JP2005062442A - Waveform connection apparatus, waveform connection method and program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a waveform connecting apparatus etc., for generating a synthesized voice which has small noise and is natural. <P>SOLUTION: A cepstrum analysis part 1 specifies the frequency of a pitch component of a speech that a waveform signal supplied from an input end IN represents. A variable BPF 2 generates a pitch signal by filtering the waveform signal with a characteristic of a band-pass filter having as its center frequency that frequency specified by the cepstrum analysis part 1. A zero-cross detection part 3 specifies the timing where the pitch signal raises to cross zero and a pitch mark addition part 5 adds a pitch mark indicating the timing to the waveform signal. A waveform connection processing part 6 sections a plurality of waveform signals given pitch marks nearby the timing where the pitch signal crosses zero so that a precedent waveform signal and a following waveform signal have nearly equal momentary values at the end point of the former waveform signal and at the start point of the latter waveform signal, and then connects them to each other to generate a waveform signal representing a synthesized speech. <P>COPYRIGHT: (C)2005,JPO&NCIPI

Description

  The present invention relates to a waveform connecting device, a waveform connecting method, and a program.

  In recent years, speech synthesized by speech synthesis technology has been widely used. Specifically, it is used in many scenes such as text-to-speech software, telephone number guidance, stock guidance, travel guidance, store guidance, traffic information, and the like.

The speech synthesis methods are roughly classified into a rule synthesis method and a waveform editing method (corpus method).
The rule synthesis method is a method of generating speech by performing morphological analysis on a text to be synthesized and performing phonological processing on the text based on the analysis result. In the rule synthesis method, there are few restrictions on the content of text used for speech synthesis, and texts with various contents can be used for speech synthesis. However, in the rule synthesis method, the quality of the output voice is inferior compared to the waveform editing method.

  On the other hand, the corpus method is a method of obtaining a target voice by recording voices actually spoken by humans and connecting constituent parts obtained by subdividing the recorded voices. The waveform editing method is more advantageous than the rule synthesis method in terms of voice quality, and a voice with a real voice can be obtained.

However, the waveform of the connection part that connects the constituent parts of the recorded voice is generally discontinuous when they are simply connected, and this becomes a source of noise, leading to a decrease in the quality of the synthesized voice. Therefore, as a technique for reducing this noise caused by the discontinuity of the connected portion, when connecting two waveforms, the portion of the preceding waveform that is as close as possible to the rear end and the waveform that is behind are as much as possible. A method is conceived in which a point at which the instantaneous value and the gradient of the tangential line substantially match each other is found one by one from a portion close to the front end, and these points are connected to each other (see, for example, Patent Document 1). . Further, for example, as shown in FIG. 5, a waveform that comes as close to the rear end as possible (a waveform shown as “S1” in FIG. 5) and a waveform that comes after (a waveform shown as “S2” in FIG. 5) A method of finding and connecting one point at a time when the instantaneous value becomes 0 (zero crossing) from a portion as close to the front end as possible is also conceivable.
JP 2003-15681 A

  However, in general, a speech waveform has periodicity at a time interval called a pitch period, and in order to obtain a natural synthesized speech, an end point of a pitch period of a preceding unit speech and a pitch period of a subsequent unit speech are obtained. Select the appropriate start point (the point that will be the end point or start point when a person actually utters these unit sounds), and connect the end point and start point to connect the audio. It is necessary to match.

  On the other hand, for example, in the above-described method of connecting points that cross zero, as shown in FIG. 5, the end generated by the break is not an appropriate point as the start point or end point of the pitch period, and the resultant synthesized speech (Sound having a waveform shown as “S0” in FIG. 5) is often unnatural (in FIG. 5, the section shown as “D1” is the pitch period of the waveform S1, and “D2”) The section shown is the pitch period of the waveform S2.) A similar phenomenon also occurs in the above-described method of connecting points at which the instantaneous value and the gradient of the tangential line substantially coincide with each other.

  The present invention has been made in view of the above circumstances, and an object thereof is to provide a waveform connecting device, a waveform connecting method, and a program for generating a natural synthesized speech with little noise.

