JP2001312291A - Method for generating numeral voice waveform and method and device for synthesizing numerical voice - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a waveform generating method, a synthesizing method, and a synthesizing device for numeral voice which has natural intonation when a numeral having plural digits is continuously spoken. SOLUTION: While the lower two digits (ZZ' of a 4-digit numeral (XX'ZZ') are fixed to an arbitrary 2-digit or 1-digit, the higher two digits (XX') are varied from 00 to 99 and the 4-digit numeral is continuously spoken; and the continuous speech of the 4-digit numeral is recorded to obtain a numeral speech waveform of the higher two digits (XX'). Further, while the higher two digits (ZZ') of a 4-digits numeral (ZZ'XX') are fixed to a arbitrary 2-digit or 1-digit numeral, the lower two digits (YY') are varied from 00 to 99 and the 4-digit numeral is continuously spoken; and the continuous speech of the 4-digit numeral is recorded to obtain a numeral speech waveform of the lower two digits (YY'). When a 4-digit numeral voice is synthesized, fragments of the waveforms of the higher two digits (XX') and lower two digits (YY') are cut out of the numeral speech waveform and the 4-digit numeral voice is synthesized.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method for forming a numeric voice waveform, a method and an apparatus for synthesizing a numeric voice,
The present invention relates to a method of generating a numeric voice waveform, a method of synthesizing a numeric voice, and a device capable of generating a numeric voice having a natural intonation when a plurality of numeric voices are continuously generated.

[0002]

2. Description of the Related Art Conventionally, a service for synthesizing numeral voices and providing a telephone number to a customer has been put to practical use as seen in a telephone number guidance service and the like. In such a service, numbers are pronounced independently one by one, and when heard throughout, it feels unnatural.

Conventionally, a plurality of digits, for example, 4
As shown in FIG. 7, the fundamental frequency pattern that sounds natural when the digits of a digit are continuously uttered is such that the first two digits, for example, “zero” and “one” form one ridge, and Two digits, for example "Knee" and "Yon" form another ridge,
It is known that the height of the first ridge is “high” and the height of the second ridge is “medium”. In the case of a three-digit number, the first two digits have the same ridge shape and height as the first two digits in FIG. 7, and the third digit has the same ridge shape and height as the last two digits in FIG. Is known to be "medium" or "low".

[0004]

In the prior art such as the telephone number guidance service described above, the provided numeric voice sounds unnatural, and the customer pays attention to this unnaturalness and plays a correct numeric voice. The problem of oversight
In other words, there has been a problem that errors are likely to occur in information transmission.

[0005] Also, a naturally utterable utterance pattern of a plurality of digits, for example, a four-digit number, is prepared in advance,
Whenever an utterance request is made, the requested utterance pattern is reproduced and uttered.
Up to 9 utterance patterns, that is, 10 4 data must be prepared, and there is a problem in that the amount of data stored in the memory becomes very large. The problem is
When the number of digits to be uttered increases, there is a problem that the number increases by a power of 10.

SUMMARY OF THE INVENTION It is an object of the present invention to provide a method and an apparatus for synthesizing a numeric voice which can be heard as a numeric voice with natural inflections when a plurality of digits are continuously uttered. It is in. It is another object of the present invention to provide a method of creating a numeric voice waveform for reproducing a numeric voice of a plurality of digits with natural inflections even if the amount of data stored in the memory is small, and a method and apparatus for synthesizing numeric voices. .

[0007]

In order to achieve the above-mentioned object, the present invention relates to a method in which a single uttered waveform varying from 0 to 9 is combined with the last one or two digits of a three- or four-digit number. A waveform in which the first two digits are changed from 00 (zero zero) to 99 (cue queue) while the first two digits are fixed to an arbitrary one or two digits, and a three- or four-digit number is continuously uttered, is a fundamental frequency. The first two digits of a three- or four-digit number are fixed to an arbitrary two-digit number, and the second digit is set to 0 to 9
The waveform in which the three digits are continuously uttered while changing the second digit and the waveform in which the last two digits are continuously uttered by changing the two digits from 00 (zero zero) to 99 (cue cue) are converted into a low fundamental frequency number. A first feature is that the sound waveform is recorded. According to this feature, it is possible to greatly reduce the data amount of the numeric voice waveform stored in the memory, which is the basis of the voice synthesis of the numbers.

