JP2001350500A - Speech speed changer - Google Patents

Abstract

PROBLEM TO BE SOLVED: To obtain a speech speed changer which automatically conducts a high quality speech speed change by waveform expanding and compressing a vowel segement in accordance with the speech speed change rate specified by a user. SOLUTION: Code vectors relative to learning voice data are recorded in a VQ code book 6 with appearance probabilities of vowels and consonants of each code vector. Then, input voice 1 is vector quantized by a segment discriminating means 7 while referring to the VQ code book to select code vectors. Then, waveform compressing and expanding is performed for the segement which is discriminated to be the vowel segment of the selected code vector by employing a speech speed change rate 10 given to a waveform expanding and compressing means 9.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a speech speed changing device which can freely change the speech speed of pre-recorded speech or speech directly input from a microphone or the like according to a user's request. Things.

[0002]

2. Description of the Related Art In recent years, car navigation systems and PDAs (P
personal Digital Assistant
s), speech synthesis S / W applicable to electronic toys and other applications
The demand for (software) is increasing. In such applications, it is desired that versatility such as changes in speech speed and pitch be realized with a small amount of computation and a small amount of memory. For such applications, the use of a small-scale text-to-speech S / W is conceivable, but in actual applications, only text-to-speech synthesis is rarely used. From the viewpoint of the quality of synthesized speech, fixed messages recorded,
Or, a combination of fixed messages and rule-based synthetic speech is often used.

[0003] In such S / W, a fixed message only accumulates a recorded voice and reproduces it. However, a voice recorded in units of words or phrases is connected, or connected with a rule synthesized voice to form a sentence. In order to obtain the continuity of, it is necessary to change the prosody such as the speech speed. In addition, it is effective for the user to be able to change the utterance speed in accordance with his or her preference and easiness of hearing, and there is a high demand for applications such as toys.

[0004] As described above, a method of changing the utterance speed of the original voice while maintaining the pitch and phonology has already been studied. As such a conventional speech speed changing device, there is, for example, a "voice speech speed conversion method" disclosed in JP-A-1-93795. FIG. 13 is a block diagram showing an example of the configuration of a conventional speech speed changing device based on the technique disclosed in the above publication. In the figure, 1 is an input voice, 2 is an analysis unit,
Reference numeral 3 denotes a control unit, 4 denotes a waveform connection unit, and 5 denotes a synthesized voice.

Next, the operation will be described. When the input voice 1 is input to the analysis unit 2, the analysis unit 2 outputs the input voice 1
, Voiced and unvoiced and voiced and unvoiced are discriminated, and for voiced, linear prediction analysis and pitch analysis are performed. Then, a time change of the resonance frequency and the bandwidth is obtained from the result of the linear prediction analysis, and the voiced sound is separated into a vowel and a consonant using the obtained result. The control unit 3 sets the expansion / contraction ratio between the voiced consonant and the unvoiced consonant to a value smaller than that of the vowel or the non-voice based on the knowledge that the change in the utterance speed of the consonant is smaller than that of the vowel or the non-voice. The waveform connection unit 4 expands and contracts the utterance time length of each section according to the expansion / contraction ratio determined by the control unit 3. The voiced section changes the speech speed by repeating or thinning out the pitch waveform based on the pitch period, and the silent section length or the unvoiced section changes the speech speed by deleting or repeatedly inserting a waveform having a length corresponding to the expansion / contraction ratio. Output.

[0006] In addition, as a document having a description related to such a conventional speech speed changing device, other than this, for example,
JP-A-6-266391, JP-A-6-274195, JP-A-10-78795, etc. relate to the prosody generation rules for determining the vowel length and consonant length of a synthesized sound. Japanese Patent Application Laid-Open No. 10-145897 discloses a method for controlling the waveform amplitude value of
Japanese Patent Laid-Open Publication No. Hei 6-222793 discloses a technique for determining a vowel length and a consonant length of a synthesized sound in accordance with a separately specified utterance tempo. JP-A-10-70790
No. Gazette.

[0007]

Since the conventional speech speed changing device is configured as described above, it is necessary to classify vowels and consonants and compress or decompress each of the vowels and consonants with a unique waveform expansion / contraction ratio. In addition, since it is segmented based on the results of linear prediction analysis, vowels and voiced consonants are likely to be incorrectly distinguished,
The boundary tends to be unclear, and it is necessary to consider the type of consonant in order to perform more natural waveform expansion and contraction. This framework is not sufficient, and the analysis unit 2 needs to perform analysis. The amount of calculation is large, and there is a problem that it is difficult to realize the operation in real time.

SUMMARY OF THE INVENTION The present invention has been made to solve the above-described problems, and is directed to vector quantization (Vector).
A vowel and a consonant are discriminated using parameter information obtained by quantification (hereinafter sometimes abbreviated as VQ), and a unique waveform expansion / contraction rate is given to each vowel section and consonant section based on the result of the determination to expand and contract the waveform. It is an object of the present invention to provide a speech speed changing device which can perform the waveform expansion and contraction in real time with high quality.

[0009]

A speech speed changing apparatus according to the present invention records a code vector of learning speech data in a VQ codebook together with the appearance probabilities of vowels and consonants of each code vector, and performs section discrimination. Means for selecting a code vector by vector-quantizing the input voice with reference to the VQ codebook, and determining whether the section is a vowel section or a consonant section from the appearance probability of a vowel or consonant in the selected code vector. When the speech speed change rate is given to the waveform expansion / contraction means, the waveform expansion / contraction is performed on the section determined as the vowel section by the section determination means based on the given speech speed change rate. .

The speech speed changing device according to the present invention defines a unique expansion / contraction ratio for each of a vowel section and a consonant section, or for each of a vowel section, a consonant section and a silent section.
According to the specified expansion / contraction ratio, the waveform expansion / contraction means obtains a waveform expansion / contraction rate for each section of a vowel section, a consonant section, and a silent section, and performs a waveform expansion / contraction based on the waveform expansion / contraction rate. is there.

The speech speed changing device according to the present invention is provided with an expansion / contraction ratio table for storing a unique expansion / contraction ratio defined for each type or category of phoneme. Refers to the expansion / contraction ratio table, finds a waveform expansion / contraction ratio for each section corresponding to the expansion / contraction ratio for each phoneme or phoneme category of the input voice, and performs waveform expansion / contraction based on the ratio.

The speech speed changing apparatus according to the present invention stores a VQ codebook in which a code vector of learning speech data is recorded together with an expansion / contraction ratio specific to each code vector.
Referring to the waveform expansion and contraction means, vector quantization is performed on the input voice to select a code vector, and when a speech speed change rate is given, the waveform expansion and contraction for each section corresponding to the expansion and contraction ratio for each selected code vector is performed. The ratio is obtained, and the waveform is expanded and contracted based on the ratio.

In the speech speed changing device according to the present invention, a phoneme label of an input voice is extracted by a labeling means, and a rhythm perception point table which stores a unique rhythm perception point for each phoneme sequence in the rhythm perception point extraction means is referred to. Then, the rhythm perception point of the input voice is extracted from the phoneme label, and when the speech speed change rate is given to the waveform expansion / contraction means, the waveform expansion / contraction rate for each section is determined so that the expansion / contraction rate of the rhythm perception point interval is constant. Waveform expansion and contraction are performed based on this.

The speech rate changing device according to the present invention defines a unique expansion / contraction ratio for each vowel section and consonant section, or for each vowel section, consonant section and silent section, by means of the waveform expanding / contracting means. When the rate is given, the expansion and contraction ratio of the vowel section and the consonant section or the expansion and contraction ratio of the vowel section, the consonant section and the silent section so that the expansion and contraction rate of the rhythm perception point interval becomes constant. The waveform expansion / contraction ratio is obtained, and the waveform expansion / contraction is performed based on the obtained ratio.