In order to achieve the above object, a waveform connecting device according to a first aspect of the present invention is provided by:
A first waveform signal representing a voice waveform and a second waveform signal to be subsequent to the first waveform signal are acquired, and for each acquired waveform signal, the pitch component of the voice represented by the waveform signal is obtained. Pitch component frequency specifying means for specifying the frequency;
For each of the first and second waveform signals, the self passband characteristic is changed so that the passband characteristic is cut off except for the components in the vicinity of the frequency of the pitch component specified by the pitch component frequency specifying means. Variable filter means for filtering the waveform signal and extracting a pitch component;
Phase detection means for identifying the timing at which the pitch component extracted by the variable filter means has a predetermined phase for each of the first and second waveform signals;
By dividing the first and second waveform signals in the vicinity of the portion corresponding to the timing detected by the phase detection means, the end point of the first waveform signal and the start point of the second waveform signal are obtained. Connecting means for determining and connecting the first and second waveform signals to each other at the end point of the first waveform signal and the start point of the second waveform signal;
The connecting means divides the first and second waveform signals so that the end point of the first waveform signal and the start point of the second waveform signal are substantially equal to each other. ,
It is characterized by that.

Moreover, the waveform connecting device according to the second aspect of the present invention is:
A first waveform signal representing a voice waveform and a second waveform signal to be subsequent to the first waveform signal are acquired, and for each acquired waveform signal, the pitch component of the voice represented by the waveform signal is obtained. Pitch component frequency specifying means for specifying the frequency;
For each of the first and second waveform signals, the self-pass so as to have a passband characteristic of a bandpass filter having a center frequency substantially coincident with the frequency of the pitch component specified by the pitch component frequency specifying means. Variable filter means for changing a band characteristic and filtering the waveform signal to extract a pitch component;
Phase detection means for identifying the timing at which the pitch component extracted by the variable filter means has a predetermined phase for each of the first and second waveform signals;
The first waveform signal and the second waveform signal are delimited at locations corresponding to the timing detected by the phase detection means, thereby determining the end point of the first waveform signal and the start point of the second waveform signal. Connection means for connecting the first and second waveform signals to each other at the end point of the first waveform signal and the start point of the second waveform signal;
It is characterized by that.

  The pitch component frequency specifying means may specify, for each of the first and second waveform signals, a frequency at which a cepstrum of the waveform signal takes a maximum value as a frequency of a pitch component of the waveform signal. Good.

  The phase detection means specifies the timing at which the value of the pitch component extracted by the variable filter means zero-crosses with a change in sign in a predetermined direction for each of the first and second waveform signals. There may be.

Moreover, the waveform connection method according to the third aspect of the present invention is as follows:
A first waveform signal representing a voice waveform and a second waveform signal to be subsequent to the first waveform signal are acquired, and for each acquired waveform signal, the pitch component of the voice represented by the waveform signal is obtained. A pitch component frequency specifying step for specifying a frequency;
For each of the first and second waveform signals, variable filtering that extracts a pitch component by filtering with passband characteristics such that components other than those in the vicinity of the frequency of the pitch component specified in the pitch component frequency specifying step are blocked. Steps,
For each of the first and second waveform signals, a phase detection step for specifying a timing at which the pitch component extracted in the variable filtering step becomes a predetermined value phase;
By dividing the first and second waveform signals in the vicinity of the portion corresponding to the timing detected in the phase detection step, the end point of the first waveform signal and the start point of the second waveform signal are obtained. Determining and connecting the first and second waveform signals to each other at an end point of the first waveform signal and a start point of the second waveform signal,
In the connecting step, the first and second waveform signals are separated such that the end point of the first waveform signal and the start point of the second waveform signal are substantially equal to each other,
It is characterized by that.

A program according to the fourth aspect of the present invention is
Computer
A first waveform signal representing a voice waveform and a second waveform signal to be subsequent to the first waveform signal are acquired, and for each acquired waveform signal, the pitch component of the voice represented by the waveform signal is obtained. Pitch component frequency specifying means for specifying the frequency;
For each of the first and second waveform signals, the self passband characteristic is changed so that the passband characteristic is cut off except for the components in the vicinity of the frequency of the pitch component specified by the pitch component frequency specifying means. Variable filter means for filtering the waveform signal and extracting a pitch component;
Phase detection means for identifying the timing at which the pitch component extracted by the variable filter means has a predetermined phase for each of the first and second waveform signals;
By dividing the first and second waveform signals in the vicinity of the portion corresponding to the timing detected by the phase detection means, the end point of the first waveform signal and the start point of the second waveform signal are obtained. A program for determining and functioning the first and second waveform signals as connection means for connecting the end point of the first waveform signal and the start point of the second waveform signal to each other. ,
The connecting means divides the first and second waveform signals so that the end point of the first waveform signal and the start point of the second waveform signal are substantially equal to each other. ,
It is characterized by that.

