JP6507867B2 - Voice generation device, voice generation method, and program - Google Patents

Description

  The present invention relates to an audio generation device, an audio generation method, and a program.

  Some voice generation devices change voice quality or the like in real time according to a change in a predetermined input value. Examples of the predetermined input value include the position of a slider used to adjust voice quality and the like, and the noise level around the device.

  This type of voice generation device holds a plurality of voice data having different combinations of voice quality etc. in order to generate voice data of one output target (message), and a plurality of voice data are generated according to the input value. Select one to play. Then, when the input value changes during reproduction of the audio data, the audio data is switched to the audio data according to the changed input value (see, for example, Patent Document 1). Such a voice generation device has a lighter processing load at the time of voice generation as compared to a method (for example, refer to Patent Document 2) of selecting the voice quality and the like of the voice to be generated according to the spectrum.

  A plurality of audio data held by the above-described audio generation device is created by morphing. Morphing is a method of generating a synthetic voice having an intermediate voice quality by mixing two voice data having different voice quality etc. at a desired ratio (morphing rate). When morphing two voice data, it is possible to not only morph the entire voice data at a constant morphing rate, but also to designate the morphing rate in units of phonemes and syllables (for example, Patent Document 3) reference).

  The above-described voice generation device can be applied to voice guidance in an environment where the noise level is not constant, such as notifying a worker of an instruction on facility operation in a factory and the operation status of the facility.

JP, 2006-178052, A Unexamined-Japanese-Patent No. 2-210497 Unexamined-Japanese-Patent No. 2006-227589

  When the above-described voice generation device is applied to an environment where the noise level is not constant, the noise level around the facility is used as a predetermined input value. Also, a plurality of voice data for one message are a first voice data created under conditions that are easy to hear when the noise level is low, and a second voice data created under conditions that are easy to hear when the noise level is high. Create by morphing with. Then, when selecting voice data according to the noise level, voice data of the morphing rate corresponding to the noise level around the device is selected based on a conversion table representing the correspondence between the noise level and the morphing rate. This prevents the message from being difficult to hear when the noise level in the vicinity of the facility rises during the reproduction of voice data (message), thereby preventing the message from being missed.

  Further, when the noise level around the facility is low during the reproduction of the voice data (message), the voice data created under the condition easy to hear when the noise level is low is output. Therefore, it is possible to prevent tiredness and the like by listening to a voice that is easy to hear when the noise level is high in an environment where the noise level is low.

  However, the plurality of voice data generated by morphing the first voice data and the second voice data also differ in the fundamental frequency that is correlated with the accent. Therefore, when switching to audio data having a different morphing rate according to a change in noise level during reproduction of one accent phrase in audio data, the fundamental frequency is changed and the accent is changed. If the accent phrase changes in this way, workers will hear the message with the wrong accent. Therefore, there is a possibility that the worker feels uncomfortable and it is difficult to understand the contents of the message.

  In one aspect, it is an object of the present invention to make it easy to hear reproduced voice even if it is switched to voice data having a different morphing rate according to a change in input value, and to prevent accent change.

  An audio generation apparatus according to an aspect of the present invention includes a morphing rate determination unit and an audio reproduction unit. The morphing rate determining unit determines two or more morphing rates including a voice quality morphing rate and a fundamental frequency morphing rate based on input values from the input device. The audio reproduction unit selects and reproduces audio data based on the morphing rate. Here, the morphing rate determining unit includes a first morphing rate determining unit and a second morphing rate determining unit. The first morphing rate determination unit determines the voice quality morphing rate based on the input value at the reproduction position of the voice data. The second morphing rate determination unit determines the morphing rate of the fundamental frequency based on the input value when the beginning of the accent phrase including the reproduction position is reproduced.

  According to the above-described aspect, even when switching to audio data having different morphing rates in accordance with a change in input value, the reproduced audio is easy to hear and the accent does not change.

It is a functional block diagram of the speech generation device concerning a 1st embodiment. It is a functional block diagram of the morphing rate determination unit in the first embodiment. It is a figure which shows the structure of an audio | voice database. It is a figure which shows the correspondence of a reproduction | regeneration position. It is a figure which shows the structural example of one audio | voice data set. It is a flowchart (the 1) which shows the audio | voice production | generation processing which concerns on 1st Embodiment. It is a flowchart (the 2) which shows the audio | voice production | generation process which concerns on 1st Embodiment. It is a figure showing the graph explaining the relationship between the reproduction | regeneration position of audio | voice data, and a morphing rate. It is a figure showing the graph explaining an accent when a noise level changes during reproduction of an accent phrase. It is a hardware block diagram of a computer. It is a figure which shows another application example of the audio | voice production | generation apparatus which concerns on 1st Embodiment. It is a figure which shows the further another application example of the audio | voice production | generation apparatus which concerns on 1st Embodiment. It is a figure which shows the structural example of the e-learning system which concerns on 2nd Embodiment. It is a figure which shows the structural example of the work window displayed on a display apparatus. It is a functional block diagram of the voice generation device concerning a 2nd embodiment. It is a functional block diagram of the synthetic speech production part in a 2nd embodiment. It is a flowchart (the 1) which shows the audio | voice production | generation processing which concerns on 2nd Embodiment. It is a flowchart (the 2) which shows the audio | voice production | generation process which concerns on 2nd Embodiment.

First Embodiment
In the present embodiment, the configuration of the voice generation device, the voice generation method, and the like will be described in the case where the present invention is applied to a voice generation device for notifying a worker of an instruction of operation of facilities in the factory and the operating status of the facilities.

FIG. 1 is a functional block diagram of the voice generation device according to the first embodiment.
As shown in FIG. 1, the voice generation device 1 according to the present embodiment includes an input value processing unit 100, a morphing rate determination unit 101, a conversion table 102, a voice reproduction unit 103, and a voice database 104. Further, the audio reproduction unit 103 includes an audio data selection unit 103a and a reproduction control unit 103b. The voice generation device 1 reproduces voice according to the surrounding noise level by acquiring noise levels at regular intervals and repeating the operation of generating and playing back voice corresponding to this in frame units.

  The input value processing unit 100 calculates the noise level around the facility 3 based on the audio signal (input value) input from the microphone 2. When the input value processing unit 100 receives a control signal instructing reproduction of audio data from the control unit 300 of the facility 3, the input value processing unit 100 starts acquisition of the audio signal. Further, when the input value processing unit 100 receives the signal indicating that the reproduction of the audio data from the reproduction control unit 103 b is finished, the input value processing unit 100 ends the acquisition of the audio signal from the microphone 2 and the calculation of the noise level.