The speech speed changing device according to the present invention is provided with an expansion / contraction ratio table to store a specific expansion / contraction ratio defined for each phoneme type or phoneme category. The expansion / contraction means refers to the expansion / contraction ratio table and calculates a waveform expansion / contraction ratio for each section corresponding to the expansion / contraction ratio for each phoneme or each phoneme category of the input voice so that the expansion / contraction ratio of the rhythm perception point interval is constant. , And expands and contracts the waveform based on it.

In the speech speed changing apparatus according to the present invention, a phoneme label of an input voice is extracted by a labeling means, and a rhythm perception point unique to the phoneme sequence and a contraction ratio specific to the phoneme sequence are extracted by the rhythm perception point extraction means. With reference to the rhythm perception point and expansion / contraction ratio table that stores the rhythm perception section and the rhythm perception point of the input voice and the expansion / contraction ratio for each phoneme section from the phoneme label, The waveform expansion / contraction ratio is determined for each phoneme section in accordance with the expansion / contraction ratio for each phoneme section so that the expansion / contraction rate of the perceived point interval is constant, and the waveform is expanded / contracted based on the waveform expansion / contraction rate.

In the speech speed changing device according to the present invention, the vowel section voice is extracted by referring to the VQ codebook by the section discriminating means, and the vowel center position in the extracted vowel section voice is determined by the vowel center extracting means. The waveform expansion / contraction means defines a specific expansion / contraction ratio for each vowel section and consonant section, or each vowel section, consonant section, and silent section, and when a speech speed change rate is given, the expansion / contraction rate of the vowel center interval becomes The expansion and contraction ratio between the vowel section and the consonant section or the expansion and contraction ratio of each section according to the expansion and contraction ratio between the vowel section, the consonant section and the silence section is determined so that the waveform expansion and contraction is performed based on the obtained ratio. It is.

In the speech speed changing device according to the present invention, the vowel section voice is extracted by referring to the VQ codebook by the section discriminating means, and the vowel center position in the extracted vowel section voice is determined by the vowel center extracting means. When the speech rate change rate is given, the waveform expansion / contraction means by the waveform expansion / contraction means makes the waveform expansion / contraction rate in the vowel center position approach the maximum / minimum at the vowel center position, and approaches the waveform expansion / contraction rate in the consonant section as it approaches the boundary with the consonant section. Is determined, and the waveform expansion / contraction of each section is performed based on this.

The speech speed changing device according to the present invention determines the pitch approximate reference position from the maximum value and the minimum value of the approximate pitch obtained by the pitch frequency analysis of the input voice by the pitch analysis means. Means, and when the speech speed change rate is given, the expansion / contraction ratio of the vowel section and the consonant section, or the vowel section, by the waveform expansion / contraction means so that the expansion / contraction rate of the pitch approximate reference position interval becomes constant. And a waveform expansion / contraction rate for each section corresponding to the expansion / contraction ratio between the consonant section and the silent section, and the waveform expansion / contraction is performed based on the waveform expansion / contraction rate.

In the speech speed changing apparatus according to the present invention, the pitch outline is determined based on the pitch frequency analysis result of the pitch analysis means, and the pitch outline dividing means includes the provisional consonant section voice from the section determination means in the vowel section. Vowel section and consonant section, and the vowel section and consonant section are discriminated again, and when the speech speed change rate is given to the waveform expansion / contraction means, the vowel section and consonant section are discriminated again. , A specific expansion / contraction ratio is determined, and the waveform expansion / contraction is performed based on the waveform expansion / contraction ratio for each section obtained according to the expansion / contraction ratio.

[0021]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS One embodiment of the present invention will be described below. Embodiment 1 FIG. FIG. 1 is a block diagram showing a configuration example of a speech speed changing device according to Embodiment 1 of the present invention. In the figure, reference numeral 1 denotes an input voice which is input to the speech speed changing device and is prepared by voice data prepared in advance in the system, and 5 denotes a speech speed changed by the speech speed changing device and output to the outside. This is a synthesized speech. Reference numeral 6 denotes a V which records a code vector of the learning speech data together with a probability of occurrence of a phoneme such as a vowel or a consonant of each code vector.
It is a Q codebook. 7 refers to the VQ codebook 6, performs vector quantization on the input speech 1 input, selects a code vector, and, based on the appearance probability of phonemes in the selected code vector, determines that the section is a vowel section. A section discriminating means for discriminating whether there is a consonant section or a consonant section. Reference numeral 8 denotes a vowel section voice output from the section discriminating means 7 based on the result of the judgment. 9 is the vowel section voice 8 determined by the section determination means 7
Waveform expansion / contraction means for performing waveform compression or decompression at a desired speech rate change rate to generate and output a synthesized voice 5, and 10 is a speech rate change rate given to the waveform expansion / contraction means 9.

Next, the operation will be described. Here, it is assumed that the input voice 1 is voice data prepared in advance by a system that outputs voice. The VQ codebook 6 is used for vector-quantizing the audio data of the input audio 1, and is a pattern (code vector) obtained by dividing the audio waveform spectrum data in the learning audio data by a general clustering method. And the code set. In the VQ codebook 6 according to the first embodiment, the vowel appearance probability described later is stored together with the code vector. The section discriminating means 7 having received the input of the input voice 1 outputs the V
With reference to the Q codebook 6, a vowel and consonant section in the input voice 1 is determined, and a vowel section voice 8, which is the input voice 1 in the vowel section portion, is sent to the waveform expansion / contraction unit 9. Note that a section determination method using the VQ codebook 6 will be described later.

The vowel section voice 8 is input to the waveform expansion / contraction means 9 and a speech speed change rate 10 is also given. The speech speed change rate 10 is a value given when the user changes the speech speed of the voice, and is input to the waveform expansion / contraction means 9 as an expansion / contraction ratio with respect to the original voice. For example, if it is desired to output the voice at a rate β times the original voice (β> 1), the speech rate change rate α is α = β. The change rate α is α = 1 / β. The waveform expansion / contraction means 9 expands / contracts the waveform of the vowel section voice 8 in accordance with the speech speed change rate 10, and does not perform waveform expansion / contraction for other sections. In this case, the waveform expansion / contraction rate of the vowel section is determined as follows based on the vowel section length and the consonant section length.

For example, the waveform expansion / contraction ratio α ′ of the vowel part is determined so that α = {α ′ · Lv + 1 · (L−Lv)} / L as the entire voice section L and the vowel section length Lv.
That is, the waveform expansion / contraction ratio α ′ of the vowel portion is obtained from the following expression obtained by modifying this expression. α ′ = 1 + (α−1) · L / Lv Incidentally, the vowel section length Lv is obtained at the time of section discrimination.

Here, as a method of expanding and contracting the vowel section voice 8 by the waveform expanding and contracting means 9, a method based on a known PSOLA method is generally used, but an Oscillator Mo is used.
It is also effective to use a known method called "del" having no pitch dependency. The speech whose speech speed has been changed by the waveform expansion / contraction means 9 is output as the synthesized speech 5.

The section discrimination using the VQ codebook 6 will be described below. First, vector quantization of training speech data labeled as a vowel or consonant is performed in advance using a VQ codebook 6 consisting of only a code and its code vector. The quantization scale at that time is a spectral parameter. Next, for each code vector in the VQ codebook 6, the appearance frequency at which a vowel or consonant is selected is determined, and the appearance probability of this vowel or consonant is recorded in the VQ codebook 6 together with the code vector. The section discriminating means 7 refers to the VQ codebook 6 and inputs the input voice 1
Is subjected to spectral analysis to perform vector quantization.
By this vector quantization, a code vector is selected from the VQ codebook 6, and it is determined from the appearance probability recorded therein whether the section is a vowel section or a consonant section. For example, in a code vector selected by vector quantization for a certain voice section, if the appearance probability of a vowel recorded therein is 70%, the section is determined to be a vowel section, and if 30%, the section is determined to be a consonant section. . In this example, the threshold for discriminating a vowel section and a consonant section is set to 50%.

