JPH07129196A - Sound waveform segmenting device, sound waveform shaping device, and sound synthesizing device - Google Patents

Abstract

PURPOSE:To provide a voice synthesizing device which can reduce labor of segmenting the pitch waveform in voice synthesization and can suppress noise generation at the termination of the segmented pitch waveform. CONSTITUTION:Each zero-cross point immediately before the voice waveform goes beyond the envelope which is prepared by a pitch waveform envelope preparing device 13, and the next zero-cross point is sensed within a certain time before and after the pitch period presumed by a pitch period presuming device 12, and the voice waveform is cut out by a pitch segmenting device 15 while this zero-cross interval is made one pitch of voice waveform. In accordance with any pitch period, a necessary window is set by a waveform forming device 16, and pitch waveforms formed are connected with each other by a waveform connecting device 17, and a resultant synthetic sound is emitted from an output part 18.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a voice slicing device, a voice waveform shaping device and a voice synthesizing device which are used for automatic guide broadcasting.

[0002]

2. Description of the Related Art In recent years, in an automatic guide broadcasting apparatus, when changing the sound to be transmitted, the same announcer newly makes a new recording, which requires considerable labor and cost, so that labor and cost reduction are strongly required. It has been demanded. For this reason, when changing only the pitch cycle without changing the character arrangement of the voice, such as when changing the accent, one pitch waveform of the voice is cut out, and this is cut down on the time axis to connect or reverse. The accent was changed by adding silence after the 1-pitch waveform and then connecting the pitch waveforms together.

A conventional speech synthesizer will be described below. FIG. 4 shows the configuration of a conventional speech synthesizer. In FIG. 4, 41 is a speech analyzer for linear prediction analysis of input speech, 42 is a pitch approximate period estimator for estimating an approximate pitch period by a modified autocorrelation method, 43 is a 1-pitch waveform clipper, and 44 is a pitch. , 45 is a pitch expander, 46 is a pitch connector, and 47 is a composite waveform output section.

The operation of the speech synthesizer configured as above will be described below. First, the digitized voice signal is analyzed by the voice analyzer 41,
Based on the result, the pitch approximate period estimator 42 estimates the approximate pitch period. Next, referring to the obtained approximate pitch period, the operator cuts out one pitch waveform corresponding to the approximate pitch period from the voice waveform by visual inspection using the cutout device 43. When increasing the pitch, the shortener 44 cuts off the latter half of one pitch waveform so as to match the pitch cycle interval, and connects the pitch waveforms from the connector 46. On the other hand, when the pitch is to be lowered, 1 is cut out by the expander 45 so as to have the pitch cycle interval.
After adding silence to the latter half of the pitch waveform, the pitch waveforms are connected from the connector 46. In the conventional speech synthesizer, the pitch cycle is changed by the above operation.

[0005]

However, the above-mentioned conventional speech synthesizer has a problem that the time and effort required for speech synthesis are not sufficiently reduced because it requires skill in cutting out a pitch waveform. Further, since the 1-pitch waveform is cut out by the rectangular window, there is a problem that an abnormal sound may be generated at the end of the cut-out 1-pitch waveform.

The present invention solves the above-mentioned conventional problems, and a speech waveform slicing device for automating pitch waveform slicing, a speech waveform slicing device for shaping a sliced one pitch waveform, and a pitch waveform slicing automation, It is also an object of the present invention to provide an excellent voice synthesizer capable of preventing the generation of abnormal sound.

[0007]

In order to achieve the above-mentioned object, a speech waveform slicing device of the present invention comprises means for creating an envelope corresponding to a pitch waveform of speech, and speech immediately before the speech waveform exceeds this envelope. And a detection means for detecting a zero-cross point of the waveform, and a voice waveform is output at intervals of each zero-cross point detected in this way.

Further, it is provided with an estimating means for estimating an approximate pitch period of the voice, and is detected within a certain time before and after the estimated approximate pitch cycle from the time of the first zero-cross point detection detected by the zero-cross point detecting means. The interval between the second zero-cross point and the zero-cross point is output as a 1-pitch voice waveform.

Further, there are provided an estimating means for estimating the pitch approximate period of the voice and a comparing means for comparing the interval between the respective zero cross points of the voice waveform detected by the detecting means for detecting the zero cross points with the pitch approximate period. When the comparison result is less than a certain value, this is output as a one-pitch waveform.