  The morphing rate determination unit 101 determines the morphing rate based on the noise level calculated by the input value processing unit 100, the conversion table 102, and the information indicating the accent phrase boundary from the reproduction control unit 103b. The conversion table 102 is a table showing the correspondence between the noise level and the morphing rate. Further, the information indicating the accent phrase boundary is information indicating whether the frame currently being processed is the accent phrase boundary.

  The voice reproduction unit 103 reads voice data from the voice database 104 based on the information specifying the voice data to be output from the control unit 300 of the facility 3 and the morphing rate determined by the morphing rate determination unit 101, and the speaker Output to 4. The voice database 104 stores voice data morphed at various morphing rates in advance. The voice data selection unit 103a receives the information specifying the voice data and the morphing rate. The voice data selection unit 103a searches the voice database 104 using the information specifying the voice data and the morphing rate as key information, and identifies the corresponding voice data. Further, when the audio data selection unit 103a specifies the audio data, the audio data selection unit 103a notifies the reproduction control unit 103b of ID information of the specified audio data. The reproduction control unit 103 b reads audio data from the audio database 104 based on the notified ID information, determines a frame to be reproduced, and outputs the frame to the speaker 4. Also, the reproduction control unit 103 b transmits information indicating accent phrase boundaries to the morphing rate determination unit 101. Furthermore, when the output (reproduction) of audio data is completed, the reproduction control unit 103b notifies the input value processing unit 100 that the reproduction is ended.

FIG. 2 is a functional block diagram of the morphing rate determination unit in the first embodiment.
As shown in FIG. 2, the morphing rate determination unit 101 in the present embodiment includes an instant morphing rate determination unit 101a and an accent phrase morphing rate determination unit 101b. Further, the morphing rate determination unit 101 includes a voice quality morphing rate determination unit 101c, a fundamental frequency morphing rate determination unit 101d, and a continuous length morphing rate determination unit 101e.

  Based on the noise level calculated by the input value processing unit 100 and the conversion table 102, the instantaneous morphing ratio determining unit 101a determines the morphing ratio according to the current noise level.

  The accent phrase morphing rate determination unit 101 b determines the morphing rate based on the noise level when the first frame of the accent phrase including the frame currently being processed is reproduced and the conversion table 102. Hereinafter, the noise level when the first frame of the accent phrase is reproduced is also referred to as the noise level at the beginning of the accent phrase. The noise level at the beginning of the accent phrase including the frame currently being processed is based on the noise level received from the input value processing unit 100 and the information indicating the accent phrase boundary received from the reproduction control unit 103b. By maintaining the noise level in Then, each time a noise level is received from the input value processing unit 100, the morphing rate is determined based on the noise level at the beginning of the accent phrase including the frame currently being processed and the conversion table 102.

  The voice quality morphing rate determination unit 101c determines the voice quality morphing rate in the voice data to be reproduced. Here, the voice quality of voice data is a parameter represented by mel cepstrum, mel generalized cepstrum, or spectrum. The voice quality morphing rate determination unit 101c in the present embodiment sets the morphing rate determined by the instantaneous morphing rate determination unit 101a as the voice quality morphing rate.

  The fundamental frequency morphing rate determination unit 101d determines the morphing rate of the fundamental frequency in the audio data to be reproduced. Here, the fundamental frequency is also called F0, and is a parameter that represents the height of the voice. The fundamental frequency morphing rate determination unit 101d in the present embodiment sets the morphing rate determined by the accent phrase morphing rate determination unit 101b as the morphing rate of the fundamental frequency.

  The continuous length morphing rate determination unit 101e determines the morphing rate of the continuous length in the audio data to be reproduced. Here, the continuation length is a parameter that represents the length of the phoneme. The continuous length morphing rate determination unit 101e in the present embodiment sets the morphing rate determined by the accent phrase morphing rate determination unit 101b as the morphing rate of the continuous length. In the continuous length morphing rate 101 e, the morphing rate determined by the instantaneous morphing rate determination unit 101 a may be the morphing rate of the continuous length.

  FIG. 3A is a diagram showing the configuration of a speech database. FIG. 3B is a view showing the correspondence between reproduction positions. FIG. 3C is a diagram showing an example of the configuration of one audio data set.

  As shown in FIG. 3A, the voice database 104 of the voice generation device 1 according to the present embodiment is composed of a plurality of voice data sets including a first voice data set 104-1 and a second voice data set 104-2. . Here, one voice data set is a set of voice data in which the combination of the voice quality morphing rate for one message and the morphing rate of the fundamental frequency (and duration) is different. For example, although a plurality of voice data included in the first voice data set 104-1 are voice data of a message "Turn handle to the right" all, voice quality morphing rate and fundamental frequency The combination with the morphing rate is different. Also, each audio data holds information on accent phrase boundaries in advance. For example, in the case of voice data uttering "Turn handle to the right", the position on the voice data that corresponds to "|" of "| handle | turn to the right | Hold the information. However, the text information may not be present, and at least information indicating whether each position on the audio data is an accent phrase boundary is held. Furthermore, as shown in FIG. 3B, each audio data holds a reference time which advances by 1.0 per mora, for example, as shown in FIG. 3B so as to correspond to the reproduction position.

  Further, in the present embodiment, as in the first voice data set 104-1 shown in FIG. 3C, the morphing rate MP of the voice quality and the morphing rate MA of the fundamental frequency in one voice data set are each from 0 to 1. It is changed by 0.1. The morphing rates MP and MA are morphing the first voice data created under conditions that are easy to hear when the noise level is low and the second voice data created under conditions that are easy to hear when the noise level is high. Represents the ratio of the first audio data at the time of

  Further, in FIG. 3C, speech data MD (MP, MA) {MP = 0 to 1, MA = 0 to 1} represent speech data in which the voice quality morphing rate is MP and the fundamental frequency morphing rate is MA, respectively. ing. Further, in FIG. 3C, MDGn {n = 0 to 10} is a group of voice data MD (MP, MA) {MP = 0 to 1} in which the morphing rate MA of the fundamental frequency is the same value and the morphing rate MP of voice quality is different. Represents

  Further, the first voice data and the second voice data used to create a voice data set may be voice data converted from text data by voice synthesis processing, or voice data obtained by recording voice uttered by a person May be.

Next, the sound generation processing in the sound generation device 1 of the present embodiment will be described.
FIG. 4A is a flowchart (part 1) illustrating a voice generation process according to the first embodiment. FIG. 4B is a flowchart (part 2) showing the sound generation process according to the first embodiment.

