JP2007041012A - Voice quality conversion device and speech synthesis device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To perform flexible voice quality conversion in good sound quality and to intuitively adjust voice quality conversion. <P>SOLUTION: In a voice quality converter, deformation is added in accordance with voice quality conversion control part 103 to a sound feature parameter string 100 as a result of analyzing voice based on a voice conversion parameter data base 105, and when at least one of feature variables by which the sound parameter string 100 or feature variables calculated from the sound feature parameter exceeds a predetermined threshold value 106, it is judged that voice quality is deteriorated as a result of voice quality conversion, and a content of voice quality conversion processing is adjusted so that quality deterioration may not arise. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention relates to a voice quality conversion device that converts voice quality and a voice synthesis device that can change voice quality.

  As a method for converting the voice quality of synthesized speech in a speech synthesizer that artificially generates speech, there are techniques described in Patent Document 1 (Japanese Patent Laid-Open No. 07-319495) and Patent Document 2 (Japanese Patent Laid-Open No. 08-152900). Exists.

  In the technique described in Patent Document 1, synthesis unit data obtained from voices of a plurality of speakers is stored in advance in a database. When a synthesis unit and voice quality to be used for speech synthesis are designated, first, synthesis unit data closest to the designated synthesis unit is selected from the database. Next, it is checked how much the voice quality of the selected synthesis unit data is different from the specified voice quality. If it is more than a predetermined level, the voice quality conversion is performed on the synthesis unit data so that it is close to the specified voice quality. Is done. Here, selection is performed by performing codebook mapping from the speaker codebook of the selected synthesis unit data (information expressing voice quality characteristics) to the codebook of the speaker with voice quality that matches the specified voice quality. The voice quality of the synthesized unit data is converted to the specified voice quality.

In the technique described in Patent Document 2, the voice quality of synthesized speech is converted by converting the sampling frequency when digital audio data is converted into an analog audio signal. Furthermore, so-called prosodic information (spectrum parameters) such as the fundamental frequency and phoneme duration is appropriately set according to the change of the sampling frequency so that the output sound is appropriate.
JP 07-319495 A Japanese Patent Laid-Open No. 08-152900

  However, when the conventional method is used, the degree of freedom of voice quality conversion is narrow or the voice quality is very difficult to adjust.

  For example, in the method of switching speech unit databases as in Patent Document 1, one speech unit database is required for one type of voice quality, and further, continuous change in voice quality cannot be expressed.

  Further, for example, in the method of converting the spectral parameters as in Patent Document 2, when voice quality conversion is performed simultaneously from a plurality of viewpoints such as “make a bright and angry voice”, the voice quality is not broken. It is difficult to maintain voice quality.

  The voice quality conversion apparatus according to the present invention includes a feature parameter conversion unit, a voice quality conversion control unit, and a voice quality parameter adjustment unit. The feature parameter conversion unit performs a predetermined conversion process on a plurality of parameters (acoustic feature parameter string) representing the acoustic features of the speech. The voice quality conversion control unit gives a conversion instruction according to the desired voice quality to the feature parameter conversion unit. The feature parameter conversion unit performs a conversion process on the acoustic feature parameter string in accordance with a conversion instruction from the voice quality conversion control unit. When the voice quality parameter adjustment unit assumes that the conversion process according to the conversion instruction from the voice quality conversion control unit is performed on the acoustic feature parameter string, the value of the acoustic feature parameter string after the conversion process is within a predetermined range. Voice quality conversion control so that the parameter falls within the predetermined range when at least one parameter of the converted acoustic feature parameter string exceeds the predetermined range. Change the conversion instruction from the section to the feature parameter conversion section.

  According to the present invention, it is possible to perform voice quality conversion with good sound quality and high flexibility, and it is possible to intuitively adjust voice quality conversion.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings. In the drawings, the same or corresponding parts are denoted by the same reference numerals, and the description thereof will not be repeated.

(Embodiment 1)
FIG. 1 shows a configuration of a voice quality conversion apparatus according to Embodiment 1 of the present invention, which is obtained as a result of analyzing speech, and an acoustic feature parameter sequence from which an original speech waveform can be obtained by performing resynthesis based on this 100 can be converted by the feature parameter conversion unit 101 to obtain a converted acoustic feature parameter sequence 102. The content of the conversion process performed by the feature parameter conversion unit 101 is instructed by the voice quality conversion control unit 103, and the voice quality conversion parameter adjustment unit 104 converts the content of the conversion process instructed by the voice quality conversion control unit 103 into the modified acoustic feature parameter 102. It is possible to detect the case where the deterioration occurs and work on the voice quality conversion control unit 103 to adjust the content of the conversion process so as not to cause the sound quality deterioration. The voice quality conversion parameter database 105 holds data serving as a template when the feature parameter conversion unit 101 performs voice quality conversion processing. The voice synthesis parameter limit value database 106 stores limit conditions (such as limit values of each acoustic feature parameter) for obtaining a synthesized speech maintaining naturalness with respect to the acoustic feature parameter sequence, and the voice quality conversion parameter The adjustment unit 104 can also adjust the content of the voice quality conversion process based on the data stored in the speech synthesis parameter limit value database 106.