On the other hand, the voice waveform shaping apparatus of the present invention comprises output means for outputting a voice waveform and window creating means for producing a window for shaping a voice waveform according to an arbitrary pitch period,
The voice waveform output from the output unit is shaped by the window created by the window creating unit.

Further, the speech synthesizer of the present invention comprises an estimating means for estimating a pitch pitch cycle of speech, a means for creating an envelope corresponding to the pitch waveform of the speech, and a speech waveform immediately before the speech waveform exceeds this envelope. Of the zero crossing point, a window creating means for creating a window for shaping a voice waveform according to an arbitrary pitch cycle, and a connecting means for connecting the pitch waveforms to each other. First
The zero-cross point interval with the second zero-cross point detected within a fixed time before and after the estimated pitch approximate period from the time of zero-cross point detection is output as a 1-pitch voice waveform,
The output voice waveform is shaped by the window created by the window creating means, and the shaped pitch waveforms are connected to each other by the connecting means to synthesize the voice in an arbitrary pitch cycle.

Further, an estimating means for estimating the approximate pitch of the voice, a means for creating an envelope corresponding to the pitch waveform of the voice, and a detecting means for detecting a zero cross point of the voice waveform immediately before the voice waveform exceeds the envelope. And comparing means for comparing the interval between the respective zero-cross points of the voice waveform detected by the detecting means with the pitch approximate period estimated by the estimating means,
A window creating means for creating a window for shaping a voice waveform according to an arbitrary pitch period, and a connecting means for connecting the pitch waveforms to each other are provided, and when the comparison result in the comparing means is less than a certain value, the detecting means At each interval of each detected zero-cross point, the output means outputs a voice as a 1-pitch voice waveform, and the output voice waveform is shaped by the window created by the window creating means. ,
Connect the molded pitch waveforms by connecting means,
The speech synthesis is performed at an arbitrary pitch cycle.

[0013]

With this configuration, a zero-cross point for each pitch waveform of the voice waveform can be detected, and the one-pitch waveform can be automatically cut out based on the detected zero-cross point, and the pitch waveform can be terminated at any pitch period. Can be set to 0 smoothly.

[0014]

DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below with reference to the drawings.

In FIG. 1, reference numeral 11 is a speech analyzer, and 12 is a pitch approximate period estimator, which has the same functions as the speech analyzer 41 and the pitch approximate period estimator 42 in the prior art.
Reference numeral 13 denotes a pitch waveform envelope generator including a counter 21, a memory 22, an envelope generator 23, and a switch G2.
Reference numeral 4 is a waveform envelope / speech comparator, 15 is a pitch clipper that cuts out one pitch of voice, 16 is a waveform shaper for the obtained 1 pitch waveform, 17 is a connector that connects the formed pitch waveforms, 18 Is the output section of the composite waveform.

1, 2 and 3 (a), 3
The operation will be described with reference to FIG. First, a method of automatically cutting out a one-pitch waveform will be described. As in the conventional case, a digitized voice signal is input, and the voice analyzer 11 and the pitch approximate period estimator 12 estimate the pitch approximate period f. The pitch approximate period f thus estimated is updated every 5 pitches. On the other hand, the pitch waveform envelope generator 13 detects the maximum amplitude and creates the envelope of the pitch waveform by the following method. FIG. 2 shows the comparator 1.
4 input signal and switches G1 and G2 ON
-OFF timing is shown.

In FIG. 2, when the power is turned on at time T1, the maximum amplitude has an initial value of 0, and the maximum amplitude and the envelope input to the comparator 14 remain 0 until the sound comes in at time T2. Is. When there is a voice input exceeding the envelope (= 0) at time T2, a signal from the comparator 14 that temporarily turns on the switch G1 (several μs) and a signal that resets the counter 21 in the pitch waveform envelope generator 13. Is output. When the counter 21 is reset, the initial value of the counter 21 becomes zero. The switch G2 is also turned on. With the above operation, the maximum value of the amplitude of the audio signal appearing after time T2 is written in the memory 22, and at the same time, the amplitude value is input to the envelope creating unit 23 to create the envelope.
The creation of the envelope will be described later.