Further, the present invention is characterized in that a numeral voice waveform having a high fundamental frequency and a numeral voice waveform having a low fundamental frequency are combined to synthesize a three- or four-digit numeral voice. There are features. According to this feature, it is possible to reproduce a numeric voice with natural intonation.

Further, according to the present invention, a numeral string is divided into intervals of two digits from the beginning, a numeral voice waveform having a high fundamental frequency is designated for a numeral in an odd period, and a low-frequency sound waveform having a low fundamental frequency is designated for a numeral in an even period. A third feature is that a numeral voice waveform is designated and these numeral voice waveforms are combined to synthesize the voice of the numeral string. According to this feature, it is possible to reproduce a multi-digit numeric voice with natural intonation.

The present invention also provides a voice waveform storage unit for storing a numeric voice waveform having a high fundamental frequency and a numeric voice waveform having a low fundamental frequency, and a numeric string for dividing the numeric string into two-digit sections from the beginning. A dividing unit, a fundamental frequency height designating unit for designating a fundamental frequency height for each section, and a numeric speech waveform having a height designated by the fundamental frequency height designating unit are extracted from the speech waveform accumulating unit. A fourth characteristic lies in that an audio waveform extracting unit for performing the above-mentioned operations and an audio signal output unit for combining and outputting the numeric audio waveforms extracted from the audio waveform extracting unit are provided. According to this feature, it is possible to provide a digit voice synthesizing apparatus that reproduces a multi-digit numeric voice with natural intonation.

[0011]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS The present invention will be described below in detail with reference to the drawings. First, a method for producing a numeric voice waveform according to the present invention will be described.

When four-digit numbers are uttered continuously, the first two digits form a first ridge having a high fundamental frequency, and the second two digits form a second ridge having a medium or low fundamental frequency. Focusing on this, in the present invention, the waveform of the first two digits of the numeric utterance and the waveform of the second half of the two digits are separately generated, and these waveforms are arbitrarily combined to synthesize the four-digit numeric voice. To do it.

Therefore, in the present invention, as shown in FIG. 1, the first two digits XX 'are changed from 00 to 99 while the latter two digits are fixed to any two or one digit. Then, the four-digit number is continuously uttered, and the continuous utterance of the four-digit number is recorded and stored as the first two-digit number utterance waveform. For example, fix the last two digits to "knee" and "yon"
The first two digits XX ′ are changed from 00 (zero zero) to 99 (cue queue), and four digits are uttered continuously. That is, "zero""zero""knee""yon","zero"
"Ichi""Knee""Yon","Zero""Knee""Knee"
“Yeon”,…, “Cue” “Cue” “Knee” “Yon”
And record. As a result, 100 numeric utterance waveforms are stored as waveforms including the first two digits of XX 'utterance.

Next, on the contrary, as shown in FIG. 2, the first two digits are fixed to an arbitrary two-digit number, for example, "Knee" or "Yon", and the second half is YY. 'To "zero"
Change from "zero" to "queue""queue", 4
A continuous utterance is made with digit numbers, and the recorded utterance is stored as a waveform including the utterance of the last two digits YY '. As a result, 100 numeric utterance waveforms are stored as waveforms including the utterance of the last two digits YY '.

Further, assuming that a one-digit number is uttered, the ten uttered waveforms from the one-digit number “zero” to “cue” are converted into the first two half-digit numbers XX ′ of the 100 digits. It is stored in addition to the utterance waveform. Furthermore, assuming the case of uttering a three-digit number, the first two digits are fixed to an arbitrary two-digit number, and the last one digit is changed from “zero” to “cue” to three digits. Speak and record numbers continuously. Then, the ten utterance waveforms are stored in addition to the 100 latter two-digit YY 'numeral utterance waveforms.

Therefore, in the present invention, the first half number utterance waveform is 110 and the second half number utterance waveform is 110.
Individual waveforms will be accumulated.

Next, an embodiment of the method of synthesizing a numeric voice according to the present invention will be described.

(1) When the number to be uttered is one digit, the corresponding one-digit number utterance waveform is taken out of the first half number utterance waveform obtained by the above recording and reproduced.

(2) When the number to be uttered is two digits, a corresponding two-digit number utterance waveform fragment is cut out of the first half number utterance waveform obtained by the above-mentioned recording and reproduced.