As described above, according to the first embodiment, it is possible to automatically perform waveform expansion and contraction for a vowel section, and to change all voice sections uniformly at the same expansion and contraction rate. This can solve the problem of the deterioration of naturalness due to the above, and has an effect that it is possible to change the utterance speed satisfactorily.

Embodiment 2 FIG. 2 is a block diagram showing a configuration example of a speech speed changing device according to a second embodiment of the present invention. In the figure, 1 is an input voice, 5 is a synthesized voice, 6 is a VQ codebook, 7 is a section discriminating means, 8 is a vowel section voice, 10 is a speech speed change rate, and these are given the same reference numerals in FIG. This is a part equivalent to each component and data according to the first embodiment. Reference numeral 11 denotes a consonant section voice output from the section determination means 7 based on the determination result. Reference numeral 9 denotes waveform expansion / contraction means corresponding to the components denoted by the same reference numerals in FIG. 1, and defines a specific expansion / contraction ratio for each vowel section and consonant section, or each vowel section, consonant section, and silent section. Given a speech speed change rate of 10, a waveform expansion / contraction ratio is calculated for each section according to the expansion / contraction ratio between a vowel section and a consonant section, or an expansion / contraction ratio between a vowel section, a consonant section, and a silent section. Embodiment 1 in that
Is different from that of

Next, the operation will be described. Again,
The input sound 1 is assumed to be sound data prepared in advance by a system for outputting sound. The VQ codebook 6 is used for performing vector quantization of the audio data 1, and has the same configuration as that of the first embodiment. The section discriminating means 7 obtains a section between a vowel and a consonant in the input speech 1, and the input speech 1 in the vowel section is used as the vowel section speech 8, and the input speech 1 in the consonant section is used as the consonant section speech 11, and the waveform expansion and contraction means 9 respectively. send. In addition,
As an example of the section determination, a method similar to that of the first embodiment can be used.

Similarly to the first embodiment, the speech speed change rate 10 is a value given when the user changes the speech speed of the voice, and is input to the waveform expansion / contraction means 9 as an expansion / contraction ratio with respect to the original voice. . The expansion / contraction ratio of the vowel section and the consonant section is defined in the waveform expansion / contraction means 9. It is generally known that the degree of expansion and contraction due to a change in speech speed is larger in a vowel section than in a consonant section. The value of the expansion / contraction ratio is obtained from the correlation between the average time length of the entire vowel and the entire consonant obtained from the learning speech data having different speech speeds and the speech speed.

The user changes the speech speed change rate 1 to the waveform stretching means 9.
When 0 is given, the waveform expansion / contraction ratio for each vowel section and consonant section is determined from the expansion / contraction ratio between the vowel section and the consonant section defined above. Here, the vowel section length is Lv, and the consonant section length is L
Assuming that c, the total expansion / contraction rate of the vowel section length Lv and the consonant section length Lc is the speech speed change rate α (α> 0). For example, if the expansion ratio of a vowel section is β and the expansion ratio of a consonant section is γ,
Regarding the expansion / contraction ratio between the vowel section and the consonant section, 0 <γ <1 <β in the case of waveform expansion, and 0 <β <1 in the case of waveform compression.
If <γ, it is determined by obtaining a coefficient α ′ such that the expansion and contraction rate of the vowel section is α′β and the expansion and contraction rate of the consonant section is α′γ.

That is, from α = α ′ (β · Lv + γ · Lc) / (Lv + Lc) α ′ = α (Lv + Lc) / (β · Lv + γ · Lc), the waveform expansion / contraction ratio of the vowel section is α′β = α・ Β (Lv + Lc) / (β ・ Lv + γ ・ L
c) The waveform expansion / contraction rate in the consonant section is α′γ = α · γ (Lv + Lc) / (β · Lv + γ · L
c) Incidentally, it is assumed that the vowel section length Lv and the consonant section length Lc are obtained at the time of section determination.

As described above, the waveform expanding / contracting means 9 outputs the input voice 1
The vowel section voice 8 and the consonant section voice 11 are expanded and contracted at a predetermined waveform expansion and contraction rate. The method of expanding and contracting the waveform is the same as that of the first embodiment. The speech whose speech speed has been changed by the waveform expansion / contraction means 9 is the synthesized speech 5
Is output as

As described above, according to the second embodiment, it is possible to automatically perform the waveform expansion and contraction by the individual expansion and contraction ratios for the vowel section and the consonant section, and to uniformly perform the same expansion and contraction for all voice sections. This makes it possible to solve the problem of naturalness degradation due to the change at the rate, thereby achieving an effect that a good change in the utterance speed can be realized.

Embodiment 3 FIG. FIG. 3 is a block diagram showing a configuration example of a speech speed changing device according to Embodiment 3 of the present invention. Parts corresponding to the respective components and data of the first embodiment or the second embodiment are denoted by the same reference numerals as in FIG. 1 or FIG. 2, and description thereof is omitted.
In the figure, reference numeral 12 denotes an expansion / contraction ratio table that stores types of phonemes and specific expansion / contraction ratios defined for each phoneme category. The waveform expansion / contraction means 9 is a waveform expansion / contraction means corresponding to a component of the first or second embodiment shown in FIG. 1 or FIG.
When 0 is given, the waveform expansion / contraction is performed by referring to the expansion / contraction ratio table 12 and calculating the waveform expansion / contraction ratio for each section corresponding to the expansion / contraction ratio for each phoneme or phoneme category of the input speech 1. Are different.

Next, the operation will be described. Again,
The input sound 1 is assumed to be sound data 1 prepared in advance by a system for outputting sound. The VQ codebook 6 is used for performing vector quantization of audio data, and has the same configuration as that of the first embodiment. The section discriminating means 7 finds a section of a vowel and a consonant in the input speech 1, and converts the vowel section speech 8, which is the input speech 1 of the vowel section, and the consonant section speech 11, which is the input speech 1 of the consonant section, into the waveform expansion and contraction means 9. Send to However, the consonant section voices are classified for each phoneme category described later. The example of the section determination is the same as that in the second embodiment.

Similarly to the first embodiment, the speech speed change rate 10 is a value given when the user changes the speech speed of the voice, and is input to the waveform expansion / contraction means 9 as an expansion / contraction ratio with respect to the original voice. . Note that the expansion ratio table 12 describes the expansion ratio for each phoneme category as described above. Examples of this phoneme category include vowels, fricative consonants, explosive consonants, nasal sounds, flowing sounds, and sound repellents. Determine the stretch ratio. The value of the expansion / contraction ratio for each category is obtained from the correlation between the average time length for each phoneme category and the utterance speed obtained from learning speech data having different utterance speeds.

The user changes the speech speed change rate 1 to the waveform stretching means 9 by the user.
When 0 is given, the expansion / contraction ratio table 12 is referred to, and the waveform expansion / contraction ratio between the vowel section and the consonant section is determined based on the expansion / contraction ratio specific to each phoneme category or phoneme category stored in the expansion / contraction ratio table 12. I do. Note that the method of determining the waveform expansion / contraction ratio from the expansion / contraction ratio given to the vowel section and the consonant section is the same as in the second embodiment. As described above, the waveform expanding / contracting means 9 expands / contracts the input speech 1 for each of the vowel sections and the sections classified according to the consonant categories. The method of expanding and contracting the waveform is the same as in the first embodiment.
The speech whose speech speed has been changed by the waveform expansion / contraction means 9 is output as the synthesized speech 5.