Further, in the counter 21, the counter 2
At the same time that 1 is reset, the pitch approximate period estimator 12
The pitch approximate period f estimated at is loaded, and counting is started. On the other hand, in order to eliminate erroneous detection when the voice signal temporarily exceeds the envelope due to unnecessary input such as noise, the comparator 14 compares the envelope and the voice signal only for a certain period before and after the approximate pitch period f. There is a need to. Therefore, a fixed value is set in advance to, for example, 5 ms, and the value of the counter 21 is kept in the counter 2 while the pitch approximate period f ± 5 ms
1 outputs a signal for turning on the switch G2.

When the voice signal exceeds the envelope (time T
3), the switch G1 is temporarily turned on, and the audio signal is input to the memory 22. At the same time, the maximum value of the amplitude is written in the memory 22, and at the same time, the maximum amplitude value is input to the envelope creating section 23. Then, a new envelope is created based on this, and the counter 21 is reset. On the contrary, the value of the counter 21 is a constant value (5
voice that exceeds the envelope while it has not reached
4 is input (time T4) and the switch G1 is turned on, the switch G2 is in the off state as described above, and the envelope continues to be attenuated as it is. False detections such as when the signal exceeds the envelope can be eliminated.

As described above, the amplitude of the speech waveform exceeding the envelope is detected as the maximum amplitude for the first time within a certain time before and after the approximate period, and the envelope is created based on this maximum amplitude.

The waveform envelope E (t) at time t starting from the time point of maximum amplitude of the voice waveform is calculated by the following equation.

E (t) = Aexp (-πBt) where A is the maximum amplitude value and B is the average bandwidth of the first formant, which is about 50 Hz.

Reference numerals 31, 32, and 33 in FIG. 3A are maximum amplitudes detected in this way, and 34, 35, and 36.
Is the envelope of the pitch waveform created from each of the maximum amplitudes 31, 32, and 33 as a starting point based on the above equation.

When the voice signal exceeds the envelope of the pitch waveform created in this way within a certain period before and after the approximate pitch period f, a cutout signal is output to the pitch cutout device 15. The pitch cutout device 15 receives the input voice signal for 10 ms.
Can be buffered to some extent. With respect to the buffered input signal, when the cutout signal is input, it goes back in time from the signal input time, and the time when the audio signal becomes a negative value for the first time is detected (37, 3 in FIG. 3A).
8, 39). This negative value time point is called a zero cross point. When each zero-cross point is detected in this way, the interval (3
7-38, 38-39) as one pitch, and the pitch waveform is cut out by the pitch cutout device 15 from the input voice, the pitch waveform is automatically cut out.

In order to eliminate erroneous detection, the method of turning on the switch G2 to create the envelope only within the time of the pitch approximate period f ± 5 ms as described above is not adopted, but the detected zero-cross point is detected. Interval (37-38, 38
-39) is detected, and this time interval is directly compared with the pitch approximate period estimated by the approximate period estimator 12, and if the comparison result is a fixed value or less, for example, a difference within about ± 5 ms, this is 1 pitch. It is also possible to automatically cut out the pitch waveform by handling it as a zero-cross point.

Next, the shaping of the voice waveform cut out in this way will be described. The waveform shaping unit 16 applies a window to the 1-pitch voice waveform cut out by the pitch cutout unit 15 so that the end of the voice waveform becomes zero. Such a window is created by the following equation.

[0027]

[Equation 1]

Further, T is as follows depending on whether the pitch is to be raised or lowered.

That is, when trying to raise the pitch,

[0030]

[Equation 2]

It becomes Also, when trying to lower the pitch,

[0032]

[Equation 3]

It becomes Figure 3 shows such a window.
An example of the waveform formed by this window is shown at 51 in (b) and at 52.

The end of the waveform thus shaped is smoothly 0, and these pitch waveforms are connected to each other by the connector 17 and output from the output section 18, so that an arbitrary pitch period is obtained. On the other hand, it is possible to synthesize a voice in which no abnormal sound is generated at the end of the pitch waveform.

As described above, according to the present embodiment, since the pitch waveform can be automatically cut out, the labor of voice synthesis can be reduced, and the pitch cycle can be changed by using the waveform-shaped pitch waveform. It is possible to prevent abnormal sounds from being generated at the ends of the cut pitch waveform.

