JP2601302B2 - Pitch frequency generator in speech synthesizer - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Summary] The present invention relates to a pitch frequency generator in a speech synthesizer,
In particular, when generating a pitch frequency of a time-varying voice of an output voice of a voice synthesizer that inputs information about a text, converts the text into voice information to be read, and outputs the voice information, a time-series The present invention relates to a pitch frequency generator in a speech synthesizer that synthesizes a plurality of pitch patterns having characteristics that monotonically decrease from a peak value at each time, and can generate a pitch pattern in as short a time as possible. In order to generate a pitch frequency of a time-varying voice of a voice output from a voice synthesizer that inputs information about a text, converts the text into voice information to be read, and outputs the voice,
A pitch frequency generation device in a speech synthesis device having a plurality of pitch patterns that are input in a time series according to the characteristics of a sentence and have a characteristic of monotonically decreasing from a peak value at each time, and that are synthesized. A tendency detecting means for generating a decreasing tendency signal when the pattern is monotonically decreasing, a threshold detecting means for producing a threshold signal when the pitch pattern is equal to or less than a predetermined threshold, and receiving the decreasing tendency signal and the threshold signal. At this time, the apparatus is configured to have an overlay stopping means for stopping the generation of the pitch pattern.

[Industrial applications]

The present invention relates to a pitch frequency generation device in a speech synthesizer, and in particular, to input information related to a sentence, convert the sentence into speech information to be read out, and output the time-varying pitch of the output speech of the speech synthesis device. When generating frequencies, they are input in chronological order according to the features of the text,
The present invention relates to a pitch frequency generation device in a voice synthesis device having a superposition unit that superposes and synthesizes a plurality of pitch patterns having a characteristic that monotonously decreases from a peak value at each time.

[Conventional technology]

In general, there are the following speech synthesizers for inputting information about a sentence, converting the information into speech information for reading out the sentence, and outputting the speech information. This is a device for proofreading a printed matter, for example, a newspaper, and proofreading whether a sentence once input to the paper-sheet forming device matches the document based on the document. That is, a sentence mixed with kanji and kana, which has been input to the paper composition device once, is read out by the voice synthesizing device. .

FIG. 7 shows an example of such a speech synthesizer. In the figure, reference numeral 11 denotes a sentence analysis unit that analyzes a sentence and identifies words by referring to a dictionary 12 when a character string that is a mixture of kanji and kana is input, and decomposes the sentence into a phonological part and a prosodic part. A prosody analyzer that analyzes the prosody and generates a pitch command that specifies the prosody, a pitch signal generator that generates a pitch pattern based on the pitch command, and an actual pronunciation based on the above phoneme. A phonetic symbol generation unit for generating a specified phonetic symbol, a speech unit combining unit for taking out a voice unit from a voice unit memory 17 storing a voice unit from the phonetic symbols and combining them to generate an actual phoneme signal, 18 A voice synthesis unit that synthesizes the pitch pattern and the phoneme signal and outputs an actual synthesized voice from the speaker 19 is shown.

Then, the prosody (pitch frequency) analyzed by the phoneme analysis unit in this example is displayed using a so-called Fujisaki model (F model) and synthesized. This Fujisaki model is
The pitch frequency is described separately for a phrase component for displaying the intonation of a sentence and an accent component for displaying the accent of a word, and these components are combined, and are represented by the following formula. Here, the pitch frequency at time t, F ₀ (t), is Is displayed. Here, G _pi = α _i ² t exp (−α _i t)... When t ≧ 0, G _pi = 0... When t <0, G _pj = _Min [1- (1 + β _j t) exp (−β _j t) , Θ] ... t
When ≧ 0, G _pj = 0... When t <0, where I is the number of phrase commands J is the number of accent commands A _pi is the size of the i-th frame command A _aj is the size of the i-th accent command T _oi is the start time of the i-th phrase command T _1j is the start time of the j-th accent command T _2j is the end time of the j-th accent command α _i is the natural angular frequency β of the phrase control mechanism for the i-th phrase command _j is the natural angular frequency of the accent control mechanism for the j-th accent command. θ is the upper limit of the step response function of the accent control mechanism.