  Hereinafter, the flow of processing when performing voice quality conversion processing in the present embodiment will be specifically shown.

  FIG. 2 shows an initial operation state of the voice quality conversion apparatus according to the present embodiment.

  The voice quality conversion control unit 103 presents the user with voice quality conversion parameter adjustment controls corresponding to each of one or more voice qualities. The voice quality conversion parameter adjustment control includes a movable range corresponding to the convertible amount of the corresponding voice quality, and a pointer that can indicate an arbitrary position within the movable range. The voice quality conversion control unit 103 causes the user to operate the pointer of the voice quality conversion parameter adjustment control to designate the voice quality and the conversion amount. For example, as shown in FIG. 2, the voice quality conversion control unit 103 sets the voice quality conversion parameters corresponding to convertible voice qualities (in this case, brightness, darkness, masculinity, and fast speech) in the range of 0 to 10, respectively. A slider (or a volume switch or the like) that can be continuously adjusted is presented to the user. The user designates a voice quality to be converted and a conversion amount by moving a slider corresponding to each voice quality conversion parameter within a range of 0 to 10. Here, it is assumed that 0 is not applied and 0 is approached as the degree of adjustment increases. In FIG. 2, the brightness is shown as “bright”, the darkness as “dark”, the masculinity as “male”, and the quick mouth as “early”. The acoustic feature parameter column 100 includes, as adjustable acoustic feature parameters, a fundamental frequency (F0), a first formant frequency (F1), a second formant frequency (F2), a frame duration (FR), and a sound source for each analysis frame. It has 5 parameters of power (PW). The voice quality conversion parameter database 105 holds a value to be added to each acoustic feature parameter value when the value of the voice quality conversion parameter of the voice quality conversion control unit 103 is adjusted (when it is increased by 1). In FIG. 2, since the set values of all voice quality conversion parameters are 0 on the voice quality conversion control unit 103, the contents of the output acoustic feature parameter sequence 102 are the same as the input acoustic feature parameter sequence 100.

  FIG. 3 shows an operation example in a state where the voice quality conversion control unit 103 sets the brightness voice quality conversion parameter to 5 and the fast voice quality conversion parameter to 3. When focusing on an analysis frame in the input acoustic feature parameter sequence 100, for example, if the value of the acoustic feature parameter F0 in the analysis frame is 300, F0 in the corresponding analysis frame in the output acoustic feature parameter sequence 102 is assumed. Since the voice quality conversion parameter database 105 shows that the value of +1 becomes +5 every time the voice quality conversion parameter of brightness is increased by 1 and becomes +1 every time the voice quality conversion parameter of the quick mouth is increased by 1, 300 + 5 × (+5) + 3 × ( +1) = 328. Similar calculations may be performed for other acoustic feature parameter values.

  FIG. 4 shows the result of setting the darkness voice quality conversion parameter to 7 by the voice quality conversion control unit 103 after the state shown in FIG. According to the voice quality conversion parameter database 105, the darkness voice quality conversion parameter has the opposite contents to the brightness voice quality conversion parameter. Therefore, when the darkness voice quality conversion parameter is increased, the voice quality conversion parameter of brightness is increased. If is greater than 0, the same effect can be obtained by first reducing the voice conversion parameter for brightness instead of increasing the voice conversion parameter for darkness. In the case of FIG. 4, the voice conversion parameter for darkness is changed to 7 from the state where the voice conversion parameter for brightness is 5 in FIG. 3. Since the same effect as increasing the voice quality conversion parameter by 5 is obtained, instead of setting the dark voice quality conversion parameter to 7, the brightness voice quality conversion parameter should be set to 0 and the dark voice quality conversion parameter should be set to 2. . As described above, the voice quality conversion processing specified by the voice quality conversion control unit 103 is adjusted by the voice quality conversion parameter adjustment unit 104 to a combination that minimizes the value of each voice quality conversion parameter. A conversion control interface is realized.