As described above, according to the third embodiment, it is possible to automatically perform waveform expansion / contraction based on individual expansion / contraction ratios for vowel sections and consonant categories. In addition, it is possible to solve the problem of deterioration of naturalness caused by uniformly changing all voice sections at the same expansion / contraction ratio, and it is possible to obtain an effect that it is possible to achieve a favorable change in speech rate.

Embodiment 4 FIG. FIG. 4 is a block diagram showing a configuration example of a speech speed changing device according to Embodiment 4 of the present invention. In the figure, 1 is an input voice, 5 is a synthesized voice, 10
Is a speech speed change rate, which is equivalent to each component and data in the first to third embodiments, which are denoted by the same reference numerals in FIGS. 1 to 3. Reference numeral 13 denotes a VQ with an expansion / contraction ratio that records a code vector for learning speech data together with an expansion / contraction ratio specific to each code vector.
It is a code book. 14 performs vector quantization on the input speech 1 with reference to the VQ codebook 13 with expansion / contraction ratio, selects a vector code, and changes the speech speed change rate 1
This is a VQ waveform expansion / contraction unit that obtains a waveform expansion / contraction ratio for each section corresponding to the expansion / contraction ratio of each selected code vector when 0 is given, and performs waveform expansion / contraction based on the obtained ratio.

Next, the operation will be described. Again,
The input sound 1 is assumed to be sound data prepared in advance by a system for outputting sound. In addition, V with expansion ratio
The Q codebook 13 is used to perform vector quantization of the audio data, and its configuration will be described later. The speech speed change rate 10 is also the same as that in the first embodiment, and is a value given when the user changes the speech speed of the voice, and is input to the VQ waveform expansion / contraction unit 14 as an expansion / contraction ratio with respect to the original voice. The VQ waveform expanding / contracting means 14 refers to the VQ codebook 13 with the expansion / contraction ratio, and expands / contracts the waveform based on the expansion / contraction ratio stored therein and the speech speed change rate 10 given by the user. The VQ waveform expansion / contraction means 1
4 will be described later. Thus,
The voice whose speech speed has been changed by the VQ waveform expanding / contracting means 14 is output as the synthesized voice 5.

The structure of the VQ codebook 13 with the expansion ratio will be described below. The VQ codebook 13 with expansion and contraction ratio used for the vector quantization of the audio data includes a pattern (code vector) obtained by dividing the audio waveform spectrum data in the learning audio data by a general clustering method, and a set of the code. Consists of For the VQ codebook 13 with the expansion / contraction ratio, the vector quantization of the learning voice data labeled as a vowel or a consonant is performed in advance. The quantization scale at that time is a spectral parameter.

Next, for each code vector in the VQ codebook 13 with expansion / contraction ratio, the appearance frequency at which a vowel or consonant is selected is determined, and the appearance probability of a vowel section or consonant section together with the code vector is added to the VQ code with expansion / contraction ratio. Record in Book 13. Furthermore, it is determined whether each code vector corresponds to a vowel or a consonant from the appearance probability of the vowel section and the consonant section recorded for each code vector, and the vowel section for each code vector is determined. Alternatively, a specific expansion / contraction ratio is given to the consonant section and stored. For example, assuming that a threshold value for determining a vowel section and a consonant section is 50%, if the vowel appearance probability of the code vector is 70%, the waveform expansion rate of the vowel section is 30%, and the waveform expansion rate of the consonant section is 30%. Record. The expansion ratio here is the same as that of the second embodiment.

Next, waveform expansion and contraction in the VQ waveform expansion and contraction means 14 will be described. First, the VQ waveform expansion / contraction means 14
Referring to the VQ codebook 13 with the expansion ratio, the input voice 1
Is subjected to spectral analysis and vector-quantized. Based on this vector quantization, a code vector is selected from the VQ codebook 13 with an expansion ratio, and the expansion ratio stored therein is read. Speech rate change rate from user 10
Is given, the waveform expansion / contraction ratio for each vowel section and consonant section is determined based on the expansion / contraction ratio read from the VQ codebook 13 with the expansion / contraction ratio. The method of determining the waveform expansion / contraction ratio is the same as that in the second embodiment. The waveform of the input voice 1 in the section is expanded / contracted in accordance with these waveform expansion / contraction rates, and the resultant is output as the synthesized voice 5. The method of expanding and contracting the waveform is the same as in the first embodiment.

As described above, according to the fourth embodiment, by recording the unique expansion / contraction ratio in the code vector, the waveform expansion / contraction at the individual expansion / contraction ratio for the vowel section and the consonant section can be performed easily and automatically. And the problem that naturalness is degraded by changing all voice sections uniformly at the same expansion / contraction ratio can be solved, and a good speech rate change can be realized. Is obtained.

Embodiment 5 FIG. FIG. 5 is a block diagram showing a configuration example of a speech speed changing device according to a fifth embodiment of the present invention. The components corresponding to the components and data of the first to fourth embodiments are described with reference to FIGS.
The same reference numerals are given and the description is omitted. In the figure,
Reference numeral 15 denotes a labeling unit that determines a phoneme boundary and a phoneme type in the input speech 1 and extracts a phoneme label from the identification result, and 16 denotes the phoneme label extracted by the labeling unit 15. Reference numeral 17 denotes a rhythm perception point table that stores a rhythm perception point unique to a phoneme sequence. Reference numeral 18 denotes a rhythm perception point table.
This is a rhythm perception point extraction unit that extracts a rhythm perception point of the input speech 1 from the phoneme label 16 extracted by the labeling unit 15. Although the waveform expanding / contracting means 9 is shown in FIGS. 1 to 3 with the same reference numerals, it also corresponds to a component of the first to third embodiments, but a speech speed change rate 10 is given. In this case, the waveform expansion and contraction ratio is determined for each section so that the expansion and contraction ratio of the rhythm perception point interval is constant, and the waveform is expanded and contracted based on the waveform expansion and contraction ratio. Is different.

Next, the operation will be described. Again,
The input sound 1 is assumed to be sound data prepared in advance by a system for outputting sound. When the input speech 1 is input, the labeling means 15 determines a phoneme boundary and a phoneme type in the input speech 1, extracts a phoneme label 16 based on the discrimination result, and sends it to the rhythm perception point extraction means 18. send. As a method of determining the phoneme label 16, for example, there is a method in which a phoneme boundary and a phoneme type are manually given in advance from the input speech 1 and its utterance content.

On the other hand, the rhythm perception point table 17 stores information on rhythm perception points corresponding to combinations of consonants (C) and vowels (V). The rhythm perception point is the position on the audio waveform where the human perceptually perceives the rhythm, and is a CV (consonant-
A unique value is taken for each phoneme sequence such as vowels or VCV (vowel-consonant-vowel). The interval between the rhythm perception points corresponds to the rhythm of the natural voice, and the ratio between the rhythm perception point intervals is preserved for the same combination of phonemes even when the speech speed changes.

FIG. 6 shows an example of the contents of the rhythm perception point table 17. FIG. 6 shows the rhythm perception point table 17.
As shown in (a), a phoneme sequence and a rhythm perception point corresponding to the phoneme sequence are described. That is, FIG.
In the example of (a), a rhythm perception point corresponding to a CV or VCV phoneme sequence by preceding V, C, and succeeding V is described. Further, in the example of FIG. 6A, the rhythm perception point indicates the frame position in the speech unit of the CV or VCV, and the correspondence with the approximate power of the speech unit is shown in FIG. 6B. Position. Rhythm perception point extracting means 18
Is the rhythm perception point table 1 from the set of phoneme labels 16.
7, the position of the rhythm perception point on the input voice 1 is determined.