[0036]

As described above, the present invention utilizes a detected zero-cross point and compares a pitch waveform with an approximate pitch period to automatically cut out a pitch waveform, and a pitch waveform cutting device for this speech. The 1-pitch waveform output from is shaped by waveform shaping according to the pitch cycle to be created, and by connecting the shaped pitch waveforms to each other to provide a device for synthesizing speech of an arbitrary pitch cycle, it is cut out. It is also possible to realize an excellent voice synthesizer capable of preventing the generation of abnormal sounds at the ends of the pitch waveform.

[Brief description of drawings]

FIG. 1 is a device configuration diagram of a speech synthesizer according to an embodiment of the present invention.

FIG. 2 is an input signal to a comparator 14 and a switch G1.
Waveform diagram showing ON-OFF timing of G2 and G2

FIG. 3 (a) Voice waveform diagram (b) Shaped voice waveform diagram

FIG. 4 is a device configuration diagram of a conventional speech synthesizer.

[Explanation of symbols]

 11 Speech Analyzer 12 Pitch Approximate Period Estimator 13 Pitch Waveform Envelope Creator 14 Comparator 15 Pitch Cutout Device 16 Waveform Shaper 17 Connector 18 Output Unit 21 Counter 22 Memory 23 Envelope Creating Unit

Claims (6)

[Claims] 1. An envelope creating means for creating an envelope corresponding to a voice waveform, a detecting means for detecting a zero-cross point of the voice waveform immediately before the voice waveform exceeds the envelope, and a zero-cross point thus detected. An audio waveform slicing device, comprising: an output unit that outputs an interval as an audio waveform of one pitch. 2. An estimation means for estimating a pitch approximate cycle of a voice is provided, and the speech is detected within a certain time before and after the estimated pitch approximate cycle from the time of the first zero-cross point detection detected by the zero-cross point detecting means. The voice waveform cutting device according to claim 1, wherein the interval between the zero cross points and the second zero cross point is output as a one-pitch voice waveform. 3. Estimating means for estimating an approximate pitch period of a voice, and comparing means for comparing an interval between respective zero-cross points of a voice waveform detected by a zero-cross point detecting means with an approximate pitch cycle estimated by the estimating means. The audio waveform cutting device according to claim 1, wherein the zero-cross point interval is output as a one-pitch audio waveform when the comparison result is less than a certain value. 4. An output means for outputting a voice waveform, and a window creating means for creating a window for shaping the voice waveform according to an arbitrary pitch period, the voice waveform output from the output means being created as the window. An audio waveform shaping device that shapes an audio waveform using a window created by the means. 5. An estimating means for estimating an approximate pitch of a voice, an envelope producing means for producing an envelope corresponding to a voice waveform, and a zero-cross point of the voice waveform immediately before the voice waveform exceeds the envelope. Detecting means, window creating means for creating a window for shaping a voice waveform according to an arbitrary pitch cycle, and connecting means for connecting pitch waveforms to each other, and first zero-cross point detection detected by the detecting means. From time to time, the zero-cross point interval with the second zero-cross point detected within a certain time before and after the estimated pitch approximate period is set to 1
A voice output as a pitch voice waveform, the voice waveform is shaped by the window created by the window creating means, and the formed pitch waveforms are connected by the connecting means to synthesize a voice of an arbitrary pitch cycle. Synthesizer. 6. An estimating means for estimating an approximate pitch of a voice, an envelope producing means for producing an envelope corresponding to a voice waveform, and a zero-cross point of the voice waveform immediately before the voice waveform exceeds the envelope. Detecting means, comparing means for comparing the interval between the respective zero-cross points of the voice waveform detected by the detecting means with the pitch approximate period estimated by the estimating means, and a window for shaping the voice waveform according to an arbitrary pitch period. A window creating means for creating, and connecting means for connecting the pitch waveforms, when the comparison result in the comparing means is less than a certain value, for each interval of each zero-cross point detected by the detecting means, Using this as a 1-pitch voice waveform, the voice is cut out by the cutting-out means, and the cut-out voice waveform is shaped by the window created by the window creating means. A voice synthesizing device for synthesizing voices having an arbitrary pitch period by connecting the formed pitch waveforms by the connecting means.