  Furthermore, it is assumed that when the value of the acoustic feature parameter F0 exceeds 350, the sound quality of the synthesized sound output by re-synthesis of the acoustic feature parameter sequence is significantly deteriorated. FIG. 5 shows a state in which the voice quality conversion parameter for brightness is set to 10 and the value of F0 is 350 in an analysis frame having an acoustic feature parameter sequence output as a result.

  FIG. 6 shows a state where the voice quality conversion parameter value of the quick mouth is further increased to 5 from the state of FIG. When the voice quality conversion parameter value of the fast mouth is increased by 5, the value of the acoustic feature parameter F0 increases by 5. However, since the value of the acoustic feature parameter F0 is already 350 in a certain analysis frame in FIG. By reducing the value by 1, the value of the acoustic feature parameter F0 is set to 350. As a result, the voice quality conversion parameter value for brightness is 9, and the voice quality conversion parameter value for early speech is 5. Thus, by automatically adjusting the value of the voice quality conversion parameter by the voice quality conversion parameter adjustment unit 104, the user can perform the voice quality conversion operation without being aware of the limit value of each acoustic feature parameter.

  The voice quality parameter adjustment unit 104 determines that the value of the acoustic feature parameter string after the conversion process is predetermined when the conversion process according to the conversion instruction from the voice quality conversion control unit 103 is performed on the acoustic feature parameter string 100. If at least one parameter of the converted acoustic feature parameter string exceeds the predetermined range, the voice quality is set so that the parameter falls within the predetermined range. A conversion instruction from the conversion control unit 103 to the feature parameter conversion unit 101 is changed.

  Note that the limit condition stored in the speech synthesis parameter limit value database 106 (for example, the limit value of each acoustic feature parameter such that the value of the acoustic feature parameter F0 should not exceed 350, the acoustic feature parameter F0 and the acoustic The voice quality conversion parameter adjustment unit 104 may refer to the data (the data that the result of adding the value of the feature parameter F2 should not exceed 2000) as necessary.

  The change given to the acoustic feature parameter sequence 100 by the feature parameter conversion unit 101 may not be uniform for all analysis frames, and the contents of the voice quality conversion parameter database 105 change for each analysis frame. Also good.

  Note that the voice quality conversion parameter value adjustment by the voice quality conversion parameter adjustment unit 104 may be automatically performed using a constraint satisfaction algorithm. An example of the constraint satisfaction algorithm is the Indigo algorithm of Reference 1. A portion relating to the acoustic feature parameter F0 of the constraint condition for adjusting the voice quality conversion parameter shown in FIGS. 5 and 6 is described in the constraint hierarchy of the Indigo algorithm as follows.

REQUIRED constraint C1: Output F0 ≦ 350
REQUIRED constraint C2: input F0 = 300
REQUIRED constraint C3: Brightness × 5 = t1
REQUIRED constraint C4: darkness × −5 = t2
REQUIRED constraint C5: masculinity x-3 = t3
REQUIRED constraint C6: Hayaku × 1 = t4
REQUIRED constraint C7: t1 + t2 = t5
REQUIRED constraint C8: t3 + t4 = t6
REQUIRED constraint C9: t5 + t6 = t7
REQUIRED constraint C10: input F0 + t7 = t8
REQUIRED constraint C11: t8 = output F0
STRONG restriction C12: Hayato = 5
WEAK restriction C13: masculinity = 0
WEAK constraint C14: Darkness = 0
WEAK restriction C15: Brightness = 10
The above constraint conditions are represented in the form of a constraint graph as shown in FIG. Variables t1 to t8 are variables for holding intermediate results of constraint satisfaction calculation. Although omitted for simplicity, in order to obtain a more desirable result, it is desirable to provide a REQUIRED constraint that constrains the value of each voice quality conversion parameter to a value between 0 and 10.

  An outline of processing when the above constraint conditions are solved by the Indigo algorithm is shown below.

Initial state: The range of all variables is [−∞, + ∞]
C1 addition: The range of the output F0 becomes [−∞, 350] C2 addition: The range of the input F0 becomes [300, 300] C3 to C10 addition: No change in the range of each variable C11 addition: The range of t8 is [−∞, 350]
Propagation of C10 and the value range of t7 becomes [−∞, 50].
Propagation of C6 and the value range of t4 becomes [5,5] C13 addition: The value range of masculinity becomes [0,0]
Propagates C5 and the range of t3 becomes [0,0]
Propagates C8 and the range of t6 becomes [5,5]
Propagation of C9 and the value range of t5 becomes [−∞, 45] C14 addition: Darkness value range becomes [0, 0]
C4 is propagated and the range of t2 becomes [0,0]
Propagates C7 and the value range of t1 becomes [−∞, 45]
Propagation of C3 and brightness value range becomes [−∞, 9] C15 addition: Brightness value range becomes [9, 9] <Reference 1>
A. Borning, R. Anderson, B. Freeman-Benson: The Indigo Algorithm, TR 96-05-01, Department of Computer Science and Engineering, University of Washington, July 1996
(Embodiment 2)
An example of an arrangement determination method for voice quality conversion parameter adjustment control according to Embodiment 2 of the present invention will be described with reference to FIG.