  As described above, according to the present invention, a waveform connecting apparatus, a waveform connecting method, and a program for generating a synthesized speech with little noise or natural are realized.

In the following, an embodiment of the present invention will be described with reference to the drawings, taking a speech synthesizer as an example.
FIG. 1 is a block diagram showing the configuration of this speech synthesizer. As shown, the speech synthesizer includes an input terminal IN, a cepstrum analysis unit 1, a variable BPF (bandpass filter) 2, a delay unit 4, a zero cross detection unit 3, a pitch mark addition unit 5, The waveform connection processing unit 6 and an output terminal OUT are included.
When a waveform signal obtained by subdividing a prerecorded voice into individual unit voices (for example, a voice representing one word or a short sentence) is supplied from the input terminal IN, A synthesized voice signal obtained by synthesizing the supplied waveform signal is output from the output terminal OUT.

The cepstrum analysis unit 1, the variable BPF 2, the zero cross detection unit 3, the delay unit 4, the pitch mark addition unit 5 and the waveform connection processing unit 6 are respectively a processor such as a DSP (Digital Signal Processor) and a CPU (Central Processing Unit), It is composed of a RAM (Random Access Memory), a memory such as a hard disk device, or a dedicated circuit. Each of the delay units 4 may be configured by a delay circuit such as a shift register or a delay line.
The same processor may perform a part or all of the functions of the cepstrum analysis unit 1, the variable BPF 2, the zero cross detection unit 3, the delay unit 4, the pitch mark addition unit 5, and the waveform connection processing unit 6.

  Each time a waveform signal is supplied from the input terminal IN, the cepstrum analysis unit 1 performs cepstrum analysis on the waveform signal to specify the frequency of the pitch component of the voice represented by the waveform signal. (Note that the data format of the waveform signal is arbitrary, for example, the PCM (Pulse Code Modulation) format may be adopted.)

As specific processing of cepstrum analysis, for example, when the cepstrum analysis unit 1 is supplied with a waveform signal from the input terminal IN, first, the intensity of the waveform signal is changed to a value substantially equal to the logarithm of the original value. And convert. (The base of the logarithm is arbitrary, and may be a common logarithm, for example.)
Next, the cepstrum analysis unit 1 uses a fast Fourier transform technique (or other arbitrary data that generates a result of Fourier transform of discrete variables) on the spectrum of the waveform signal (ie, the cepstrum) whose value has been converted. This method is used. Then, the minimum value among the frequencies giving the maximum value of the cepstrum is specified as the frequency (basic frequency) of the pitch component. Then, the cepstrum analysis unit 1 controls the pass band characteristic of the variable BPF 2 so that the frequency of the identified pitch component is the center frequency of the variable BPF 2 (the center frequency of the pass band).

The variable BPF 2 performs a function of a FIR (Finite Impulse Response) type or IIR (Infinite Impulse Response) type band pass filter whose center frequency is variable.
Specifically, the variable BPF 2 sets its center frequency to a value according to the control of the cepstrum analysis unit 1. On the other hand, the variable BPF 2 receives a waveform signal from the input terminal IN, filters the same waveform signal that the cepstrum analysis unit 1 used to determine the center frequency, and outputs the filtered waveform signal (pitch signal), This is supplied to the zero cross detection unit 3.

  Note that the pitch signal may be composed of digital data having a sampling interval substantially the same as the sampling interval of the waveform signal, for example. Further, it is desirable that the bandwidth of the variable BPF 2 always falls within twice the frequency of the pitch component specified by the cepstrum analysis unit 1.

  The zero-cross detection unit 3 identifies the timing at which the instantaneous value of the pitch signal supplied from the variable BPF 2 changes from negative to positive (timing to rise and zero-cross), generates a signal (zero-cross signal) representing the identified timing, It supplies to the mark addition part 5. In other words, the timing at which the pitch signal rises and zero-crosses is the timing at which the phase of the sine wave becomes 0 when the pitch signal is regarded as a sine wave.