  The voice generation device 1 according to the present embodiment receives voice data of morphing rate according to the noise level around the facility 3 when receiving a control signal requesting output of voice (message) from the control unit 300 of the facility 3 Is generated and output in frame units. At this time, as shown in FIG. 4A, the voice generation device 1 first initializes the noise level when the head of the accent phrase including the reproduction position is reproduced (step S1). The accent phrase morphing rate determination unit 101b performs step S1.

  Next, the voice generation device 1 obtains a voice signal (input value) from the microphone 2 and calculates the noise level at the present time (step S2). Step S2 is performed by the input value processing unit 100. The input value processing unit 100 calculates the noise level based on, for example, a prepared correspondence table between the input power of the audio signal and the noise level. Further, the input value processing unit 100 passes the calculated noise level to the instant morphing rate determination unit 101a and the accent phrase morphing rate determination unit 101b of the morphing rate determination unit 101.

  Upon receiving the noise level at the present time from the input value processing unit 100, the instantaneous morphing rate determining unit 101a obtains the instantaneous morphing rate based on the noise level at the current time and the conversion table 102 (step S3a). . After that, the instant morphing rate determination unit 101a passes the obtained instant morphing rate to the voice quality morphing rate determination unit 101c. Then, the voice quality morphing rate determination unit 101c sets the received instantaneous morphing rate as the voice quality morphing rate MP (step S3b).

  On the other hand, upon receiving the noise level at the present time from the input value processing unit 100, the accent phrase morphing factor determination unit 101b holds the current noise level as shown in FIG. 4A (step S4a). Subsequently, the accent phrase morphing rate determination unit 101b confirms whether the noise level at the beginning of the accent phrase, that is, the noise level when reproducing the beginning of the accent phrase including the reproduction position has been set (step S4b). . When the noise level is not set (step S4b; No), the accent phrase morphing factor determination unit 101b sets the current noise level as the noise level when the head of the accent phrase including the reproduction position is reproduced (step S4c). ). Thereafter, the accent phrase morphing rate determination unit 101b obtains an accent phrase morphing rate based on the noise level when reproducing the beginning of the accent phrase and the conversion table 102 (step S4 d). If the noise level has already been set (step S4b; Yes), the accent phrase morphing rate determination unit 101b skips the process of step S4c and performs a process of obtaining the accent phrase morphing rate (step S4d). After step S4d, the accent phrase morphing rate determination unit 101b passes the obtained accent phrase morphing rate to the basic frequency morphing rate determination unit 101d and the continuous length morphing rate determination unit 101e. Then, the fundamental frequency morphing rate determination unit 101d sets the received accent phrase morphing rate as the morphing rate MA of the fundamental frequency (step S4e). Similarly, the continuous length morphing rate determination unit 101e sets the received accent phrase morphing rate as the morphing rate of the continuous length (step S4e).

  Thus, when the voice quality, the fundamental frequency, and the morphing rate of the duration are determined, the morphing rate determination unit 101 determines the determined voice quality, the fundamental frequency, and the morphing rate MP, MA of the duration as the voice data selection unit 103a of the voice reproduction unit 103. Pass to As shown in FIG. 4B, the voice data selection unit 103a is a voice based on the information for specifying the voice data to be output from the control unit 300 of the facility 3, the morphing rate MP of voice quality, and the morphing rate MA of the fundamental frequency. Voice data for output is determined from the database 104 (step S5). At this time, the voice data selection unit 103a specifies the voice data set from the voice database 104 based on the information specifying the voice data. Further, the speech data selection unit 103a determines the speech data MD (MP, MA) in the identified speech data set based on the morphing rate MP of the voice quality and the morphing rate MA of the fundamental frequency. Thereafter, the audio data selection unit 103a passes information related to the audio data MD (MP, MA) to the reproduction control unit 103b.

  When the reproduction control unit 103b receives the information on the sound data MD (MP, MA), the reproduction control unit 103b reads the sound data MD (MP, MA) from the sound database 104 and outputs the sound data to the speaker 4 from the reproduction position based on the current reference time Step S6).

  Further, when the audio data is output to the speaker 4, the reproduction control unit 103b checks whether the reproduction position has reached the end position of the audio data (step S7). If the reproduction position has not reached the end position of the audio data (step S7; No), the reproduction control unit 103b then checks whether the reproduction position matches the accent phrase boundary (step S8). When the playback position matches the accent phrase boundary (step S8; Yes), the playback control unit 103b cooperates with the accent phrase morphing rate determination unit 101b to reproduce the noise level at the time of playing the beginning of the accent phrase at this point. The noise level is updated (step S9). Thereafter, the reproduction control unit 103b changes the reproduction position to the beginning of the next frame (step S10), and causes the input value processing unit 100 to perform the process of step S2. Thereafter, the sound generation device 1 repeats steps S2 to S10 until the reproduction position reaches the end position of the sound data.

  When the reproduction position reaches the end position of the audio data (step S7; Yes), the reproduction control unit 103b ends the output processing with the output of the end position. Thus, the voice generation device 1 is in the standby state. When the audio generation device 1 in the standby state receives a new control signal from the control unit 300 of the facility 3, the audio generation device 1 generates and outputs audio data according to the control signal.

  FIG. 5 is a graph showing the relationship between the reproduction position of audio data and the morphing rate. FIG. 5 shows a graph in which the noise level L at each reproduction position of voice data is plotted, a graph in which the voice quality morphing rate MP is plotted, and a graph in which the fundamental frequency morphing rate MA is plotted vertically arranged.

  For example, as shown in FIG. 5, the noise level L when reproducing certain audio data fluctuates in the range of L1 ≦ L ≦ L2. In the example shown in FIG. 5, the noise level L when the n-th accent phrase is reproduced is L = L1 at the reproduction position P1 (accent phrase boundary Bn) at the beginning of the accent phrase. It rises and changes to L = L2.

  In the voice generation process according to the present embodiment, the voice quality morphing rate MP is determined based on the current noise level L. Therefore, when the noise level L at the time of reproducing the reproduction position P4 included in the n-th accent phrase is L = L2, the voice quality morphing ratio MP with respect to the reproduction position P4 has a value MP (L2) corresponding to the noise level L2. Become.

  On the other hand, in the voice generation process according to the present embodiment, the morphing rate MA of the fundamental frequency and the duration is determined based on the noise level when the head of the accent phrase including the playback position is played back. Therefore, even if the noise level L at the time of reproducing the reproduction position P4 is L = L2, the morphing rate MA of the fundamental frequency and the continuation length with respect to the reproduction position P4 is the noise at the time of reproducing the beginning of the n-th accent phrase. The value MA (L1) corresponds to the level L1.