  Assume that the difference value data of each acoustic feature parameter stored in the voice quality conversion parameter database 105 is as shown in FIG. If the closeness of the content of the change that each voice quality conversion parameter has on the acoustic feature parameter is defined as the Euclidean distance between the difference values of each acoustic feature parameter stored in the voice quality conversion parameter database 105, each distance is shown in Table 800. Can be calculated as follows. After that, the voice quality conversion parameter adjustment control may be arranged so that the closer the distance is, the closer it is arranged. In Table 800, the voice quality conversion parameters of masculinity and darkness are the closest, followed by brightness and fast mouth, and then fast mouth and darkness. Therefore, the arrangement shown in 801 can be performed.

  By performing such an arrangement and further adjusting the voice quality conversion parameters by the method as described below, the user can perform the voice quality conversion operation in an intuitive and easily understandable manner.

  STEP 0: The value of a certain voice quality conversion parameter is changed by the user. For example, as shown in 802a and 802b of FIG. 9, it is assumed that the voice quality conversion parameter of brightness (bright) is changed to 10 from the state where all voice quality parameters are 0. Here, it is assumed that conditions such as the input acoustic feature parameter value and the limit value of the acoustic feature parameter are the same as those shown in FIG.

  STEP 1: When the voice quality conversion parameter i is further increased by Xi, a number Xi is calculated such that any one of the acoustic feature parameters or feature variables that can be calculated based on the value of the acoustic feature parameter falls outside the preset value range. This is calculated for all voice quality conversion parameters i. For example, when the voice quality conversion parameter for brightness is set to 10 as in 802b of FIG. 9, the value of the acoustic feature parameter F0 exceeds the limit value 350 when the voice quality conversion parameter for early mouth (early) is set to greater than 0. As a result, the sound quality of the synthesized sound deteriorates. Therefore, Xi for the fast voice conversion parameter is 0.

  STEP 2: The current value of each voice quality conversion parameter i is Yi, Xi + Yi is the maximum value of the voice quality conversion parameter adjustment control of the voice quality conversion parameter i, 0 is the minimum value, and voice quality conversion parameter adjustment control is performed so that Yi is the current value. Redraw. This is performed for all voice quality conversion parameters i. In the case of the above example, redrawing is performed so that the maximum value = 0, the minimum value = 0, and the current value = 0 as shown in 802b of FIG. If the brightness voice quality conversion parameter is lowered from 10 to 9 in this state (802b), the maximum value of the early voice quality conversion parameter is as shown by 802c in FIG. Increase from 0 to 5.

  According to the above method, when the value of a certain voice quality conversion parameter is increased, a voice quality conversion parameter whose content of change applied to the acoustic feature parameter is similar to that is calculated based on which acoustic feature parameter or acoustic feature parameter. Since there is no room for the value of the characteristic variable to be increased without deviating from the predetermined range, Xi in STEP1 is reduced, and the current setting value Yi is set to the maximum settable value by redrawing in STEP2. You can draw as you approach. In the present embodiment, since the content of the change given to each acoustic feature parameter is arranged closer to the voice quality conversion parameter adjustment control, when the user operates a certain voice quality conversion parameter adjustment control, the user is placed closer. The voice quality conversion parameter adjustment control that is displayed is moved in the same direction, and the voice quality conversion parameter adjustment control that is arranged farther is moved in the opposite direction. Therefore, the user can intuitively understand how the voice quality is converted by operating the slider (voice quality conversion parameter control). In addition, since the maximum value that can be set for each voice quality conversion parameter is automatically changed in conjunction with the voice quality conversion parameter setting, each acoustic feature parameter that allows the user to convert voice quality without breaking the voice quality Voice quality conversion operation can be performed without being aware of the limit value.