  Each time the waveform signal is supplied from the input terminal IN, the delay unit 4 delays the waveform signal for a predetermined time and supplies the waveform signal to the pitch mark adding unit 5. The time length that the delay unit 4 delays the waveform signal is the timing specified by the zero cross signal supplied from the zero cross detection unit 3 to the pitch mark adding unit 5 (that is, the pitch signal rises). It is assumed that a portion corresponding to the timing of zero crossing is selected to a value that is supplied to the pitch mark adding unit 5 earlier than this zero cross signal (or simultaneously with this zero cross signal).

Each time the pitch mark adding unit 5 is supplied with the waveform signal from the delay unit 4 and the zero cross signal is supplied from the zero cross detection unit 3, the pitch mark adding unit 5 sets the boundary of the section corresponding to one pitch of the waveform signal to the supplied waveform signal. Pitch marks shown are added and sequentially supplied to the waveform connection processing unit 6.
Specifically, the position indicated by the pitch mark added by the pitch mark adding unit 5 is, as schematically shown in FIG. 2, the timing specified by the zero cross signal supplied from the zero cross detecting unit 3 (that is, the pitch signal). Is a location (a location indicated as “PM” in FIG. 2) corresponding to the timing at which zero rises and zero crossing). FIG. 2 is a diagram exemplifying a case where pitch marks are added to waveform signals representing waveforms illustrated as “S1” and “S2”.

  In speech actually uttered by a person, in general, the pitch component of the preceding unit speech often crosses zero at the boundary between two adjacent unit speeches, and the pitch component of the subsequent unit speech at this boundary. Are often zero-crossed. For this reason, the pitch mark added to the waveform signal by the pitch mark adding unit 5 indicates a position that can be a boundary between unit sounds when the waveform signal is actually uttered by a person. In other words, the pitch mark indicates the position of an appropriate end point or start point of the pitch period.

  When the waveform signal with the pitch mark added thereto is sequentially supplied from the pitch mark adding unit 5, the waveform connection processing unit 6 temporarily stores these waveform signals. Then, these waveform signals are connected to each other, and the obtained waveform signal is output from the output terminal OUT as a synthesized speech waveform signal.

  However, for example, as shown in FIG. 3, the waveform connection processing unit 6 is as late as possible of the preceding waveform signal (a waveform signal representing a waveform shown as “S1” in the example shown in FIG. 3). The instantaneous value and the tangent gradient from the portion close to the pitch mark at the end and the portion of the subsequent waveform signal (the waveform signal representing the waveform shown as “S2” in the example shown in FIG. 3) as close to the pitch mark as possible. Are searched for one point at a time, and the waveform signals are connected so that these points (that is, the end point of the preceding waveform signal and the start point of the subsequent waveform signal) overlap each other, and the synthesized speech (In the example shown in FIG. 3, a waveform signal representing a waveform shown as “S3”) is generated. The part after the end point of the preceding waveform signal and the part before the start point of the subsequent waveform signal are discarded.

  Note that the waveform connection processing unit 6 can specify any order in which the waveform signals to which the pitch marks are added are connected in any order. For example, the waveform connection processing unit 6 has the pitch marks added thereto. The waveform signals may be connected to each other in the order in which they are supplied to them. Alternatively, data specifying the connection order may be acquired from the outside in accordance with a user operation or the like, and the connections may be made in the order indicated by the data.

  By performing the operations described above, this speech synthesizer can generate a waveform signal at the timing when the pitch signal rises and zero-crosses (that is, at a position that can become the boundary of unit speech when actually uttered by a person). After the separation, the separated waveform signals are connected to each other so as to be connected smoothly. For this reason, the waveform signal of the synthesized speech represents natural speech.

The configuration of the speech synthesizer is not limited to that described above.
For example, instead of adding a pitch mark to the waveform signal, the pitch mark adding unit 5 sends data indicating the timing specified by the zero cross signal supplied to the pitch mark adding unit 5 by the zero cross detecting unit 3 to the waveform connection processing unit 6. You may make it supply. In this case, the waveform connection processing unit 6 may perform the above-described processing assuming that the position indicated by this data is the position to which the pitch mark is added. Further, the zero cross detection unit 3 may supply the zero cross signal to the waveform connection processing unit 6. In this case, the waveform connection processing unit 6 may perform the above-described processing on the waveform signal, assuming that the position corresponding to the timing specified by the zero cross signal is the position where the pitch mark is added.

  Also, the zero cross detection unit 3 is the timing at which the instantaneous value of the pitch signal supplied from the variable BPF 2 changes from positive to negative (the timing of falling and zero crossing. In other words, the sine when the pitch signal is regarded as a sine wave. The timing at which the phase of the wave becomes π [radian] is specified, and a signal representing this timing may be generated as a zero-cross signal.