Similarly to the first embodiment, the speech speed change rate 10 is a value given when the user changes the speech speed of the voice, and is input to the waveform expansion / contraction means 9 as an expansion / contraction ratio with respect to the original voice. . The waveform expansion / contraction means 9 calculates the speech speed change rate 10
Expands and contracts the input voice 1 according to the input voice 1
The upper rhythm perception point interval is expanded or contracted by the same speech speed change rate 10. That is, first, the labeling means 1
Based on the phoneme label 16 extracted in step 5, the vowel section and the consonant section within the rhythm perception point interval are determined. The method of determining the waveform expansion / contraction ratio between a vowel section and a consonant section from the expansion / contraction ratio between a vowel and a consonant defined by the waveform expansion / contraction means 9 is the same as in the second embodiment. Corrugated expansion / contraction means 9
Compresses or decompresses the input speech 1 in each section according to these values. The method of expanding and contracting the waveform is the same as in the first embodiment. The voice whose speech speed has been changed by the waveform expansion / contraction means 9 is output as the synthesized voice 5.

As described above, according to the fifth embodiment, the expansion and contraction of the waveform can be automatically performed, and the expansion and contraction of the waveform in consideration of the rhythm perception point which is the basis of the time structure of the natural voice can be performed. Thus, an effect is obtained in which the utterance speed can be changed more naturally than when all voice sections are converted uniformly.

Embodiment 6 FIG. FIG. 7 is a block diagram showing a configuration example of a speech speed changing device according to Embodiment 6 of the present invention. The components corresponding to the respective components and data in the first to fifth embodiments are denoted by the same reference numerals as those in FIGS. 1 to 5, and description thereof is omitted. The speech speed changing device according to the sixth embodiment is obtained by connecting the expansion / contraction ratio table 12 to the waveform expanding / contracting means 9 of the speech speed changing device according to the fifth embodiment. In this case, the waveform expansion and contraction is performed with reference to the expansion and contraction ratio table 12, which is different from the fifth embodiment.

Next, the operation will be described. Again,
The input sound 1 is assumed to be sound data prepared in advance by a system for outputting sound. Labeling means 15
Determines a phoneme boundary and a phoneme type in the input speech 1 and uses the result as a phoneme label 16 for the rhythm perception point extracting means 1.
Send to 8. Note that the phoneme label 16 is determined according to the fifth embodiment.
The same as in the case of. The rhythm perception point table 17 is a table in which information on rhythm perception points corresponding to combinations of phonemes (C and V) is stored.
This is the same as the case of the fifth embodiment shown in FIG. The rhythm perception point extracting means 18 refers to the rhythm perception point table 17 from the set of phoneme labels 16 extracted by the labeling means 15,
The rhythm perception point position on the input voice 1 is determined.

Similarly to the first embodiment, the speech speed change rate 10 is a value given when the user changes the speech speed of the voice, and is input to the waveform expansion / contraction means 9 as an expansion / contraction ratio with respect to the original voice. . The waveform expansion / contraction means 9 calculates the speech speed change rate 10
Expands and contracts the input voice 1 according to the input voice 1
The upper rhythm perception point interval is expanded or contracted by the same speech speed change rate 10. That is, first, the labeling means 1
Based on the extracted phoneme label 16 of 5, the breakdown of the section for each phoneme category such as vowels and burst consonants within the rhythm perception point interval is determined. Examples of this phoneme category are the same as those described in the third embodiment. By referring to the expansion / contraction ratio table 12, the expansion / contraction ratio corresponding to the phoneme category for each section is obtained. The configuration of the expansion / contraction ratio table 12 and the method of determining the waveform expansion / contraction ratio are the same as those in the third embodiment. The waveform expansion / contraction means 9 compresses or expands the input voice 1 in each section based on these values. The method of expanding and contracting the waveform is the same as in the first embodiment. The speech whose speech speed has been changed by the waveform expansion / contraction means 9 is output as the synthesized speech 5.

As described above, according to the sixth embodiment, the waveform expansion and contraction can be automatically performed, and the waveform expansion and contraction in consideration of the rhythm perception point and the phoneme type, which are the basics of the time structure of natural speech. Is performed, it is possible to achieve an effect that a more natural change in the utterance speed can be realized than when all voice sections are converted uniformly.

Embodiment 7 FIG. FIG. 8 is a block diagram showing a configuration example of a speech speed changing device according to a seventh embodiment of the present invention. In FIG. 8, parts corresponding to the respective components and data of the first to sixth embodiments are not shown. 1 to 5 and FIG. 7, and the description thereof is omitted. In the figure, reference numeral 19 denotes a rhythm perception point unique to the phoneme sequence and an expansion / contraction ratio specific to the phoneme sequence.
Is a rhythm perception point and expansion / contraction ratio table referenced by. In the seventh embodiment, the expansion ratio table 12 and the rhythm perception point table 17 prepared separately in each of the above embodiments are combined into one rhythm perception point / expansion ratio table 19 in the seventh embodiment. This is different from the sixth embodiment.

Next, the operation will be described. Again,
The input sound 1 is assumed to be sound data prepared in advance by a system for outputting sound. Labeling means 15
Performs a discrimination between a phoneme boundary and a phoneme type in the input speech 1, and sends the discrimination result as a phoneme label 16 to the rhythm perception point extracting means 18. The determination of the phoneme label 16 is the same as in the fifth embodiment. The rhythm perception point and expansion / contraction ratio table 19 stores information on rhythm perception points corresponding to combinations of phonemes (C and V) and expansion / contraction ratios of phoneme waveforms corresponding to the combinations. The content of the rhythm perception point is the same as that of the fifth embodiment shown in FIG. The expansion and contraction ratio of the waveform is obtained by calculating the expansion and contraction ratio for each phoneme section of CV or VCV and recording the obtained value.

The rhythm perception point extracting means 18 refers to the rhythm perception point and expansion / contraction ratio table 19 based on the set of phoneme labels 16 extracted by the labeling means 15, and determines the rhythm perception point position on the input voice 1. To obtain the expansion / contraction ratio corresponding to the phoneme sequence. Note that the speech speed change rate 10 is the same as that in the first embodiment, and is a value given when the user changes the speech speed of the voice, and is input to the waveform expansion / contraction means 9 as an expansion / contraction ratio with respect to the original voice. The waveform expansion / contraction means 9 expands / contracts the input voice 1 in accordance with the speech speed change rate 10, and at this time,
Speech rate change rate 10 with the same rhythm perception point interval on input speech 1
To be stretched.

That is, the breakdown of the section for each phoneme category such as vowels and burst consonants within the rhythm perception point interval is first determined by the phoneme label 16. An example of this phoneme category is the same as in the third embodiment. The waveform expansion and contraction ratio for each section is obtained from the expansion and contraction ratio obtained by referring to the rhythm perception point and expansion and contraction ratio table 19 and the given speech speed change ratio 10. The method of determining the waveform expansion / contraction ratio is the same as that of the third embodiment. Corrugated expansion / contraction means 9
Expands and contracts the input voice 1 in each section according to these values. The method of expanding and contracting the waveform is the same as in the first embodiment. The speech whose speech speed has been changed by the waveform expansion / contraction means 9 is output as the synthesized speech 5.

As described above, according to the seventh embodiment, the waveform expansion and contraction can be automatically performed, and the waveform expansion and contraction in consideration of the rhythm perception points and phoneme types, which are the basics of the time structure of natural speech. Is performed, it is possible to achieve an effect that a more natural change in the utterance speed can be realized than when all voice sections are converted uniformly.

Embodiment 8 FIG. FIG. 9 is a block diagram showing a configuration example of a speech speed changing device according to Embodiment 8 of the present invention. In the figure, 1 is an input voice, 5 is a synthesized voice, 6 is a VQ codebook, 7 is a section discriminating means, 8 is a vowel section voice, 9 is a waveform expanding / contracting means, and 10 is a speech speed change rate.
5 and FIGS. 7 and 8 are the same as the components and data in the first to seventh embodiments shown with the same reference numerals.