  In this example, the maximum value of the related voice quality conversion parameter is changed in conjunction with the voice quality conversion parameter setting by the user. However, as shown in 802d to 802f in FIG. It is also possible to change related voice quality conversion parameters. For example, as shown in FIG. 802e, when the voice quality conversion parameter of brightness is set to 10, the quick voice quality conversion parameter is also set to 10 in conjunction with this. However, here, the quick voice quality conversion parameter slider is simply moved to the position of 10, and the conversion of the acoustic feature parameter is not performed in response to the quick voice quality conversion parameter being set to 10. As described above, when the voice quality parameter of brightness is set to 10, if the voice quality parameter of the fast mouth is set to 1 or more, the value of the acoustic feature parameter F0 exceeds the limit value 350 and the sound quality of the synthesized sound is deteriorated. However, as shown in FIG. 802e, since the voice quality conversion parameter for brightness is set to 10 in conjunction with the voice quality conversion parameter for brightness, the value of the voice quality conversion parameter for early speed is also set to 10 (maximum value). Can not be increased any more. When the voice quality conversion parameter for brightness is lowered from 10 to 9 in this state (802e), the early voice quality conversion parameter is lowered from 10 to 5 as shown at 802f in FIG. However, in this case as well, the quick voice quality conversion parameter slider is simply moved to the position 5, and the acoustic feature parameter conversion is not performed in response to the quick voice quality conversion parameter being set to 5. In this way, since the value of each voice quality conversion parameter is automatically changed in conjunction with the setting of the voice quality conversion parameter, the user can limit the acoustic feature parameters for converting the voice quality without breaking the voice quality. Voice quality conversion can be performed without being aware of the value.

(Embodiment 3)
An example of an arrangement determination method for voice quality conversion parameter adjustment control according to Embodiment 3 of the present invention will be described with reference to FIG.

  The table 800 can be created from the voice quality conversion parameter database 105 by a method similar to the method described in the second embodiment. Based on the table 800, the lower limit of each voice quality conversion parameter adjustment control is summarized as a single point, and in the table 800, the voice quality conversion parameters closer to each other are arranged so that the angle between the voice quality conversion parameter adjustment controls becomes smaller. Shown in

  When such an arrangement is performed and the voice quality conversion parameter is adjusted by the same method as described in the second embodiment, another voice quality conversion parameter adjustment control is operated as shown in 901 by another operation. The voice quality conversion parameter adjustment control also appears to move in the same direction. This makes it possible to provide a voice quality conversion interface that is intuitively easy for the user to understand.

(Embodiment 4)
The configuration of a speech synthesizer according to Embodiment 4 of the present invention is shown in FIG. This device is a variable voice quality speech synthesizer capable of performing speech synthesis with various voice qualities by using text information as an input.

  In this configuration, the speech synthesizer 1002 selects data matching the input text information 1001 from the speech units stored in the speech synthesis database 1000, and generates the acoustic feature parameter sequence 100 based on the selected data. The acoustic feature parameter sequence 100 is deformed by the method shown in the first, second, and third embodiments, and the transformed acoustic feature parameter sequence 102 is output. The waveform generation unit 1003 generates a synthesized sound waveform 1004 using the deformed acoustic feature parameter sequence 102 and reproduces it by the speaker device 1005, whereby the voice quality conversion control unit 103 converts arbitrary text information 1001 input by the user. It can be played back with the synthesized voice of the specified voice quality.

(Embodiment 5)
FIG. 12 shows the configuration of a speech synthesizer according to Embodiment 5 of the present invention. This device is a variable voice quality speech synthesizer capable of performing speech synthesis with various voice qualities. This configuration enables the same effect as that of the above-described fourth embodiment with another configuration. The difference between this configuration and the fourth embodiment is that, in the fourth embodiment, the voice quality conversion parameter adjustment unit 104 uses the acoustic feature parameter sequence output from the voice synthesis unit 1002 to detect quality degradation of the output speech. In this embodiment, some or all of the data stored in the speech synthesis database 1000 is used, and acoustic feature parameter values appearing in the acoustic feature parameter sequence 100 output from the speech synthesis unit 1002 based on the speech synthesis database 1000 are used. This is the point of predicting the trend and adjusting the voice quality conversion parameter. For example, the voice quality conversion parameter adjustment unit 104 selects all the phoneme segment data one by one from the speech synthesis database 1000, and the quality of the output speech deteriorates when the segment data is converted according to the voice quality conversion parameter adjustment unit 104. The voice quality conversion parameter adjustment process may be performed based on whether or not to do so.

  The advantage of the present embodiment over the fourth embodiment is that the content of the adjustment processing of the voice quality conversion parameter is the same regardless of what text information 1001 is input unless the speech synthesis database 1000 is replaced in the present embodiment. Is all you need to do. However, if the contents of the acoustic feature parameter sequence 100 differ greatly depending on the contents of the text information 1001, the quality of the output voice may be deteriorated depending on the contents of the text information 1001.