  In addition, when a pitch mark is added to a unit sound whose pitch component has a predetermined phase (not necessarily 0) at the boundary with another unit sound when a person actually utters, the zero cross detection unit 3 However, the timing at which the pitch signal substantially reaches the phase may be specified, and a signal representing the specified timing may be supplied to the pitch mark adding unit 5 as a signal in place of the zero cross signal.

  However, the process of specifying the presence or absence of the zero crossing of the pitch signal and the change of the sign before and after the zero crossing can be performed more easily with a simple configuration than the process of specifying an arbitrary phase of the pitch signal. For this reason, if all the unit sounds handled by this speech synthesizer can be regarded as “the phase of the pitch component becomes almost zero at the boundary with other unit sounds when a person actually utters”, Thus, it is possible to simplify the configuration of the zero-cross detection unit 3 by handling it as such.

  In addition, the waveform connection processing unit 6 does not necessarily have to have the tangent gradients substantially coincide with each other when searching for two points to be connected to each other. One point at which the instantaneous values substantially match each other may be searched for from the portion close to and the portion of the subsequent waveform signal as close as possible to the front end pitch mark.

  Further, the cepstrum analysis unit 1 does not necessarily need to make the center frequency of the variable BPF 2 coincide with the frequency of the pitch component, and has a passband characteristic that blocks other components in the vicinity of the frequency of the pitch component. What is necessary is just to control the pass-band characteristic of variable BPF2. However, the phase delay when a signal whose frequency substantially matches the center frequency of the variable BPF 2 passes through the variable BPF 2 is substantially zero. For this reason, it is desirable that the center frequency of the variable BPF 2 substantially matches the frequency of the pitch component.

  Further, since the phase delay when the signal whose frequency substantially matches the center frequency of the variable BPF 2 passes through the variable BPF 2 is substantially 0 as described above, the waveform signal in which the center frequency of the variable BPF 2 is supplied to the input terminal IN. If the pitch component substantially matches the frequency of the pitch component, the instantaneous value of the portion of the waveform signal to which the pitch mark is added can be expected to be almost zero. Therefore, when the center frequency of the variable BPF 2 substantially matches the frequency of the pitch component of the waveform signal supplied to the input terminal IN, the two waveform signals supplied to the waveform connection processing unit 6 are replaced with the pitch mark. What is necessary is just to mutually connect in the location to which "." Was added.

  Further, the delay unit 4 is not always necessary, and the pitch mark adding unit 5 may be directly supplied with the waveform signal from the input terminal IN. In this case, for example, the pitch mark adding unit 5 may temporarily store the waveform signal for a predetermined time or more. Alternatively, the portion corresponding to the timing specified by the zero cross signal in the waveform signal supplied from the input terminal IN to the pitch mark adding unit 5 is simultaneously with the zero cross signal (or later than the zero cross signal). The pitch mark adding unit 5 does not necessarily need to temporarily store the waveform signal.

  In general, an IIR filter has a feature that the delay of a signal passing therethrough does not increase even when the number of stages is increased. For this reason, when the variable BPF 2 is formed of an IIR type filter, the delay unit 4 can often be omitted.

  The speech synthesizer may include a plurality of units each including an input terminal IN, a cepstrum analysis unit 1, a variable BPF 2, a zero cross detection unit 3, a delay unit 4, and a pitch mark addition unit 5. A unit may generate a waveform signal with a pitch mark using a waveform signal supplied to its input terminal IN and supply the waveform signal to the waveform connection processing unit 6. In this case, the waveform connection processing unit 6 may generate a synthesized speech waveform signal using the waveform signal with pitch marks supplied from each unit.

  Further, the waveform signal supplied to the input terminal IN may represent a silent state. By combining the waveform signal representing the sound state and the waveform signal representing the silence state, a portion including the end of the signal representing the sound state (specifically, for example, the beginning and end of the voice, or the breathing portion) Etc.) can be avoided, and this part can be heard naturally.