  As described above, in the voice generation process according to the present embodiment, when the noise level L largely changes during reproduction of one accent phrase, only the voice quality morphing rate changes according to the noise level, and the fundamental frequency and the duration length are changed. The morphing rate does not change. That is, while playing back one accent phrase, it is only the voice quality morphing rate that has a correlation with audibleness that changes according to the noise level, and the fundamental frequency morphing rate that has a correlation with the accent does not change. Therefore, it is possible to prevent the accent phrase from changing while the audio data is being reproduced.

  FIG. 6 is a graph showing an accent when the noise level changes during the reproduction of the accent phrase. Note that FIG. 6 shows the noise level and frequency when the accent phrase “handle the wheel” is reproduced, and the accent of the reproduced sound.

  In FIG. 6, a curve F (L1) shows the relationship between the reproduction position and the fundamental frequency in the audio data created under the condition that the noise level L is L = L1. Further, a curve F (L2) indicates the relationship between the reproduction position and the fundamental frequency in the audio data created under the condition that the noise level L is L = L2 so as to be easily audible. A curve Fout indicates the relationship between the reproduction position and the fundamental frequency when the audio data is generated according to the morphing rate of the fundamental frequency determined based on the noise level L.

  In the conventional voice generation process, when the noise level L changes during playback of the accent phrase, the morphing rate of the fundamental frequency also changes. At this time, as shown in the graph at the top of FIG. 6, the sound of the section of the noise level L2 from the playback position 0 (the beginning of the accent phrase) to P1 (between “n” and “d”) is a curve It is reproduced at the fundamental frequency of F (L2). Similarly, the sound in the section of the noise level L1 from the playback position P2 (between “D” and “L”) to P3 (end position of the accent phrase) is played at the fundamental frequency of the curve F (L1) Ru. Also, in the case where the noise level L gradually decreases from L = L2 to L = L1 as in the section from the playback position P1 to P2, the speech with the fundamental frequency morphing rate MA according to the noise level is Play while changing the frequency.

  Therefore, the fundamental frequency in the case where the accent phrase “handle” is reproduced by the conventional voice generation processing is as shown by a curve Fout shown in the graph at the top of FIG. That is, after the reproduction position P1 at which the noise level L decreases, reproduction is performed at a frequency lower than the basic frequency F (L2) of the audio data at the time of start of reproduction. Thus, the reproduced accent phrase “handle” sounds stronger only at the “n” portion as shown in the upper part of FIG. However, when the accent phrase "handle" is pronounced with a standard accent, as shown in the lower graph of FIG. 6, the four sounds of "noodle" are stronger than "ha" and almost the same I hear it. Therefore, when the morphing rate of the fundamental frequency changes in the accent phrase according to the change of the noise level as in the conventional voice generation processing, the accent may change, which may make the worker feel uncomfortable. In addition, in the case of a message including an accent phrase having homonyms having different accents, the content of the message may be difficult to understand.

  On the other hand, in the voice generation process according to the present invention (the present embodiment), as described above, the morphing rate MA of the fundamental frequency during reproduction of one accent phrase has a large change in noise level in the middle, It remains the morphing rate when playing the beginning of the accent phrase. That is, as shown in the middle and lower graphs of FIG. 6, if the noise level L when reproducing the beginning of the accent phrase is L = L2, the fundamental frequency after the reproduction position P1 at which the noise level L changes is also It remains at the time of the noise level L2. Therefore, the curve Fout representing the frequency of the reproduced "handle" accent phrase coincides with the curve F (L2). Therefore, as shown in the lower graph of FIG. 6, the four accents of "Noodle" sound stronger and almost equal in strength to "ha" as shown in the lower graph of FIG. Therefore, even if the noise level changes in the middle, it can be heard with a standard accent, and it is possible to prevent the worker from feeling uncomfortable or becoming difficult to understand the contents.

  As described above, according to the first embodiment, by determining the morphing rate of voice quality based on the noise level at the current time (the current reproduction position), it is possible to prevent the voice from being difficult to hear due to the change in the noise level. . Furthermore, by determining the morphing rate of the fundamental frequency at the current playback position based on the noise level when the beginning of the accent phrase including the current playback position is played back, the accent may change and it may be difficult to understand the content. It can prevent.

  In a facility such as a factory, it is desirable to accurately notify in real time the operating status and operation guidance of a facility so that workers can operate the facility safely and correctly. As described above, the voice generation device 1 of the present embodiment changes the voice quality in real time according to the noise level around the facility 3 while changing the fundamental frequency in accent phrase units. Therefore, even in an environment where the noise level is not constant, it is easy to hear a voice message, and it is not difficult to understand the content due to an accent mistake. Therefore, it can be said that the voice generation device 1 of the present embodiment is optimal for supporting safe and accurate work in facilities such as a factory.

  In the voice generation device 1 according to the first embodiment, the instantaneous morphing rate determining unit 101a and the voice quality morphing rate determining unit 101c illustrated in FIG. 2 may be integrated into one determining unit. Similarly, the accent phrase morphing rate determination unit 101b, the fundamental frequency morphing rate determination unit 101d, and the continuous length morphing rate determination unit 101e illustrated in FIG. 2 may be one integrated determination unit. Also, the continuous length morphing rate determination unit 101e may set the morphing rate determined by the instantaneous morphing rate determination unit 101a as the morphing rate of the continuous length.

  Also, the voice generation device 1 according to the first embodiment can be realized by, for example, a computer and a program that causes the computer to execute the processing illustrated in FIGS. 4A and 4B. The audio generation device 1 realized by the computer and the program will be described with reference to FIG.

FIG. 7 is a hardware configuration diagram of a computer.
As shown in FIG. 7, the computer 5 operated as an audio generation device includes a processor 50, a main storage device 51, an auxiliary storage device 52, an input device 53, an output device 54, and a communication interface device 55. Prepare. These elements 50 to 55 in the computer 5 are connected to one another by a bus 59 so that data can be passed between the elements.

  The processor 50 is an arithmetic processing unit such as a central processing unit (CPU) or a micro processing unit (MPU), and controls the overall operation of the computer 5 by executing various programs including an operating system.

  The main storage device 51 includes a read only memory (ROM) 51 a and a random access memory (RAM) 51 b. For example, a predetermined basic control program or the like read out by the processor 50 when the computer 5 is started is recorded in the ROM 51a in advance. The RAM 51 b is also used as a working storage area as needed when the processor 50 executes various programs. In the present embodiment, for example, the RAM 51b can be used to temporarily hold the noise level when reproducing the beginning of the accent phrase, information indicating an audio data set including audio data to be reproduced, and the like.

  The auxiliary storage device 52 is a storage device having a large capacity as compared with the main storage device 51 such as a Hard Disk Drive (HDD) or a Solid State Drive (SSD). The auxiliary storage device 52 can store various programs executed by the processor 50 and various data including the conversion table 102 and the voice database 104.