  Note that the acoustic feature parameter sequence 100 in this embodiment does not have to be generated directly from the speech synthesis database 1000 by speech synthesis processing, and the voice quality of the acoustic feature parameter sequence 100 generated by some method is speech synthesis. If the content of the database 1000 is sufficiently similar, the same effect can be obtained.

(Embodiment 6)
FIG. 13 shows the configuration of a speech synthesizer according to Embodiment 6 of the present invention. This device is a variable voice quality speech synthesizer capable of performing speech synthesis with various voice qualities. This configuration is another configuration in which the same effect as that of the fifth embodiment can be expected. The difference between this configuration and the fifth embodiment is that the voice quality conversion parameter adjustment unit 104 directly refers to the contents of the voice synthesis database 1000 in the fifth embodiment, whereas the voice synthesis database 1000 in the present embodiment. From the above, the voice synthesis database feature table 1200 in which only necessary and sufficient data for detecting quality degradation due to voice quality conversion processing is extracted in advance, and voice quality conversion parameter adjustment processing is performed based on this. For example, the voice synthesis database feature table 1200 can be obtained by extracting only the upper limit value, lower limit value, and average value of each acoustic feature parameter from all the data stored in the voice synthesis database 1000. Reference numeral 104 can perform voice quality conversion parameter adjustment processing by referring to only data necessary and sufficient for detecting quality degradation of output speech at high speed.

  As in the fifth embodiment, in this embodiment, the acoustic feature parameter sequence 100 does not have to be generated directly from the speech synthesis database 1000 by the speech synthesis process, but is generated by some method. If the voice quality of the parameter string 100 sufficiently matches the content of the speech synthesis database feature table 1200, the same effect can be obtained.

(Embodiment 7)
FIG. 14 shows the configuration of a speech synthesizer according to Embodiment 7 of the present invention. This device is a variable voice quality speech synthesizer capable of performing speech synthesis with various voice qualities by using text information as an input. In this configuration, the voice synthesis unit 1002 implements voice quality variable voice synthesis by using the data stored in the voice synthesis database 1000 for voice synthesis after performing the voice quality conversion processing by the feature parameter conversion unit 101. The voice quality conversion parameter adjustment by the voice quality conversion parameter adjustment unit 104 is performed with reference to the speech synthesis database 1000 as in the fifth embodiment.

  Which configuration of the present embodiment or the fifth embodiment is taken may be selected in view of ease of mounting.

  Note that the voice quality conversion parameter adjustment unit 104 uses the voice synthesis database feature table containing data necessary and sufficient to detect the quality deterioration of the output voice as in the sixth embodiment, and uses the voice quality conversion parameter feature table. The adjustment process may be performed.

(Embodiment 8)
FIG. 15 shows the configuration of a voice quality conversion apparatus according to the eighth embodiment of the present invention. This device is a voice quality conversion device that receives voice waveform data and outputs the converted voice waveform data. In this configuration, the input speech waveform 1400 is analyzed by the speech analysis unit 1401, and the acoustic feature parameter string 100 as an analysis result is input to the voice quality conversion parameter adjustment unit 104 and the feature parameter conversion unit 101.

  With this configuration, voice quality conversion processing can be applied to not only synthesized speech generated from text information but also to real voice waveform data through an interface that is intuitively easy to operate while preventing quality degradation.

(Embodiment 9)
FIG. 16 shows the configuration of a voice quality conversion apparatus according to Embodiment 9 of the present invention. This device is a voice quality conversion device that receives voice waveform data and outputs the converted voice waveform data. The difference between the configuration in the present embodiment and the configuration in the eighth embodiment is that the voice quality conversion parameter adjustment unit performs the voice quality conversion parameter adjustment processing with reference to the waveform feature table 1500 instead of the acoustic feature parameter sequence 100. It is. For example, the waveform feature table 1500 extracts in advance only data necessary and sufficient to detect quality degradation due to voice quality conversion processing from the result of analyzing sample speech previously uttered by the same speaker who uttered the speech waveform 1400 For example, only the upper limit value, lower limit value, and average value are extracted from each acoustic feature parameter of the analysis result of the sample voice.

  There may be a plurality of waveform feature tables 1500. For example, the waveform feature table 1500 that best represents the features of the speech waveform 1400 and the acoustic feature parameter string 100 is selected and used according to attributes such as the speaker's age and sex. You can also.

  With this configuration, the voice quality conversion parameter adjustment unit 104 can perform the voice quality conversion parameter adjustment processing prior to the input of the acoustic feature parameter sequence 100, and further, a waveform that represents the features of the acoustic feature parameter sequence 100 in a compact manner. Use of the feature table 1500 makes it possible to perform voice quality conversion parameter adjustment processing at high speed.