In addition, this speech synthesizer has a recording medium on which a waveform signal is recorded instead of the input terminal IN (for example, a floppy (registered trademark) disk, a CD (Compact Disc), an MO (Magneto-Optical Disk), etc.) Includes a recording medium drive device (for example, a floppy (registered trademark) disk drive, a CD-ROM drive, an MO drive, etc.) that reads the waveform signal from the signal and supplies it to the delay unit 4, the variable BPF 2, and the cepstrum analysis unit 1. Also good.
In addition, the speech synthesizer may include a recording medium drive device that writes the waveform signal generated by the waveform connection processing unit 6 to the recording medium instead of the output terminal OUT.
The same recording medium drive device may perform both the function of reading the waveform signal from the recording medium and the function of writing the waveform signal generated by the waveform connection processing unit 6 to the recording medium.

Although the embodiment of the present invention has been described above, the waveform connecting apparatus according to the present invention can be realized using a normal computer system, not a dedicated system.
For example, a medium (CD, which stores a program for causing a personal computer to execute the operations of the above-described cepstrum analysis unit 1, variable BPF 2, zero cross detection unit 3, delay unit 4, pitch mark addition unit 5 and waveform connection processing unit 6) By installing the program from an MO, floppy (registered trademark) disk, or the like, a speech synthesizer that performs the above-described processing can be configured.

  The personal computer that executes this program performs, for example, the process shown in FIG. 4 as a process corresponding to the operation of the speech synthesizer of FIG. FIG. 4 is a flowchart showing processing executed by the personal computer.

  That is, when the personal computer obtains data representing the waveform signal of the above unit voice from the outside (step S101 in FIG. 4), the personal computer first performs cepstrum analysis on the data to obtain the data. The frequency of the pitch component of the voice represented by is specified (step S102).

  Next, the personal computer generates data representing the pitch signal by filtering the data acquired in step S101 with the characteristics of the bandpass filter having the frequency specified in step S102 as the center frequency (step S103). .

  Next, the personal computer specifies the timing at which the pitch signal represented by the data generated in step S103 rises and zero-crosses (step S104), and the pitch is data indicating the position on the waveform signal corresponding to the specified timing. The mark is added to the data acquired in step S101 (step S105).

  When the personal computer acquires a plurality of data representing the waveform signal of the unit sound in step S101, the personal computer performs the processing of steps S102 to S105 for each of these data, and the number of data acquired in step S101 as the data with the pitch mark. The same number is generated.

  The personal computer then creates and outputs data representing the waveform signal of the synthesized speech based on the plurality of data with the pitch mark added (step S106). The data generated and output in step S106 includes a portion as close as possible to the rearmost pitch mark in the preceding waveform signal among the two waveform signals represented by the plurality of data generated in the processing in steps S102 to S105, From the subsequent waveform signal, search the point where the instantaneous value and the tangential gradient are almost identical to each other from the portion as close as possible to the pitch mark at the front end, and connect the waveform signal so that these points overlap each other. It is assumed that the waveform signal of the synthesized speech obtained as a result of execution is represented.

  The program for causing the personal computer to perform the functions of the above-described speech synthesizer may be, for example, uploaded to a bulletin board (BBS) on a communication line and distributed via the communication line. The carrier wave may be modulated by the signal, the obtained modulated wave may be transmitted, and the apparatus that has received the modulated wave may demodulate the modulated wave to restore the program. The above-described processing can be executed by starting this program and executing it under the control of the OS in the same manner as other application programs.

  When the OS shares a part of the processing, or when the OS constitutes a part of one component of the present invention, a program excluding the part is stored in the recording medium. May be. Also in this case, in the present invention, it is assumed that the recording medium stores a program for executing each function or step executed by the computer.

It is a figure which shows the speech synthesizer which concerns on embodiment of this invention. It is a figure which represents typically the state by which the pitch mark was added to the waveform signal. It is a figure explaining a mode that a waveform signal is connected. It is a flowchart which shows the process which the personal computer which performs the function of the speech synthesizer concerning embodiment of this invention performs. It is a figure explaining a mode that a waveform signal is connected in an improper location.