  The input device 53 is, for example, various buttons and switches, and the microphone 2. The button or switch is used to set the operation of the computer 5 (voice generation device 1). When the operator of the computer 5 operates various buttons and switches, input information associated with the contents of the operation is transmitted to the processor 50. Also, the microphone 2 is used to obtain the noise level around the facility 3.

  The output device 54 is, for example, a liquid crystal display or a speaker 4. The liquid crystal display displays operation guidance, setting values and the like according to display data transmitted from the processor 50 and the like. Further, the speaker 4 outputs audio data transmitted from the processor 50 or the like.

  The communication interface device 55 is a device for communicably connecting the computer 5 and the control unit 300 of the facility 3. When the computer 5 receives the control signal from the control unit 300 of the facility 3 by the communication interface device 55, the computer 5 generates and outputs a message (voice data) according to the control signal.

  In the computer 5, the processor 50 reads out the program for the above-mentioned sound generation processing from the auxiliary storage device 52 and executes it. When the processor 50 receives a control signal from the control unit 300 of the facility 3 via the communication interface device 55 during execution of a program, the processor 50 obtains the noise level around the facility 3 using the microphone 2. Also, the processor 50 generates an instantaneous morphing rate MP and accent phrase morphing based on the current noise level, the noise level when reproducing the beginning of the accent phrase, and the conversion table 102 stored in the auxiliary storage device 52 or the RAM 51b. Determine the rate MA. Then, the processor 50 sets the morphing rate of voice quality to the instantaneous morphing rate MP, and sets the morphing rate of the fundamental frequency and the duration to the accent phrase morphing rate MA. Further, the processor 50 reads voice data to be reproduced from the voice database 104 of the auxiliary storage device 52 based on the combination of the set voice quality, the fundamental frequency, and the morphing rate of the continuous length, and outputs the voice data to the speaker 4.

[Application Example of Voice Generation Device 1]
As an application example of the voice generation device 1 according to the present embodiment, FIG. 1 shows an example in which the voice generation device 1 is provided separately from the facility 3. However, the sound generation device 1 according to the present embodiment is not limited to this, and may be built in the equipment 3 as a sound generation unit. Furthermore, when the operation statuses of a plurality of facilities are applied to a facility centrally managed by one management server, it is also possible to output voice based on a control signal from the management server, not from the facility 3.

FIG. 8A is a diagram showing another application example of the voice generation device according to the first embodiment.
Facilities, such as a factory to which the voice generation device 1 according to the present embodiment can be applied, include a plurality of facilities 3 (3A, 3B) as shown in FIG. 8A. It is often managed. The management server 6 is communicably connected to each facility 3. For example, the temperature, pressure, the distance from the facility 3 to the worker, the presence or absence of a worker, from various sensors provided to each facility 3 Get information such as Then, the management server 6 monitors the operation status of each facility 3 based on the information acquired from each facility 3 and manages each facility 3 to operate normally. As described above, when the operation statuses of the plurality of facilities 3 are centrally managed by the management server 6, the operation of the plurality of voice generation devices 1 (1 A, 1 B) individually applied to the respective facilities 3 is also controlled by the management server 6 It is possible to manage. When the management server 6 centrally manages the operations of the plurality of voice generation devices 1, for example, voice data conveying that an abnormality has occurred in the facility 3A can be output toward the periphery of the facility 3B. Therefore, even if the voice data conveying the abnormality toward the periphery of the facility 3A is output but the abnormality is not dealt with for a certain period of time, the worker of the other facility 3B around the abnormality of the facility 3A It can be informed. Therefore, it is possible to prevent a failure or the like of the equipment 3A due to a delay in coping with the abnormality of the equipment 3A. In addition, when a plurality of facilities 3A and 3B are linked (cooperation), for example, an abnormality occurring in one facility can be promptly notified to a worker who is around another facility, and a chain of facility abnormalities Can be prevented.

FIG. 8B is a diagram showing still another application example of the voice generation device according to the first embodiment.
When the voice generation device 1 is applied to a facility in which the operation status of a plurality of facilities 3 (3A, 3B) is centrally managed by one management server 6, for example, as shown in FIG. 8B, it corresponds to the voice generation device 1 The voice generation unit 600 may be provided in the management server 6. By doing this, compared to the case where the voice generation device 1 is applied to each of the plurality of facilities 3, the introduction cost and the maintenance cost of the voice generation device can be reduced.

  In addition, it is needless to say that the voice generation device 1 according to the present embodiment can be applied not only to a factory but also to, for example, a device that performs voice guidance in a station yard or a downtown area.

Second Embodiment
In the present embodiment, a configuration of a voice generation apparatus, a voice generation method, and the like when the present invention is applied to an e-learning system will be described.

FIG. 9 is a view showing a configuration example of an e-learning system according to the second embodiment.
As shown in FIG. 9, in the e-learning system according to the present embodiment, a host computer 8 and a plurality of terminals (clients) 9 are connected by a communication network 10 such as the Internet. The host computer 8 is a computer that creates and provides teaching materials. On the other hand, each of the plurality of terminals 9 is a computer used when a learner learns using a teaching material.

  The host computer 8 operates as a voice generation device when creating or reproducing voice data as teaching material. The host computer 8 includes a computer main body 80, a keyboard 81, a mouse 82, a display device 83, and a speaker 84. The computer main body 80 includes a processor 50, a main storage device 51, an auxiliary storage device 52, a communication interface device 55 and the like in the hardware configuration of the computer shown in FIG. The keyboard 81 and the mouse 82 correspond to the input device 53 in the hardware configuration of the computer shown in FIG. The display device 83 and the speaker 84 correspond to the output device 54 in the hardware configuration of the computer shown in FIG.

  When the host computer 8 is operated as the voice generation device 1, the computer main body 80 is made to execute a voice data creation program. The voice data creation program is a program for creating voice data from character information (text data) input by the operator operating the keyboard 81 or the like. During execution of the voice data creation program, for example, a work window 85 as shown in FIGS. 9 and 10 is displayed on the display device 83.

FIG. 10 is a view showing a configuration example of a work window displayed on the display device.
For example, as shown in FIG. 10, an input area 85a, a play button 85b, a save button 85c, a slider 85d, and a groove 85e are provided in the work window 85 displayed on the display device 83 when generating audio data. It is provided.

  The input area 85a is an area for receiving and displaying character information input by operating the keyboard 81 or the like shown in FIG. 9 as character information for voice data creation.