  The voice quality conversion apparatus of the present invention is useful as an agent application or a text-to-speech application that uses synthesized speech, a communication apparatus that uses a voice quality conversion function, a voice quality editor apparatus, or the like.

1 is a block diagram showing a configuration of a voice quality conversion device according to Embodiment 1. FIG. It is the schematic showing the operation | movement initial state of the voice quality conversion apparatus shown in FIG. It is the schematic showing the state which operated two voice quality conversion parameter values after the state shown in FIG. It is the schematic showing the voice quality conversion parameter adjustment process at the time of operating one voice quality conversion parameter value after the state shown in FIG. It is the schematic showing the state which operated one voice quality conversion parameter value from the initial state with the voice quality conversion apparatus shown in FIG. It is the schematic showing the voice quality conversion parameter adjustment process at the time of operating one voice quality conversion parameter value after the state shown in FIG. It is a restriction | limiting graph in the case of performing a voice quality conversion parameter adjustment process with an Indigo algorithm. FIG. 10 is a schematic diagram illustrating a method for arranging voice quality conversion parameter adjustment controls according to a second embodiment. FIG. 10 is a schematic diagram illustrating a method for arranging voice quality conversion parameter adjustment controls according to a second embodiment. 10 is a schematic diagram illustrating an arrangement method and an operation example of voice quality conversion parameter adjustment control according to Embodiment 3. FIG. FIG. 10 is a block diagram illustrating a configuration of a voice quality variable speech synthesizer according to a fourth embodiment. FIG. 10 is a block diagram illustrating a configuration of a voice quality variable speech synthesizer according to a fifth embodiment. It is a block diagram which shows the structure of the voice quality variable speech synthesizer by Embodiment 6. FIG. FIG. 10 is a block diagram illustrating a configuration of a voice quality variable speech synthesizer according to a seventh embodiment. It is a block diagram which shows the structure of the voice quality conversion apparatus by Embodiment 8. FIG. It is a block diagram which shows the structure of the voice quality conversion apparatus by Embodiment 9. FIG.

Explanation of symbols

100 acoustic feature parameter sequence 101 feature parameter conversion unit 102 converted acoustic feature parameter sequence 103 voice quality conversion control unit 104 voice quality conversion parameter adjustment unit 105 voice quality conversion parameter database 106 speech synthesis parameter limit value database 800 mutual distance table 801 between voice quality conversion parameters Example of Arrangement of Voice Quality Conversion Parameter Adjustment Control in Voice Quality Conversion Control Unit 900 Example of Arrangement of Voice Quality Conversion Parameter Adjustment Control in Voice Quality Conversion Control Unit 901 Example of Operation of Voice Quality Conversion Parameter Adjustment Control Arranged in Voice Quality Conversion Control Unit 1000 Synthetic database 1001 Text information 1002 Speech synthesis unit 1003 Waveform generation unit 1004 Synthetic sound waveform 1005 Speaker device 1200 Speech synthesis database feature table 1400 Speech waveform 1401 Speech analysis unit t1 to t8 In constraint graph, each constraint in the constraint graph variables C1~C15 7 for holding the intermediate result of the constraint satisfaction calculation conditions

Claims (16)