As shown in FIG. 4, the pitch frequency of the Fujisaki model is obtained by superimposing a plurality of phrase patterns and accent patterns in time series. Looking at one phrase pattern, as shown in FIG. 5, it has a waveform phrase pattern that rises once in response to a pulse-like phrase command and monotonically decreases from the maximum value. Looking at the accent pattern, as shown in FIG. 6, an accent pattern having a waveform pattern shifted in time with respect to a rectangular wave-like accent command, gradually rising, and gradually decreasing monotonously from the maximum value is obtained. To form. Then, as shown in FIG. 4, a plurality of phrase patterns and accent patterns specified by a plurality of phrase commands and accent commands (in this example, the phrase command is 3 and the accent command is 4) are overlapped. An actual pitch pattern is generated.

[Problems to be solved by the invention]

By the way, the actual pitch signal generation device calculates the value of the pitch frequency at each time by applying each constant to the above equation according to the input pitch command. (For example, at t = T in FIG. 4, it is not possible to calculate the result of the pitch pattern unless the calculation is completed for all of the two phrase patterns and the four accent patterns), and this calculation takes time. There is a problem that it takes time to generate the pitch frequency as a whole.

Therefore, an object of the present invention is to provide a pitch pattern generation device in a speech synthesis device that can generate a pitch pattern in as short a time as possible.

[Means for solving the problem]

In the present invention, a means for solving the above-mentioned problem is a method of inputting information relating to a sentence, converting the sentence into speech information to be read out, and outputting the speech information. When generating a pitch frequency, the pitch frequency generating unit includes a superimposing unit that superimposes and synthesizes a plurality of pitch patterns having a characteristic of monotonically decreasing from a peak value at each time in a time-series manner according to the characteristics of a sentence. In the apparatus, a tendency detecting means for generating a decreasing tendency signal when the pitch frequency in the pitch pattern is monotonically decreasing, and a threshold detecting means for generating a threshold signal when the pitch frequency in the pitch pattern is equal to or less than a predetermined threshold When receiving the decreasing trend signal and the threshold signal, the pitch frequency is monotonically decreasing and the pitch frequency is It is that you have a superposition stopping means stops generating the pitch pattern is less than a constant threshold value.

[Action]

According to the present invention, the pitch frequency in the pitch pattern is monotonic, and the generation of the pitch pattern of the pitch pattern in which the pitch frequency in the pitch pattern is equal to or less than a predetermined threshold is stopped. The number of terms does not increase and the calculation time is short.
Therefore, the pitch frequency can be generated in a short time as a whole.

〔Example〕

Hereinafter, an embodiment of a pitch frequency generation device in a speech synthesis device according to the present invention will be described with reference to the drawings.

2 and 3 show a first embodiment according to the present invention. In the present embodiment, as shown in FIG. 2, the pitch frequency generation device 20 includes a phrase pattern generation unit 21 that generates a phrase pattern that is one component of the pitch frequency, and an accent pattern that is another component of the pitch frequency. An accent pattern generation unit 22 to generate the signals is provided, and these signals are added by an adder 24. The adder 24 adds the phrase pattern and the accent pattern, and is the first term of the above equation.
F _min is added.

In this embodiment, the phrase pattern generator has a configuration as shown in FIG. In the figure, reference numeral 30 denotes a buffer memory for storing a phrase command from the prosody analysis unit, and N-stage storage which can sufficiently store the number of superimposed terms I of the term (second term) specifying the phrase in the above equation. Each storage area has a size A _pi as a phrase command.
Time _Toi and each unique frequency α _i are stored. Also 3
1 is a selector for successively extracting the phrase commands stored in the buffer memory 30 by the operation of the control unit 32, 33 is a phrase pattern calculation unit that calculates a phrase pattern at time t based on the above-mentioned phrase commands, and 34 is a phrase pattern calculation unit This is an accumulating unit that stores the results of the calculations of 33 superimposed at each time.

The above-mentioned phrase pattern calculator calculates a phrase pattern P (t) according to the following equation for the current time t sent from the control unit 32.

P (t) = A _pi G _pi (t−T _oi ) where G _pi (t) = α _i ² t exp (−α _i t)... T ≧ 0 G _pi (t) = 0. A threshold determination unit that operates as a detection unit, a threshold detection unit, and an overlay stop unit. This threshold judgment unit
Reference numeral 35 denotes a state in which the value of the phrase pattern at the time t calculated by the phrase pattern calculation unit 33 is received, and the calculated value corresponding to the Mth phrase command at that time is in a decreasing state (smaller than the previous calculated value). If it is smaller than a predetermined threshold, a command is sent to the command unit 32 so as not to execute the calculation of the phrase pattern corresponding to the M-th phrase command from the next time.