  The reproduction button 85b is used when converting the character information displayed in the input area 85a into audio data and reproducing it. Further, the save button 85c is used when converting the character information displayed in the input area into voice data and storing it, that is, storing it in the storage device as an electronic file.

  The slider 85 d is used to specify the degree of emphasis of speech when converting character information displayed in the input area 85 a into speech data and reproducing the speech data. The slider 85d can be moved left and right along the groove 85e, and in the example shown in FIG. 10, the degree of emphasis is lowest when the slider 85d is moved to the left end (smooth) of the groove 85e. The closer to the right end (emphasis) of 85e, the higher the degree of emphasis. When the slider 85d is moved along the groove 85e, the slider value changes in accordance with the distance from the left end of the groove 85e. When the computer main body 80 creates audio data, based on the slider value, the audio parameter at the time of calm and the audio parameter at the time of enhancement are morphed so that the emphasis degree corresponds to the position of the slider 85 d.

  Next, a configuration example of functional blocks when the computer main body 80 is operated as the voice generation device 1 will be described with reference to FIGS. 11 and 12.

  FIG. 11 is a functional block diagram of the voice generation device according to the second embodiment. FIG. 12 is a functional block diagram of a synthetic speech generation unit in the second embodiment.

  As shown in FIG. 11, the voice generation apparatus 1 (computer main body 80) according to the second embodiment includes an input data processing unit 120, a morphing rate determination unit 121, a conversion table 122, and a synthesized speech creation unit 123. , Speech database 124. Further, the voice generation device 1 further includes a display control unit 125 and a text database 126.

  The input data processing unit 120 performs processing of receiving text data input from the input device (keyboard) 81 and processing of receiving position information of the slider 85 d input from the input device (mouse) 82. The input data processing unit 120 passes the input text data to the display control unit 125 and stores the text data in the text database 126. Further, the input data processing unit 120 passes the input position information (slider value) of the slider 85 d to the display control unit 125. Furthermore, when the input data processing unit 120 receives a signal corresponding to the operation of pressing the play button 85 b or the save button 85 c from the mouse 82 or the like, the input data processing unit 120 passes the slider value to the morphing rate determination unit 121 and synthesizes the text data as synthesized speech. Pass it to the creation unit 123.

  The display control unit 125 controls the display of the display device 83. The display control unit 125 changes the display in the input area 85a in the work window 85 displayed on the display device 83 and the position of the slider 85d based on, for example, the text data and the slider value received from the input data processing unit 120. .

  The morphing rate determination unit 121 determines the morphing rate based on the slider value received from the input data processing unit 120, the conversion table 122, and the information indicating the accent phrase boundary from the synthetic speech creation unit 123. The conversion table 122 is a table showing the correspondence between slider values and morphing rates. Further, the information indicating the accent phrase boundary is information indicating whether the reproduction position of the audio data currently output to the speaker 84 is the accent phrase boundary.

  The synthetic speech creation unit 123 creates synthetic speech based on the text data received from the input data processing unit 120 and the morphing rate determined by the morphing rate determination unit 121. Further, the synthetic speech generation unit 123 outputs the generated synthetic speech to the speaker 84. Furthermore, when the synthesized speech is generated according to the operation of pressing the save button 85 c, the synthesized speech generation unit 123 stores the generated synthesized speech in the speech database 124. When the synthesized speech is stored in the speech database 124, the synthesized speech creation unit 123 stores the data of the synthesized speech in association with the text data stored in the text database 126.

  The morphing rate determination unit 121 in the voice generation device 1 of the present embodiment determines the morphing rate of voice quality, fundamental frequency, and duration in the same manner as in the first embodiment. That is, the morphing rate determination unit 121, like the morphing rate determination unit 101 shown in FIG. 2, is an instantaneous morphing rate determination unit, an accent phrase morphing rate determination unit, a voice quality morphing rate determination unit, and a fundamental frequency morphing rate determination unit And a continuous length morphing rate determining unit.

  On the other hand, the synthetic speech creation unit 123 in the speech generation device 1 of the present embodiment creates speech data of synthetic speech based on the text data and the morphing rate. The synthetic speech creation unit 123 of the present embodiment creates speech data by a synthesis method based on a Hidden Markov Model (HMM), which is one of known speech synthesis methods. As shown in FIG. 12, the synthetic speech creation unit 123 includes a language processing unit 123a, a silent speech parameter creation unit 123b, an enhanced speech parameter creation unit 123c, a morphing processing unit 123d, and an analysis / synthesis unit 123e. Including. Further, the synthetic speech creation unit 123 further includes a plain speech HMM parameter 123 f and an emphasis speech HMM parameter 123 g.

  The language processing unit 123a reads and converts text data into phonetic text representing accents.

  The silent speech parameter creation unit 123b creates a speech parameter for a silent speech based on the phonetic text and the silent speech HMM parameter 123f. Also, the emphasized speech parameter creation unit 123c creates speech parameters for the speech during emphasis based on the phonetic text and the emphasized speech HMM parameters 123g.

  The morphing processing unit 123d morphs the speech parameter of the speech at the time of silence and the speech parameter of the speech at the emphasis at the morphing rate determined by the morphing rate determination unit 121, and creates the speech parameter for the current frame.

  The analysis and synthesis unit 123 e analyzes and synthesizes speech parameters for the current frame and converts the speech parameters into speech waveforms. Further, when the beginning of the accent phrase is included in the current frame, the analysis and synthesis unit 123 e notifies the morphing rate determination unit 121 of information indicating that the beginning of the accent phrase is included.

  Next, the voice generation processing in the voice generation device 1 (host computer 8) of the present embodiment will be described.

  FIG. 13A is a flowchart (part 1) illustrating an audio generation process according to the second embodiment. FIG. 13B is a flowchart (part 2) illustrating the sound generation process according to the second embodiment.

  The voice generation device 1 according to the present embodiment voices the text displayed in the input area 85a when the operation of pressing the play button 85b or the save button 85c of the work window 85 as shown in FIG. 10 is performed. Convert to and play. At this time, as shown in FIG. 13A, the audio generation device 1 first initializes the reproduction position of the text data and the correspondence between the slider value and the reproduction position (step S21). Step S21 is performed by the accent phrase morphing rate determining unit (not shown) of the morphing rate determining unit 121.

  Next, the speech generation device 1 acquires the slider value at the current point and passes it to the morphing rate determination unit 121 and passes the text data to the synthesized speech creation unit 123 (step S22). The input data processing unit 120 performs step S22. The input data processing unit 120 passes the acquired slider value to the instant morphing rate determination unit and the accent phrase morphing rate determination unit (not shown) of the morphing rate determination unit 121. Further, the input data processing unit 120 passes the text data to the language processing unit 123 a of the synthetic speech creation unit 123.

