42 ... RAM 43 ... CPU DESCRIPTION OF SYMBOLS 50 ... Data storage server 60 ... Voice output terminal 61 ... Information reception part 62 ... Display part 63 ... Sound output part 64 ... Communication part 65 ... Memory | storage part 67 ... Control part

Claims (4)

A speech parameter that is at least one feature amount representing a waveform of sound emitted by a person and tag data that is information including the emotion of the person who emitted the sound represented by the speech parameter are associated with each person. Parameter acquisition means for acquiring the audio parameters included in each of the audio data from a first storage device storing at least two audio data;
Classification means for classifying the group of voice parameters acquired by the parameter acquisition means into at least two groups based on the distribution of the voice parameters;
Each of the groups classified by the classification means is defined as a classification cluster, and each of the tag data associated with the speech parameter included in each of the classification clusters is acquired for each classification cluster from the first storage device. Tag acquisition means to perform,
Representative estimation means for estimating at least one representative tag data for each classification cluster, based on the tag data acquired by the tag acquisition means, information representing emotions representing each of the classification clusters;
Parameter determining means for determining, for each classification cluster, a representative parameter that is the voice parameter representing the classification cluster based on the voice parameter included in each classification cluster classified by the classification means;
Generating voice emotion data in which the representative parameter determined by the parameter determination unit and the representative tag data estimated by the representative estimation unit are associated with each corresponding classification cluster, and the second memory A data generation device comprising: data generation means stored in the device. The parameter determination means includes
Average means for deriving an average parameter, which is an average value of speech parameters included in each of the classification clusters, for each classification cluster;
A difference deriving unit for deriving for each classification cluster using, as the representative parameter, a difference between an average parameter derived by the averaging unit and a reference parameter representing a speech parameter at a specified reference value; The data generation device according to claim 1. The difference derivation means includes
Extraction means for extracting each of the voice parameters associated with the tag data representing that the emotion is a natural body from the first storage device;
The data generation apparatus according to claim 2, further comprising: a reference derivation unit that derives an average of the voice parameters extracted by the extraction unit as the reference parameter. A speech parameter that is at least one feature amount representing a waveform of sound emitted by a person and tag data that is information including the emotion of the person who emitted the sound represented by the speech parameter are associated with each person. A parameter acquisition procedure for acquiring the audio parameters included in each of the audio data from a first storage device storing at least two audio data;
A classification procedure for classifying the group of voice parameters acquired in the parameter acquisition procedure into at least two groups based on the distribution of the voice parameters;
Each of the groups classified in the classification procedure is defined as a classification cluster, and each of the tag data associated with the speech parameter included in each of the classification clusters is acquired for each classification cluster from the first storage device. Tag acquisition procedure to
Based on the tag data acquired in the tag acquisition procedure, a representative estimation procedure for estimating at least one representative tag data for each classification cluster, which is information including emotions representing each of the classification clusters;
A parameter determination procedure for determining, for each classification cluster, a representative parameter that is the voice parameter representing the classification cluster based on the voice parameter included in each classification cluster classified in the classification procedure;
Generating voice emotion data in which the representative parameter determined in the parameter determination procedure and the representative tag data estimated in the representative estimation procedure are associated with each corresponding classification cluster; A data generation method comprising: a data generation procedure stored in an apparatus.

Images