Therefore, according to the phrase pattern generation unit 22 according to this embodiment, the calculation of the phrase pattern based on each phrase command is such that the calculated value corresponding to the phrase command is in a reduced state,
Because and was prevented from being performed when predetermined smaller than the predetermined threshold has, for example, in FIG. 4, the calculation of a phrase pattern for No.1 phrase command, the value of the phrase pattern and the threshold value S ₁ or less decline Time T
_It is not necessary to execute the calculation after _s1. Similarly, the calculation of the phrase pattern after time T _s2 is not necessary for the second and subsequent phrase commands. Can be shortened.

Next, an accent pattern generation unit provided in the pitch frequency generation device according to the present embodiment will be described. The accent pattern generation unit in this embodiment uses the contents of the accent command stored in the buffer memory and the phrase pattern calculation unit as an accent pattern calculation unit, and the expression for calculating the accent pattern in the accent pattern calculation unit is the phrase pattern generation unit described above. Since only the phrase pattern calculator of FIG. 3 is replaced with the accent pattern calculator, the detailed description thereof is omitted.

In this accent pattern generation unit, the buffer memory stores the magnitude A _aj rising time T _1j , rising time T _2j , natural angular frequency β _j in the buffer memory as a rising accent command.
Is stored. The accent pattern calculating section calculates the accent pattern A (t) at time t according to the following equation.

A (t) = _Aaj { _Gaj (t- _T1j ) _-Gaj (t- _T2j )} where _Gaj (t) = _Min [1 (1 + _.beta.jt ) exp (-. _Beta.jt ), .theta.]
... _t≥0 G _aj (t) = 0 ... t <0.

According to this accent pattern generation unit, similarly to the above-described phrase pattern generation unit, when the value of the accent pattern is decreasing and is equal to or less than a predetermined threshold, the subsequent calculation is not performed. The calculation time for generating the accent pattern can be shortened.

Therefore, according to the pitch frequency generation unit according to the present embodiment,
Since the phrase pattern and the accent pattern forming the pitch frequency can be generated in a short time, the pitch frequency can be generated at high speed as a whole.

In the above-described embodiment, the pitch frequency generation device includes a phrase pattern generation unit that generates a phrase pattern and a pattern generation unit that generates a pitch pattern. However, the present invention does not necessarily need to include both generation units. It goes without saying that the present invention is also applied to a pitch frequency generator having one of the generators.

〔The invention's effect〕

As described above, according to the present invention, a pitch frequency generating device generates a decreasing tendency signal when a pitch pattern is monotonically decreasing, and a threshold signal when the pitch pattern is equal to or less than a predetermined threshold. When a threshold detecting means for generating a signal and a decreasing tendency signal and a threshold signal are received,
Since the superposition stopping means for stopping the generation of the pitch pattern is provided, the calculation is not executed for a portion which is small enough not to affect the generation of the pitch frequency. And the time required for generating the pitch frequency as a whole can be shortened.

[Brief description of the drawings]

FIG. 1 is a principle diagram of the present invention, FIG. 2 is a block diagram showing an embodiment of a pitch frequency generating device according to the present invention, and FIG. 3 is a diagram showing a phrase pattern generating unit of the pitch frequency generating device shown in FIG. FIG. 4 is a diagram showing a state of generating a pitch frequency, FIG. 5 is a diagram showing a state of a phrase pattern, FIG. 6 is a diagram showing a state of an accent pattern, FIG.
The figure shows the configuration of the speech synthesizer. DESCRIPTION OF SYMBOLS 1 ... Pitch frequency generation apparatus 2 ... Overlapping means 3 ... Trend detection means 4 ... Threshold detection means 5 ... Overlap stop means

Claims (1)

(57) [Claims] When generating a pitch frequency of a time-varying voice of an output voice of a voice synthesizing apparatus which inputs information relating to a text, converts the text into voice information to be read out, and outputs the voice information, the pitch depends on the characteristics of the text. A pitch frequency generation device having a superposition means for superimposing and synthesizing a plurality of pitch patterns having a characteristic of being monotonically decreasing from a peak value at each time, wherein the pitch frequency in the pitch pattern is monotonically decreasing. A tendency detecting means for generating a decreasing tendency signal when the tendency is detected, a threshold detecting means for generating a threshold signal when the pitch frequency in the pitch pattern is equal to or less than a predetermined threshold, and when receiving the decreasing tendency signal and the threshold signal The pitch frequency is monotonically decreasing, and the pitch frequency is equal to or less than a predetermined threshold. Superimposing stop formed pitch frequency generator in the speech synthesis apparatus and a stop means.