  After step S22, the voice generation device 1 performs a morphing rate determination process (step S23) and a voice parameter creation process. The morphing rate determining unit 121 performs the morphing rate determining process (step S23). The morphing rate determining unit 121 determines the morphing rate of the voice quality, the fundamental frequency, and the duration by the same processes as steps S3a and S3b and steps S4a to S4e illustrated in FIG. 4A. In the process of step S23 performed in the present embodiment, a slider value is used instead of the noise level. In addition, the morphing rate determination unit 121 passes the determined voice quality, the fundamental frequency, and the morphing rate of the duration to the morphing processing unit 123 d of the synthesized speech creation unit 123.

  On the other hand, the synthetic speech creation unit 123 performs the speech parameter creation process. First, in parallel with the morphing rate determination process S23, the synthetic speech creation unit 123 reads the text data and converts it into a phonetic text representing an accent (step S24a), and speech parameters for plain speech and emphasized speech. A process of creating (step S24b) is performed.

  The language processing unit 123a performs step S24a. The language processing unit 123a converts text data into phonetic text by any of known conversion methods.

  Step S24b is performed by the quiet speech parameter creation unit 123b and the emphasis speech parameter creation unit 123c. The silent speech parameter creation unit 123b creates, based on the phonetic text and the silent speech HMM parameter 123f, a speech parameter for a speech at quietness with the lowest degree of emphasis. The emphasized speech parameter creation unit 123c creates speech parameters for the speech during emphasis with the highest degree of emphasis based on the phonetic text and the emphasized speech HMM parameters 123g. The process of step S24b is performed by any of known speech parameter creation methods based on the Hidden Markov Model. The silent speech parameter creation unit 123b passes the created voice parameter to the morphing processing unit 123d. Similarly, the emphasized voice parameter creation unit 123c passes the created voice parameter to the morphing processing unit 123d.

  When receiving the speech parameter and the morphing rate, the morphing processing unit 123d creates a speech parameter for the frame based on the received speech parameter and the morphing rate, as shown in FIG. 13B (step S25). The morphing processing unit 123d creates speech parameters for the frame by any of the morphing processes in the known speech synthesis process. In addition, the morphing processing unit 123d passes the created speech parameter to the analysis and synthesis unit 123e.

  The analysis and synthesis unit 123 e analyzes and synthesizes speech parameters for the frame and converts the speech parameters into speech data (speech waveform) of the frame (step S 26). The analysis and synthesis unit 123e converts the speech parameter of the frame into speech data by any of the conversion methods in the known speech synthesis process.

  Further, the analysis and synthesis unit 123e outputs the obtained voice data (step S27). The analysis and synthesis unit 123e outputs the obtained voice data to the speaker 84. Further, in the case of speech synthesis processing according to the operation of pressing the save button 85 c of the work window 85, the speech data is stored in the speech database 124.

  Furthermore, after outputting the voice data, the analysis and synthesis unit 123 e confirms whether the frame has reached the end position of the text data (step S 28). If the frame has not reached the end position of the text data (step S28; No), the analysis and synthesis unit 123e then checks whether the frame includes an accent phrase boundary (step S29). When the accent phrase boundary is included (step S29; Yes), the analysis synthesis unit 123e cooperates with the morphing rate determination unit 121 to set the slider value of the time when the beginning of the accent phrase including the reproduction position is reproduced at the current time The slider value is updated (step S30). Thereafter, the analysis and synthesis unit 123e changes the frame to the next frame (step S31), and causes the input data processing unit 120 to perform the process of step S22. Thereafter, the voice generation device 1 repeats steps S22 to S31 until the frame reaches the end position of the text data.

  Then, when the frame has reached the end position of the text data (step S28; Yes), the analysis and synthesis unit 123e outputs the voice data of the last frame and ends the processing. Thus, the voice generation device 1 is in the standby state. When the sound generation device 1 in the standby state receives a signal corresponding to an operation of pressing the play button 85 b or the save button 85 c of the work window 85, the sound generation device 1 performs text data generation and output processing again.

  As described above, in the voice generation process of the present embodiment, when text data is converted into voice data (voice waveform), the morphing rate of voice quality that has a correlation with the degree of voice emphasis such as how to stretch voice Make a decision based on Therefore, the desired section of the voice data can be easily emphasized. For example, as shown in FIG. 10, when reproducing by inputting "The pointer is important in C language" in the input area 85a of the work window 85, the voice of the emphasis degree according to the position of the slider 85d from the speaker 84 "The pointer is important in C language" is output. At this time, the slider 85d is moved to the right at the beginning of the accent phrase "pointer is", and the slider 85d is returned to the left at the end position, whereby the "pointer is" can be emphasized.

  Further, in the voice generation process of the present embodiment, the morphing rate of the fundamental frequency having a correlation with the accent is determined based on the slider value when the head of the accent phrase is reproduced. Therefore, changing the slider value while playing the accent phrase does not change the accent. Therefore, it is possible to prevent that the accent of the accent phrase changes and it becomes difficult to understand the contents of the reproduced sound. For example, when the text data "The pointer is important in C language" is reproduced so that the accent phrase "the pointer is" is emphasized, the accent is displayed even if the position of the slider 85b changes while the "pointer is" playing. does not change.

  In the present embodiment, the case of creating speech parameters at the time of calmness and emphasis based on the Hidden Markov Model has been described as an example, but the two speech parameters used for morphing are not limited to this, and other methods may be used. You may create it.