Reference numeral 20 denotes vowel center extracting means for determining the center position of the vowel section determined by the section determining means 7.
Reference numeral 21 denotes a vowel center position extracted by the vowel center extracting means 20. The waveform expansion / contraction means 9 defines a specific expansion / contraction ratio for each vowel section and consonant section, or each vowel section, consonant section, and silent section. Extracted vowel center position 21
, So that the expansion and contraction rate of the vowel center interval is constant,
The waveform expansion / contraction ratio is determined for each section according to the expansion / contraction ratio between a vowel section and a consonant section, or the expansion / contraction ratio between a vowel section, a consonant section, and a silent section, and the waveform is expanded / contracted based on the waveform expansion / contraction rate. This is different from the waveform expanding / contracting means in each of the above-described embodiments in which 1 and the like are assigned the same reference numerals.

Next, the operation will be described. Again,
The input sound 1 is assumed to be sound data prepared in advance by a system for outputting sound. VQ Codebook 6
Are used to perform vector quantization of audio data, and the configuration is the same as that in the first embodiment. The section discriminating means 7 refers to the VQ codebook 6 to find a section between vowels and consonants in the input speech 1 and sends the vowel section speech 8 which is the input speech 1 in the vowel section to the vowel center extracting means 20. As an example of the section determination, the same method as in the first embodiment can be used.
The vowel center extracting means 20 having received the vowel section voice 8 extracts a vowel center position 21 which is a reference point in the vowel section,
It is sent to the waveform expansion / contraction means 9. As an example of the vowel center position extraction method, there is a method of determining a position where the power is maximum in a vowel section.

The speech speed change rate 10 is the same as that in the first embodiment, and is a value given when the user changes the speech speed of the voice, and is input to the waveform expansion / contraction means 9 as an expansion / contraction ratio with respect to the original voice. . The waveform expansion / contraction means 9 has a speech speed change rate 10
Expands and contracts the vowel section voice 8 and the consonant section voice 11 according to the vowel center position 2 on the vowel section voice 8.
The center position interval of 1 is expanded and contracted at the same speech speed change rate 10. The method of determining the waveform expansion / contraction ratio between the vowel section and the consonant section from the expansion / contraction ratio between the vowel and the consonant specified by the waveform expansion / contraction means 9 is the same as in the second embodiment. The waveform expanding / contracting means 9 compresses or expands the input voice 1 in each section according to these values. The method of expanding and contracting the waveform is the same as in the first embodiment. The voice changed by the waveform expansion / contraction means 9 is output as the synthesized voice 5.

As described above, according to the eighth embodiment, the expansion and contraction of the waveform can be automatically performed, and the expansion and contraction of the waveform in consideration of the center position of the vowel so that the interval between the vowels of the natural voice can be used. Is performed, it is possible to achieve an effect that a more natural change in the utterance speed can be realized than when all voice sections are converted uniformly.

Embodiment 9 FIG. 10 is a block diagram showing a configuration example of a speech speed changing device according to a ninth embodiment of the present invention. FIG. 10 is a diagram showing components corresponding to the components and data of the first to eighth embodiments. 1 to 5 and FIGS. 7 to 9 are denoted by the same reference numerals, and description thereof is omitted. In the figure, reference numeral 22 denotes a waveform expansion / contraction ratio in a vowel section when a speech speed change rate 10 is given.
0 is determined to be maximum or minimum at the extracted vowel center position 21 and to approach the waveform expansion / contraction rate of the consonant section as it approaches the boundary with the consonant section. Based on the determined waveform expansion / contraction rate, This is a vowel center waveform expansion / contraction unit that expands / contracts the waveform. Embodiment 9 is a waveform expanding / contracting means 9
Is different from the eighth embodiment in that this is replaced by the vowel central portion waveform expansion / contraction means 22.

Next, the operation will be described. Again,
The input sound 1 is assumed to be sound data prepared in advance by a system for outputting sound. The VQ codebook 6 is also used for performing vector quantization of audio data, and has the same configuration as that of the first embodiment. The section discriminating means 7 refers to the VQ codebook 6 to find a section between vowels and consonants in the input speech 1 and sends the vowel section speech 8 which is the input speech 1 in the vowel section to the vowel center extracting means 20. As an example of the section discrimination, the same method as in the first embodiment can be used. The vowel center extracting means 20 having received the vowel section voice 8 extracts a vowel center position 21 which is a reference point in the vowel section, and sends it to the vowel center waveform expanding / contracting means 22. The vowel center position 21 and its extraction method are the same as in the eighth embodiment.

The speech speed change rate 10 is the same as that of the first embodiment, and is a value given when the user changes the speech speed of the voice. Is entered. The vowel central part waveform expanding / contracting means 22 expands / contracts the vowel section voice 8 and the consonant section voice 11 in accordance with the speech speed change rate 10.
The upper vowel center interval is expanded and contracted at the same speech speed change rate 10. Note that the respective waveform expansion / contraction rates of the vowel section and the consonant section can be determined in the same manner as in the eighth embodiment.

Here, in a vowel section, expanding and contracting a transient part such as near the boundary with a consonant section and a stationary part such as the vowel center position 21 at the same expansion and contraction rate may impair the naturalness of speech. . Therefore, the waveform expansion and contraction rate of the vowel section
At the vowel center position 21 extracted by the vowel center extracting means 20, the maximum or the minimum is determined, and the waveform expansion / contraction rate of the consonant section is determined to be closer to the boundary with the consonant section. That is, when the expansion / contraction rate λ of the vowel section is λ> 1, the waveform is expanded, so that the waveform expansion / contraction rate is maximized at the vowel center position 21 and the waveform expansion / contraction of the consonant section smaller than λ at the boundary with the consonant section. Decrease to a value equal to the rate. When the expansion / contraction rate λ of a vowel section is 0 <λ <1, the waveform expansion / contraction is minimized at the vowel center position 21, and the waveform expansion / contraction of the consonant section is larger than λ at the boundary with the consonant section. Increase to the rate. The vowel center waveform expansion / contraction means 22 expands / contracts the input voice 1 in each section according to the waveform expansion / contraction rate. The method of expanding and contracting the waveform is the same as that in the first embodiment. In this way, the vowel center waveform expanding / contracting means 2
2 is output as synthesized speech 5.

As described above, according to the ninth embodiment, the expansion and contraction of the waveform can be automatically performed, and the vowel center position is considered so that the interval between the vowels of natural speech can be used. By weighting and expanding and contracting the waveform more strongly around the vowel center position than near the boundary with the consonant, it is possible to achieve a more natural change in utterance speed than converting all voice sections uniformly. The effect is obtained.

Embodiment 10 FIG. FIG. 11 is a block diagram showing a configuration example of a speech speed changing device according to Embodiment 10 of the present invention. The components corresponding to the respective components and data in the first to ninth embodiments are denoted by the same reference numerals as those in FIGS. 1 to 5 and FIGS. 7 to 10, and description thereof is omitted. In the figure, reference numeral 23 denotes pitch analysis means for analyzing the pitch frequency of the input voice 1 to obtain a pitch outline, and reference numeral 24 denotes a pitch outline extracted by the pitch analysis means 23. Reference numeral 25 denotes pitch outline reference position determining means for determining the maximum value and minimum value of the pitch outline 24 and determining the pitch outline reference position based on the maximum and minimum values. This is the pitch approximate reference position.