A feature parameter conversion unit that performs a predetermined conversion process on a plurality of parameters (acoustic feature parameter string) representing acoustic features of speech;
A voice quality conversion control unit that gives a conversion instruction according to a desired voice quality to the feature parameter conversion unit;
A voice quality parameter adjustment unit,
The feature parameter converter is
According to the conversion instruction from the voice quality conversion control unit, performs a conversion process on the acoustic feature parameter sequence,
The voice quality parameter adjustment unit
If the conversion process according to the conversion instruction from the voice quality conversion control unit is assumed to have been performed on the acoustic feature parameter string, whether the value of the acoustic feature parameter string after the conversion process is within a predetermined range When at least one parameter in the acoustic feature parameter string after the conversion process exceeds the predetermined range, the voice quality conversion control unit makes the characteristic so that the parameter falls within the predetermined range. Change the conversion instructions to the parameter converter,
A voice quality conversion device characterized by that. In claim 1,
A speech synthesis parameter limit value database that holds limit conditions for obtaining a synthesized speech that maintains naturalness for the acoustic feature parameter sequence;
The voice quality parameter adjustment unit
Perform the determination process and the conversion instruction change process with reference to the speech synthesis parameter limit value database,
A voice quality conversion device characterized by that. In claim 1,
The voice quality parameter adjustment unit
Change the conversion instruction using the constraint satisfaction algorithm.
A voice quality conversion device characterized by that. In claim 1,
For each of one or more convertible voice qualities, further comprising a voice quality conversion parameter database that associates the conversion amount of the voice quality with the conversion amount of the acoustic feature parameter sequence necessary for performing the voice conversion of the conversion amount,
The voice quality conversion control unit
Presenting one or more convertible voice qualities to the user, allowing the user to specify a voice quality and a conversion amount of the voice quality, and converting the feature parameter based on the voice quality and conversion amount specified by the user and the voice quality conversion parameter database Give conversion instructions to
A voice quality conversion device characterized by that. In claim 4,
The voice quality conversion control unit
Presenting to the user voice quality conversion parameter adjustment controls corresponding to each of the one or more voice qualities;
The voice quality conversion parameter adjustment control is
A movable range corresponding to the convertible amount of the corresponding voice quality, and a pointer capable of indicating an arbitrary position within the movable range,
The voice quality conversion control unit
Allowing the user to specify the voice quality and the conversion amount by operating the pointer of the voice quality conversion parameter adjustment control;
A voice quality conversion device characterized by that. In claim 5,
The voice quality conversion parameter adjustment control has the shape of a slider or a volume switch.
A voice quality conversion device characterized by that. In claim 5,
The voice quality conversion control unit
When the pointer of the voice quality conversion parameter adjustment control corresponding to a certain voice quality is operated by the user, the conversion amount of the acoustic feature parameter string corresponding to the conversion volume of the voice quality corresponding to the operation and the acoustic feature parameter string corresponding to the other voice quality Change the movable range of the voice quality conversion parameter adjustment control corresponding to the other voice quality based on the conversion amount of
A voice quality conversion device characterized by that. In claim 5,
The voice quality conversion control unit
When the pointer of the voice quality conversion parameter adjustment control corresponding to a certain voice quality is operated by the user, the acoustic feature parameter string corresponding to the voice quality conversion amount corresponding to the operation and the acoustic feature parameter string corresponding to the other voice quality Change the pointer position of the voice quality conversion parameter adjustment control corresponding to the other voice quality based on the conversion amount of
A voice quality conversion device characterized by that. In any one of claims 5 to 8,
The voice quality conversion parameter adjustment control is
In accordance with the operation of each voice quality conversion parameter adjustment control, it is arranged so that the content of the change added to the acoustic feature parameter sequence is closer to the similar one,
A voice quality conversion device characterized by that. In any one of claims 5 to 8,
The voice quality conversion parameter adjustment control is
It is arranged so that the angle between the controls becomes smaller as the content of the change added to the acoustic feature parameter sequence is similar according to the operation of each voice quality conversion parameter adjustment control,
A voice quality conversion device characterized by that. A voice quality conversion device according to claim 1;
A speech synthesizer that outputs text characteristic parameter strings as input text information,
The feature parameter conversion unit of the voice quality conversion device performs the predetermined conversion process on the acoustic feature parameter from the speech synthesis unit.
A speech synthesizer characterized by the above. A voice quality conversion device according to claim 1;
A speech synthesis database used for speech synthesis processing,
The voice quality conversion parameter adjustment unit of the voice quality conversion device detects that the quality degradation of the output voice occurs when performing the conversion process as instructed by the voice quality conversion control unit with reference to the voice synthesis database,
A speech synthesizer characterized by the above. A voice quality conversion device according to claim 1;
A speech synthesis database feature table containing characteristic information for detecting a conversion process that causes quality degradation in the output speech related to the speech synthesis database used in the speech synthesis process;
The voice quality conversion parameter adjustment unit of the voice quality conversion device detects that the quality degradation of the output voice occurs when performing the conversion process as instructed by the voice quality conversion control unit with reference to the voice synthesis database feature table,
A speech synthesizer characterized by the above. A voice quality conversion device according to claim 1;
A speech synthesizer that outputs an acoustic feature parameter sequence as input of text information;
A speech synthesis database used for speech synthesis processing,
The data stored in the speech synthesis database is converted by the feature parameter conversion unit, and speech synthesis is performed using the converted data.
A speech synthesizer characterized by the above. In claim 1,
A voice analysis unit for converting the voice waveform data into an acoustic feature parameter sequence;
The input voice waveform data is converted into an acoustic feature parameter string by the voice analysis unit, and converted by the feature parameter conversion unit and output, thereby performing voice quality conversion processing of the voice waveform data.
A voice quality conversion device characterized by that. In claim 15,
One or more waveform feature tables containing characteristic information for detecting a conversion process that causes degradation of quality in output speech when speech waveform data having a specific voice quality is converted,
Detecting quality degradation of the output voice using information corresponding to the input voice waveform data in the waveform feature table;
A voice quality conversion device characterized by that.

Images