Explanation of symbols

1 Cepstrum analysis unit 2 Variable BPF
3 Zero cross detection unit 4 Delay unit 5 Pitch mark adding unit 6 Waveform connection processing unit IN Input terminal OUT Output terminal

Claims (6)

A first waveform signal representing a voice waveform and a second waveform signal to be subsequent to the first waveform signal are acquired, and for each acquired waveform signal, the pitch component of the voice represented by the waveform signal is obtained. Pitch component frequency specifying means for specifying the frequency;
For each of the first and second waveform signals, the self passband characteristic is changed so that the passband characteristic is cut off except for the components in the vicinity of the frequency of the pitch component specified by the pitch component frequency specifying means. Variable filter means for filtering the waveform signal and extracting a pitch component;
Phase detection means for identifying the timing at which the pitch component extracted by the variable filter means has a predetermined phase for each of the first and second waveform signals;
By dividing the first and second waveform signals in the vicinity of the portion corresponding to the timing detected by the phase detection means, the end point of the first waveform signal and the start point of the second waveform signal are obtained. Connecting means for determining and connecting the first and second waveform signals to each other at the end point of the first waveform signal and the start point of the second waveform signal;
The connecting means divides the first and second waveform signals so that the end point of the first waveform signal and the start point of the second waveform signal are substantially equal to each other. ,
A waveform connecting device characterized by that. A first waveform signal representing a voice waveform and a second waveform signal to be subsequent to the first waveform signal are acquired, and for each acquired waveform signal, the pitch component of the voice represented by the waveform signal is obtained. Pitch component frequency specifying means for specifying the frequency;
For each of the first and second waveform signals, the self-pass so as to have a passband characteristic of a bandpass filter having a center frequency substantially coincident with the frequency of the pitch component specified by the pitch component frequency specifying means. Variable filter means for changing a band characteristic and filtering the waveform signal to extract a pitch component;
Phase detection means for identifying the timing at which the pitch component extracted by the variable filter means has a predetermined phase for each of the first and second waveform signals;
The first waveform signal and the second waveform signal are delimited at locations corresponding to the timing detected by the phase detection means, thereby determining the end point of the first waveform signal and the start point of the second waveform signal. Connection means for connecting the first and second waveform signals to each other at the end point of the first waveform signal and the start point of the second waveform signal;
A waveform connecting device characterized by that. The pitch component frequency specifying means specifies, for each of the first and second waveform signals, a frequency at which the cepstrum of the waveform signal takes a maximum value as a frequency of the pitch component of the waveform signal.
The waveform connecting device according to claim 1 or 2, wherein The phase detection means specifies the timing at which the value of the pitch component extracted by the variable filter means zero-crosses with a change in sign in a predetermined direction for each of the first and second waveform signals. is there,
The waveform connecting device according to claim 1, 2, or 3. A first waveform signal representing a voice waveform and a second waveform signal to be subsequent to the first waveform signal are acquired, and for each acquired waveform signal, the pitch component of the voice represented by the waveform signal is obtained. A pitch component frequency specifying step for specifying a frequency;
For each of the first and second waveform signals, variable filtering that extracts a pitch component by filtering with passband characteristics such that components other than those in the vicinity of the frequency of the pitch component specified in the pitch component frequency specifying step are blocked. Steps,
For each of the first and second waveform signals, a phase detection step for specifying a timing at which the pitch component extracted in the variable filtering step becomes a predetermined value phase;
By dividing the first and second waveform signals in the vicinity of the portion corresponding to the timing detected in the phase detection step, the end point of the first waveform signal and the start point of the second waveform signal are obtained. Determining and connecting the first and second waveform signals to each other at an end point of the first waveform signal and a start point of the second waveform signal,
In the connecting step, the first and second waveform signals are separated such that the end point of the first waveform signal and the start point of the second waveform signal are substantially equal to each other,
The waveform connection method characterized by the above-mentioned. Computer
A first waveform signal representing a voice waveform and a second waveform signal to be subsequent to the first waveform signal are acquired, and for each acquired waveform signal, the pitch component of the voice represented by the waveform signal is obtained. Pitch component frequency specifying means for specifying the frequency;
For each of the first and second waveform signals, the self passband characteristic is changed so that the passband characteristic is cut off except for the components in the vicinity of the frequency of the pitch component specified by the pitch component frequency specifying means. Variable filter means for filtering the waveform signal and extracting a pitch component;
Phase detection means for identifying the timing at which the pitch component extracted by the variable filter means has a predetermined phase for each of the first and second waveform signals;
By dividing the first and second waveform signals in the vicinity of the portion corresponding to the timing detected by the phase detection means, the end point of the first waveform signal and the start point of the second waveform signal are obtained. A program for determining and functioning the first and second waveform signals as connection means for connecting the end point of the first waveform signal and the start point of the second waveform signal to each other. ,
The connecting means divides the first and second waveform signals so that the end point of the first waveform signal and the start point of the second waveform signal are substantially equal to each other. ,
A program characterized by that.

Images