(3) When the number to be uttered is three digits, the corresponding two-digit number utterance waveform is cut out of the first half of the number utterance waveform obtained by the above recording, and the second half of the number utterance is further cut out. From the waveform, a corresponding one digit digit utterance waveform fragment is cut out, and the two digit utterance waveform fragments are combined and reproduced. In this case, the sound section is cut out without silence before and after the first two digits and the last one digit voice section and combined.

(4) When the number to be uttered is four digits, a fragment of the corresponding two-digit number utterance waveform is cut out from the first half number utterance waveform obtained by the above recording, and the second half number utterance is further cut out. From the waveform, a corresponding two-digit number utterance waveform fragment is cut out, and the two-digit utterance waveform fragments are combined and reproduced. In this case, the sound section is cut out without any silence before and after the two-digit voice section before and after the two-digit voice section and combined. For example, as shown in FIG. 3, the first two digits XX 'and the last two digits YY'
Are combined to synthesize a 4-digit (XX'YY ') voice. If a discontinuity remains in the four-digit voice, it is preferable to insert a silent period of about 100 msec into the joint (point A in the figure) of the two preceding and following two-digit voice waveforms. The same applies to the case of (3) and (5) described later.

(5) If the number to be uttered is 5 digits or more,
The data is divided into four digits from the beginning, and the divided numbers are reproduced by applying the above-mentioned methods (1) to (4). For example, if the number to be uttered is five digits, the first four digits are reproduced by the method (4), and the fifth digit is reproduced by the method (1). If the number to be uttered is 10 digits, 1 digit to 4 digits
Digit, 5 to 8 digits are reproduced by the method of (4), respectively,
The 0th digit is reproduced by the method of (2). In this case, processing is performed independently for each four digits from the beginning, and 300 m is interposed between these four digits.
Connect with silence for about a second.

Next, an embodiment of the numeral speech synthesizing apparatus according to the present invention will be described. This synthesizing device can be realized using, for example, a computer. FIG. 4 is a functional block diagram showing the configuration of one embodiment of the present invention.

In the figure, reference numeral 1 denotes an input numeral string,
The digit string dividing unit 2 that divides each digit into two digits from the beginning adds the height (ie, pitch) of the fundamental frequency to the divided digits.
Is a two-digit and one-digit voice waveform corresponding to the first ridge and a two-digit and one-digit number corresponding to the second ridge of the fundamental frequency pattern. Voice waveform storage unit for storing voice waveforms of
The audio waveform extraction unit 5 extracts an audio waveform having a basic frequency suitable for the position designated by the fundamental frequency height designation unit 2 from the audio waveform storage unit 3, and the speech waveform extraction unit 5 inserts silence as necessary. And an audio signal output unit for outputting an audio waveform to a speaker or a telephone line.

The fundamental frequency height designating section 2 designates a high fundamental frequency for odd-numbered sections and a low fundamental frequency for even-numbered sections of the section divided by the digit string dividing section 1. . The audio waveform storage unit 3 stores a waveform including a first two-digit audio waveform 3a which is a waveform having a high basic frequency and a waveform including a second two-digit audio waveform 3b which is a waveform having a low basic frequency. When a high fundamental frequency is designated by the height designation unit 2, a speech waveform is cut out from the first two-digit speech waveform 3a by the speech waveform extraction signal a from the speech waveform extraction unit 4, while a low fundamental frequency is designated. Then, an audio waveform is cut out from the latter two-digit audio waveform 3b.

Next, the operation of this embodiment will be described with reference to FIGS. FIG. 5 is a flowchart for explaining the operation of the present embodiment. For example, when a numeral string {an} (n is a positive integer) of a telephone number is input to the numeral string division unit 1 (step S1), the numeral string division unit 1 converts the numeral string {an} into two digits from the beginning. (Step m2) (step S2). Next, setting m = 1 (step S3), the fundamental frequency height designation unit 2 determines the height of the fundamental frequency of the two-digit number (or one-digit number) of the m-th section (the first half voice waveform or the second half). (A voice waveform) (step S4). Next, the audio waveform extraction unit 4 extracts the two-digit number (or one-digit number) audio waveform of the m-th section from the designated area in the audio waveform storage unit 3 (step S5). Then, the audio signal output unit 5 outputs the extracted audio waveform to a speaker or a telephone line (Step S6). In step S7, it is determined whether or not all of the numeral string {an} has been output to the speaker or the telephone line. If the determination is negative, the process proceeds to step S8, where 1 is added to m. Then, again at step S4
And the same operation as described above is continued. The above operation is repeatedly performed, and when the determination in step S7 becomes positive, the operation of the present embodiment ends. The audio signal output unit 5 may output the first two digits and the second two digits as necessary.
A silence is inserted between the digits to generate a multi-digit voice with natural inflection.