The following appendices will be further disclosed regarding the embodiment including each example described above.
(Supplementary Note 1)
A morphing rate determining unit that determines two or more morphing rates including a voice quality morphing rate and a fundamental frequency morphing rate based on input values from an input device;
An audio reproduction unit that reproduces audio data based on the morphing rate,
The morphing rate determining unit
A first morphing rate determining unit that determines a morphing rate of the voice quality based on the input value at the time of reproducing each frame of the voice data;
And a second morphing rate determining unit that determines the morphing rate of the fundamental frequency based on the input value when the first frame of the accent phrase including the frame in the audio data is reproduced. Voice generation device.
(Supplementary Note 2)
The morphing rate determining unit
The morphing rate of the fundamental frequency determined by the second morphing rate determining unit is determined as a morphing rate of a continuous length.
The voice generation device according to claim 1, characterized in that:
(Supplementary Note 3)
The morphing rate determining unit
The morphing rate of the voice quality determined by the first morphing rate determination unit is determined as the morphing rate of the continuous length,
The voice generation device according to claim 1, characterized in that:
(Supplementary Note 4)
The input value includes the value of the noise level in a predetermined area outside the device,
The voice generation device according to claim 1, characterized in that:
(Supplementary Note 5)
The input value is a value indicating a position within the range of a slider movable within a predetermined range.
The voice generation device according to claim 1, characterized in that:
(Supplementary Note 6)
A storage unit storing a plurality of audio data different in combination of the morphing rates;
The voice reproduction unit
A reproduction control unit for reading out audio data from the storage unit and outputting the audio data from the reproduction position based on a combination of the morphing rates determined by the morphing rate determination unit for each reproduction position of the audio data;
The voice generation device according to claim 1, characterized in that:
(Appendix 7)
The speech generation device includes a synthetic speech creation unit that creates synthetic speech based on text data based on a predetermined language,
The synthetic speech generation unit
A language processing unit for converting text data based on a predetermined language into phonetic text;
A speech parameter creation unit that creates two or more speech parameters for the phonetic text based on two or more conversion parameters different in voice type;
A morphing processing unit that morphs the two or more speech parameters based on the morphing rate to create speech parameters for the synthesis position;
An output unit that converts voice parameters generated by the morphing processing unit into voice data and outputs the converted data;
The speech synthesizer according to claim 1, further comprising:
(Supplementary Note 8)
The computer is
The morphing rate of voice quality is determined based on the current reproduction position in the audio data to be output and the input value corresponding to the reproduction position,
The morphing rate of the fundamental frequency is determined based on the input value when the head of the accent phrase including the reproduction position in the audio data is reproduced.
Generating voice data based on two or more morphing rates including the determined morphing rate,
A voice generation method characterized by performing processing.
(Appendix 9)
Determine the duration morphing rate to be the same morphing rate as the fundamental frequency morphing rate,
Generating the audio data based on the voice quality, the fundamental frequency, and the morphing rate of the duration;
The speech generation method according to appendix 8, characterized in that
(Supplementary Note 10)
Determine the morphing rate of the duration length to be the same as the morphing rate of the voice quality,
Generating the audio data based on the voice quality, the fundamental frequency, and the morphing rate of the duration;
The speech generation method according to appendix 8, characterized in that
(Supplementary Note 11)
For each of the reproduction positions of the voice data, one of a plurality of voice data having different combinations of the voice quality and the morphing rate of the fundamental frequency prepared in advance is selected based on the voice quality and the morphing rate of the fundamental frequency. Generate voice data,
The speech generation method according to appendix 8, characterized in that
(Supplementary Note 12)
The morphing rate of voice quality is determined based on the current playback position in the audio data to be output and the input value corresponding to the playback position,
The morphing rate of the fundamental frequency is determined based on the input value when the head of the accent phrase including the reproduction position in the audio data is reproduced.
Generating voice data based on two or more morphing rates including the determined morphing rate,
A program that causes a computer to execute a process.

DESCRIPTION OF SYMBOLS 1 voice synthesizer 100 input value processing unit 101, 121 morphing rate determination unit 102, 122 conversion table 103 voice reproduction unit 123 synthesized voice creation unit 104, 124 voice database 120 input data processing unit 125 display control unit 126 text database 101 a instant morphing Rate determination unit 101b Accent phrase morphing ratio determination unit 101c Voice quality morphing ratio determination unit 101d Basic frequency morphing ratio determination unit 101e Continuous length morphing ratio determination unit 103a Voice data selection unit 103b Reproduction control unit 123a Language processing unit 123b Plain speech parameter creation unit 123c Emphasized speech parameter creation unit 123d morphing processing unit 123e analysis synthesis unit 123f plain speech HMM parameter 123g emphasis speech HMM parameter 2 microphone 3, 3A, 3 B Equipment 4, 84 Speaker 5 Computer 50 Processor 51 Main storage device 52 Auxiliary storage device 53 Input device 54 Output device 55 Communication interface device 6 Management server 8 Host computer 80 Computer main body 81 Keyboard 82 Mouse 83 Display device 85 Work window 85a Input area 85b play button 85c save button 85d slider 85e groove 9 client 10 communication network

Claims (8)

A morphing rate determining unit that determines two or more morphing rates including a voice quality morphing rate and a fundamental frequency morphing rate based on input values from an input device;
An audio reproduction unit that reproduces audio data based on the morphing rate,
The morphing rate determining unit
A first morphing rate determining unit that determines a morphing rate of the voice quality based on the input value at the time of reproducing each frame of the voice data;
And a second morphing rate determining unit that determines the morphing rate of the fundamental frequency based on the input value when the first frame of the accent phrase including the frame in the audio data is reproduced. Voice generation device. The morphing rate determining unit
The morphing rate of the fundamental frequency determined by the second morphing rate determining unit is determined as a morphing rate of a continuous length.
The voice generation device according to claim 1, characterized in that: The morphing rate determining unit
The morphing rate of the voice quality determined by the first morphing rate determination unit is determined as the morphing rate of the continuous length,
The voice generation device according to claim 1, characterized in that: The input value includes the value of the noise level in a predetermined area outside the device,
The voice generation device according to claim 1, characterized in that: A storage unit storing a plurality of audio data different in combination of the morphing rates;
The voice reproduction unit
A reproduction control unit for reading out audio data from the storage unit and outputting the audio data from the reproduction position based on a combination of the morphing rates determined by the morphing rate determination unit for each reproduction position of the audio data;
The voice generation device according to claim 1, characterized in that: The speech generation device includes a synthetic speech creation unit that creates synthetic speech based on text data based on a predetermined language,
The synthetic speech generation unit
A language processing unit for converting text data based on a predetermined language into phonetic text;
A speech parameter creation unit that creates two or more speech parameters for the phonetic text based on two or more conversion parameters different in voice type;
A morphing processing unit that morphs the two or more speech parameters based on the morphing rate to create speech parameters for the synthesis position;
An output unit that converts voice parameters generated by the morphing processing unit into voice data and outputs the converted data;
The speech synthesis apparatus according to claim 1, further comprising: The computer is
The morphing rate of voice quality is determined based on the current reproduction position in the audio data to be output and the input value corresponding to the reproduction position,
The morphing rate of the fundamental frequency is determined based on the input value when the head of the accent phrase including the reproduction position in the audio data is reproduced.
Generating voice data based on two or more morphing rates including the determined morphing rate,
A voice generation method characterized by performing processing. The morphing rate of voice quality is determined based on the current reproduction position in the audio data to be output and the input value corresponding to the reproduction position,
The morphing rate of the fundamental frequency is determined based on the input value when the head of the accent phrase including the reproduction position in the audio data is reproduced.
Generating voice data based on two or more morphing rates including the determined morphing rate,
A program that causes a computer to execute a process.

Images