The waveform expansion / contraction means 9 defines a specific expansion / contraction ratio for each vowel section and consonant section, or each vowel section, consonant section and silent section, and when a speech speed change rate 10 is given,
In order to make the expansion and contraction rate of the pitch outline reference position interval constant,
A waveform expansion / contraction ratio for each section corresponding to the expansion / contraction ratio between a vowel section and a consonant section, or an expansion / contraction ratio between a vowel section, a consonant section, and a silent section is obtained. 1 to 5 and FIGS. 7 to 1 in that the waveform of the consonant section voice 11 is expanded and contracted.
This is different from the waveform expanding / contracting means 9 in each of the above embodiments, in which 0 is assigned the same reference numeral.

Next, the operation will be described. Again,
The input sound 1 is assumed to be sound data prepared in advance by a system for outputting sound. Pitch analysis means 23
When the input speech 1 is inputted, the pitch pitch in a short time is obtained, the pitch pitch is recorded in the time direction, and the pitch pitch is extracted and inputted to the pitch pitch reference position determining means 25. As a method of extracting the pitch frequency, there is a method of obtaining a cepstrum from a speech waveform for each fixed section length, obtaining a maximum value of the cepstrum in a high quefrency area, and obtaining a reciprocal thereof. The pitch outline reference position determination means 25 determines a pitch outline reference position 26 which is a reference position on the input pitch outline 24 and outputs it to the waveform expansion / contraction means 9. Here, this pitch outline reference position 26 is, for example,
Globally, this is a time position where the pitch outline 24 takes the maximum value and the minimum value.

The VQ codebook 6 is also used for performing vector quantization of audio data, and has the same configuration as that of the first embodiment. The section discriminating means 7 obtains a section between a vowel and a consonant in the input speech 1 and converts the vowel section speech 8 which is the input speech 1 of the vowel section and the consonant section speech 11 which is the input speech 1 of the consonant section into a waveform expanding / contracting means. Send to 9. In addition, as an example of this section determination,
The same method as in the first embodiment can be used.

The speech speed change rate 10 is the same as that in the first embodiment, and is a value given when the user changes the speech speed of the voice, and is input to the waveform expansion / contraction means 9 as an expansion / contraction ratio with respect to the original voice. . The waveform expansion / contraction means 9 calculates the speech speed change rate 10
The input voice 1 expands and contracts according to the pitch, and at this time, the interval between the pitch rough reference positions 26 on the input voice 1 determined by the pitch rough reference position determining means 25 expands and contracts at the same speech speed change rate 10. Linearly expand and contract as That is, when the speech speed change rate 10 is given by the user, the waveform stretching means 9 first
Determine the waveform expansion / contraction ratio for each vowel section and consonant section. In addition,
The method of determining the waveform expansion / contraction ratio is the same as that in the second embodiment. Next, the waveform expansion / contraction means 9 expands / contracts the input voice 1 in each of the vowel section and the consonant section at the determined waveform expansion / contraction rate. The method of expanding and contracting the waveform is the same as in the first embodiment. The voice whose speech speed has been changed by the waveform expansion / contraction means 9 is output as the synthesized voice 5.

As described above, according to the tenth embodiment, the expansion and contraction of the waveform can be automatically performed, and the expansion and contraction of the waveform is performed based on the reference position obtained from the pitch frequency of the natural voice. Waveform expansion and contraction can be performed without destroying the approximate pitch, and an effect that a natural change in speech rate can be realized as compared with the case where all voice sections are converted uniformly.

Embodiment 11 FIG. FIG. 12 is a block diagram showing a configuration example of a speech speed changing device according to Embodiment 11 of the present invention. The first to tenth embodiments are described.
In FIG. 7, the same reference numerals as those in FIGS. 1 to 5 and FIGS. 7 to 11 denote parts corresponding to the components and data, and a description thereof will be omitted. In the figure, reference numeral 27 denotes a provisional vowel section voice in the vowel section obtained from the input speech 1 by the section determination means 7, and reference numeral 28 denotes a provisional consonant section voice in the consonant section obtained from the input speech 1 by the section determination means 7. 29
Is a provisional consonant section voice 28 determined by the section determination means 7
Is separated into a part included in a vowel section and a consonant section based on the determination result of the pitch outline 24 by the pitch analysis means 23, and the consonant section is separated into the consonant section speech 11
And the part included in the vowel section and the provisional vowel section voice 2
7 is a pitch approximate shape dividing means which outputs the vowel section sound 7 to the waveform expanding / contracting means 9.

The waveform expanding / contracting means 9 determines a specific expansion / contraction ratio for each of the vowel section voice 8 and the consonant section voice 11 newly determined by the pitch outline dividing section 29, and is given a speech speed change rate 10. In this case, the same reference numerals are given to FIGS. 1 to 5 and FIGS. 7 to 11 in that the waveform expansion / contraction ratio for each section according to the expansion / contraction ratio is obtained and the waveform expansion / contraction is performed based on the obtained ratio. This is different from the waveform expanding / contracting means 9 in each of the above embodiments.

Next, the operation will be described. Again,
The input sound 1 is assumed to be sound data prepared in advance by a system for outputting sound. The VQ codebook 6 is also used for performing vector quantization of audio data, and has the same configuration as that of the first embodiment. The section discriminating means 7 obtains a section of a vowel and a consonant in the input speech 1, and sets the input speech 1 of the vowel section as a provisional vowel section speech 27, and the input speech 1 of the consonant section as a provisional consonant section speech 28, respectively, and roughly pitch-divides them. Output to means 29. As an example of the section determination, a method similar to that of the first embodiment can be used.

The pitch analyzing means 23 obtains a short-time pitch frequency of the input voice 1, extracts the pitch frequency recorded in the time direction as a pitch outline 24, and sends it to the pitch outline dividing means 29. . As an example of the pitch analysis, the same method as in the tenth embodiment can be used.

From the section discriminating means 7, the provisional vowel section voice 27
And the provisional consonant section voice 28 are input, and the pitch approximate shape dividing means 29 to which the pitch approximate form 24 is input from the pitch analysis means 23,
4 is divided. That is, if the pitch outline 24 of the provisional consonant section voice 28 is random or the slope of the general shape is different from that of the provisional vowel section voice 27 before and after, the section is regarded as the consonant section voice 11. The other section of the provisional consonant section voice 28 is provisional vowel section voice 2
7 together with vowel section voice 8. The vowel section speech 8 and the consonant section speech 11 newly discriminated by the pitch outline shape dividing section 29 are added to the waveform expansion / contraction means 9 in this manner.
Is entered.

When the speech speed changing rate 10 is given, the waveform expanding / contracting means 9 determines a specific expansion / contraction ratio of the vowel section voice 8 and the consonant section voice 11 from the pitch outline shape dividing section 29 based on the speech rate change rate 10. The waveform expansion / contraction ratio for each section according to the ratio is obtained, and the waveform is compressed or expanded based on the obtained ratio. Note that the speech speed change rate 10 is the same as that in the first embodiment, and is a value given when the user changes the speech speed of the voice, and is input to the waveform expansion / contraction means 9 as an expansion / contraction ratio with respect to the original voice. Also, the method of determining the waveform expansion / contraction ratio is the same as that of the second embodiment, and the method of waveform expansion / contraction is also the same as that of the first embodiment. In this way, the waveform stretching means 9
Is output as synthesized speech 5.

As described above, according to the eleventh embodiment, the expansion and contraction of the waveform can be automatically performed, and the approximate pitch is used as a criterion for determining the vowel consonant interval. The waveform can be expanded / contracted without destroying the utterance, and the effect that a natural change in the utterance speed can be realized as compared with the case where the entire voice section is converted uniformly.

The input voice 1 in the first to eleventh embodiments may be voice directly input by a narrator using a microphone instead of voice data prepared in advance in the system.

Further, as the waveform expansion / contraction rate in the vowel section in Embodiments 2 to 11, the speech speed change rate 10 input by the user may be used as it is, and the waveform expansion / contraction rate in the consonant section is set to “1”. It may be fixed.