As described above, according to the present embodiment, the voice waveform fragment of the first two digits and the voice waveform fragment of the second half are combined to reproduce the numeric voice of a plurality of digits. Therefore, it is possible to generate a natural numeral sound having a natural intonation and making it difficult for a listener to feel uncomfortable. In addition, the number of numeral voice waveforms stored in the voice waveform storage unit 3 is 110 for the first half and one-digit number and 110 for the second half and two-digit number, for a total of two.
Since only 20 waveforms are required, the amount of data stored in the audio waveform storage unit 3 can be significantly reduced as compared with the conventional one.

Next, a second embodiment of the present invention will be described with reference to FIG.
This will be described with reference to FIGS. FIGS. 6A to 6C are block diagrams showing only main parts of the present embodiment, and the other configuration is the same as or equivalent to FIG. In FIG. 3A, only the first half and one-digit audio waveforms (110 waveforms) 3a are stored in the audio waveform storage unit 3. The converter 6 receives the first two-digit audio waveform 3a as an input, converts the input to the height of the second half two-digit audio waveform, and outputs the same. Then, when the audio waveform extracting unit 4 receives the instruction of the first two-digit audio waveform, the switch 7 selects the terminal 8a, while the second half 2
When receiving the instruction of the digit voice waveform, the switch 7 selects the terminal 8b. As a result, only the first two-digit audio waveform (110 waveforms) 3a needs to be stored in the audio waveform storage unit 3, and the amount of stored data can be reduced.

FIG. 3B shows a modification of FIG. 3A, in which the second half digit and second digit audio waveforms (11
Only zero waveforms) 3b are stored. The converter 9
The latter two-digit audio waveform 3b is used as an input, and the input is converted into the height of the first two-digit audio waveform and output. Therefore, when the audio waveform extraction unit 4 receives the instruction of the first two-digit audio waveform, the switch 10 selects the terminal 11b, and when it receives the instruction of the latter two-digit audio waveform, the switch 10 selects the terminal 8a. As a result, only the latter two-digit two-digit audio waveforms (110 waveforms) 3b need to be stored in the audio waveform storage unit 3, and the amount of stored data can be reduced.

FIG. 3C shows still another modified example, in which the audio waveform storage unit 3 stores a 1-digit and 2-digit audio waveform having a middle height between the first and second half 1-digit and 2-digit audio waveforms. Only 110 waveforms) 3c are stored. First converter 12
Has the function of taking the intermediate two-digit audio waveform 3c as an input, converting the input to the height of the first two-digit audio waveform, and outputting the same. The second converter 13 receives the intermediate two-digit audio waveform 3c, converts the input into the height of the latter two-digit audio waveform, and outputs the converted signal. Then, when the audio waveform extraction unit 4 receives the instruction of the first two-digit audio waveform, the switch 1
The switch 4 selects the terminal 15a, while the switch 14 selects the terminal 15b when receiving the instruction of the latter two-digit audio waveform. As a result, the audio waveform accumulating unit 3 only needs to accumulate only two-digit audio waveforms (110 waveforms) 3c having an intermediate height.
The amount of accumulated data can be reduced.

It should be noted that, in the above-described embodiment, the sound waveform accumulating section 3 stores, as shown in FIG.
Although the continuous utterance waveform of the digit or four-digit number is stored, the present invention is not limited to this, and the necessary first half or two of the continuous utterance waveform of the three- or four-digit number,
Alternatively, only the waveform of the last one or two digits may be cut out in advance, and only these waveforms may be stored. By doing so, it is possible to reduce the storage capacity of the audio waveform storage unit 3 and the amount of calculation at the time of synthesis.

[0032]

As is apparent from the above description, according to the present invention, the numeric voice waveform having a high fundamental frequency and the numeric voice waveform having a low fundamental frequency are separately stored, so that they are stored in the memory. The amount of data to be performed can be greatly reduced as compared with the related art.

Further, according to the present invention, a numeric voice is synthesized by combining a numeric voice waveform fragment having a high fundamental frequency and a numeric voice waveform fragment having a low fundamental frequency.
A plurality of numeric voices having natural intonation can be generated, and a natural voice that does not make a listener feel uncomfortable can be synthesized. The present invention is preferably applied to an automatic voice response device or the like.

[Brief description of the drawings]

FIG. 1 is an explanatory diagram of a method for creating a first two-digit voice waveform of a four-digit numeric voice.

FIG. 2 is an explanatory diagram of a method of creating a voice waveform of the last two digits of a four-digit numeric voice.

FIG. 3 is an explanatory diagram of a method for synthesizing a four-digit numeric voice according to an embodiment of the present invention.

FIG. 4 is a block diagram showing a configuration of a numeral voice synthesizing apparatus according to an embodiment of the present invention.

FIG. 5 is a flowchart showing the operation of the embodiment.

FIG. 6 is a block diagram showing a configuration of a main part of a modified example of the present invention.

FIG. 7 is an explanatory diagram of a conventional technique.

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 1 ... Numeric string division | segmentation part, 2 ... Basic frequency height designation | designated part, 3 ... Audio waveform accumulation part, 4 ... Audio waveform extraction part, 5 ... Audio signal output part, 6, 9, 12, 13 ... Conversion part.

 ────────────────────────────────────────────────── ─── Continuing on the front page (72) Inventor Toru Shimizu 2-1-15 Ohara, Kamifukuoka-shi, Saitama F-term in K.D. Laboratory Inc. (reference) 5D045 AA09

Claims (8)

[Claims] 1. A waveform uttered independently by changing from 0 to 9 and a first half of a three- or four-digit number in a state where the last one or two digits are fixed to an arbitrary one or two-digit number. 2
A waveform in which the digits are changed from 00 (zero zero) to 99 (cue queue) and three or four digits are continuously uttered,
A method for creating a numeric voice waveform, characterized in that the waveform is recorded as a numeric voice waveform having a high fundamental frequency. 2. A waveform in which the first two digits of a three-digit or four-digit number are fixed to an arbitrary two-digit number, and the second half is changed from 0 to 9 to continuously utter a three-digit number, A waveform in which the last two digits are changed from 00 (zero-zero) to 99 (cue queue) and a four-digit number is uttered continuously, and a numerical voice waveform having a low fundamental frequency is recorded. How to make. 3. A three- or four-digit numeric speech is synthesized by combining a two-digit numeric speech waveform fragment having a high fundamental frequency and a one- or two-digit numeric speech waveform fragment having a low fundamental frequency. A method for synthesizing numeral voices, characterized in that: 4. A numeral string is divided into two-digit sections from the beginning, and a numeral voice waveform having a high fundamental frequency is designated for a number in an odd section, and a numeral speech waveform having a low fundamental frequency is designated for a number in an even section. A method for synthesizing a numeric voice, wherein the numeric voice is synthesized by combining designated numeric voice waveform fragments extracted based on these specifications. 5. The numerical voice waveform fragment having a high fundamental frequency and the numerical voice waveform fragment having a low fundamental frequency are respectively composed of the first two digits and the second half of the waveform created by the method according to claim 1 or 2. The method according to claim 3 or 4, wherein the method is a digit or a two-digit part. 6. A voice waveform storage unit for storing a numeric voice waveform having a high fundamental frequency and a numeric voice waveform having a low fundamental frequency, a numeric string dividing unit for dividing a numeric string into two-digit sections from the beginning, For each of the sections, a basic frequency height specifying unit for specifying a height of a basic frequency, and a voice waveform for extracting a numerical voice waveform fragment having a height specified by the basic frequency height specifying unit from the voice waveform storage unit. An apparatus for synthesizing a numeric voice, comprising: an extracting section; and an audio signal output section for combining and outputting the numeric audio waveform fragments extracted from the audio waveform extracting section. 7. A numerical voice waveform having a high basic frequency and a numerical voice waveform having a low basic frequency stored in the voice waveform storage section are respectively numerical voice waveforms created by the methods of claim 1 and 2. The numerical speech synthesizer according to claim 6, wherein 8. The voice waveform accumulating section stores only one of the numeric voice waveform having a high fundamental frequency and the numeric voice waveform having a low fundamental frequency, and the other numeric voice waveform is stored based on the one voice waveform. 7. The numerical speech synthesizer according to claim 6, wherein the numerical speech is created by frequency conversion.