In the first to third embodiments and the eighth to eleventh embodiments, the section discriminating unit 7 discriminates between a vowel section and a consonant section as well as the amplitude of the waveform of the input speech 1. May be provided with a function of discriminating a section in which the absolute value of all is less than an arbitrary value as a silent section. Embodiments 1 to 3, Embodiment 8, Embodiment 10, and Embodiment Waveform expansion / contraction means 9 according to mode 11
In addition, the vowel central part waveform expanding / contracting means 22 of the ninth embodiment can expand / contract a silent section, and when expanding / contracting, can set the expansion / contraction ratio of a silent section to be the largest than a vowel section and a consonant section. It may be.

Also, the section discriminating means 7 in the first to fourth embodiments and the eighth to eleventh embodiments uses the spectrum parameters analyzed in advance instead of analyzing the spectrum parameters each time. It may be stored as.

The section discriminating means 7 in the first to third embodiments and the eighth to eleventh embodiments uses, for example, 50 vowels and consonants in place of uniquely determining a vowel and a consonant using a single threshold value. At a vowel appearance probability near%, if the preceding and succeeding sections are vowels, the sections may be determined to be vowels, and section discrimination in consideration of continuity with the preceding and following sections may be performed.

Also, the section discriminating means 7 in the first to third embodiments and the eighth to eleventh embodiments does not uniquely determine vowels and consonants in each section, but uses May be determined based on the cumulative value of the probabilities.

Also, in the first to fourth embodiments, instead of the input speech 1 itself, a residual waveform and a spectrum parameter obtained by performing linear prediction analysis of the input speech are input as the input speech 1. In addition, the section discriminating means 7 in the first and third embodiments and the VQ waveform expanding / contracting means 14 in the fourth embodiment perform vector quantization and section discrimination with respect to spectral parameters. The waveform expanding / contracting means 9 in the third embodiment and the VQ waveform expanding / contracting means 14 in the fourth embodiment may expand and contract the residual waveform in a section corresponding to the spectrum parameter.

The eleventh embodiment is different from the tenth embodiment.
And the waveform expanding / contracting means 9 expands / contracts the interval between the pitch approximate reference positions obtained by the pitch approximate reference position determining means at the same speech rate change rate. May be set.

Further, the waveform expansion / contraction means 9 of the second, third, fifth, eighth, tenth, and eleventh embodiments, and the VQ waveform expansion / contraction means of the fourth embodiment. 14 and the vowel center waveform expanding / contracting means 22 of the ninth embodiment
May refer to a table that records the expansion and contraction ratio of each vowel and consonant corresponding to the mora speed, instead of defining the expansion and contraction ratio of vowels and consonants.

The rhythm perception point table 17 in the fifth and sixth embodiments and the rhythm perception point and expansion / contraction ratio table 19 in the seventh embodiment include not only the combination of phonemes but also the vowel pitch. The value of the rhythm perception point may be described in accordance with information indicating whether the subject corresponds to the height, the average speech speed, or the beginning, middle, or end of a sentence.

The labeling means 15 in the fifth to seventh embodiments may perform labeling by a simple voice recognition technique instead of manually creating phoneme labels.

[0095]

As described above, according to the present invention, the VQ in which the code vectors of the learning speech data are recorded together with the appearance probabilities of the vowels and consonants of each code vector.
By referring to the codebook, the input speech is vector-quantized to select a code vector, and a vowel section determined by the appearance probability of a vowel or consonant in the selected code vector is given to a speech given to the waveform stretching means. Since the waveform is expanded and contracted at the speed change rate, it is possible to automatically perform the waveform expansion and contraction for the vowel section, and the naturalness is deteriorated by uniformly changing all voice sections at the same expansion and contraction rate. Can be solved, and a speech speed changing device capable of realizing a good speech speed change can be obtained.

According to the present invention, a vowel section and a consonant section,
Alternatively, a specific expansion / contraction ratio is defined for each of the vowel section, consonant section, and silent section, and when a speech speed change rate is given, the waveform expansion / contraction of the vowel section, consonant section, and unvoiced section determined according to the specified expansion / contraction ratio. Since the waveform is expanded and contracted based on the rate, it is possible to automatically perform waveform expansion and contraction at separate expansion and contraction rates for vowel sections and consonant sections, and change all voice sections uniformly at the same expansion and contraction rate. This can solve the problem of the deterioration of the naturalness due to the above, and has an effect that it is possible to realize a good change in the utterance speed.

According to the present invention, when a speech speed change rate is given, an input / output ratio obtained by referring to an expansion / contraction ratio table which stores a specific expansion / contraction ratio defined for each type of phoneme and each phoneme category. Since the waveform expansion and contraction is performed based on the waveform expansion and contraction rate for each section corresponding to the expansion and contraction ratio for each phoneme or phoneme category of the voice, individual expansion and contraction rates are set for sections classified into vowel sections and consonant categories. Can automatically perform waveform expansion and contraction, and can eliminate the problem of naturalness degradation caused by changing all voice sections uniformly at the same expansion and contraction rate, and achieve a good change in speech rate There is an effect that can be.

According to the present invention, when the speech rate change rate is given, the input voice is vector-quantized by referring to the VQ codebook in which the code vector is recorded together with the expansion / contraction ratio specific to each code vector. A code vector is selected, and the waveform is expanded and contracted based on the waveform expansion and contraction ratio obtained for each section in accordance with the expansion and contraction ratio of each selected code vector. Waveform expansion and contraction at the expansion and contraction ratio can be performed easily and automatically, and the problem of deterioration in naturalness caused by changing all voice sections uniformly at the same expansion and contraction ratio can be solved, and good speech can be obtained. There is an effect that the speed can be changed.

According to the present invention, the rhythm perception point of the input voice is extracted from the phoneme label extracted by the labeling means from the input voice by referring to the rhythm perception point table storing the rhythm perception point unique to each phoneme sequence. When the speech rate change rate is given, the waveform expansion and contraction is performed based on the waveform expansion and contraction rate for each section obtained so that the expansion and contraction rate of the rhythm perception point interval becomes constant. And the basics of the natural sound time structure,
By expanding and contracting the waveform considering the rhythm perception point,
There is an effect that it is possible to realize a more natural change of the utterance speed than to perform uniform conversion with the same expansion / contraction ratio.

According to the present invention, a vowel section and a consonant section,
Alternatively, a specific expansion / contraction ratio is defined for each of the vowel section, the consonant section, and the silent section, and when a speech speed change rate is given, the specified expansion / contraction ratio is such that the expansion / contraction rate of the rhythm perception point interval becomes constant. The waveform expansion and contraction is performed based on the waveform expansion and contraction ratio of the vowel interval and consonant interval and the vowel interval, consonant interval, and silent interval determined according to the above. By performing waveform expansion and contraction in consideration of the rhythm perception point, which is the basis of the time structure of natural speech, it is possible to achieve a more natural speech rate change than a uniform conversion. is there.

According to the present invention, when the speech speed change rate is given, the rhythm perception is referred to by referring to the expansion / contraction ratio table storing the specific expansion / contraction ratio defined for each phoneme type or phoneme category. Since the waveform expansion and contraction is performed based on the waveform expansion and contraction ratio for each section corresponding to the expansion and contraction ratio for each phoneme or phoneme category of the input voice, determined so that the expansion and contraction ratio of the point interval is constant, the vowel Waveform expansion and contraction can be automatically performed at individual expansion and contraction rates for sections and sections classified by consonant categories, and waveform expansion and contraction taking into account rhythm perception points and phoneme types, which are the basics of the time structure of natural speech. Is performed, there is an effect that the utterance speed can be changed more naturally than the uniform conversion.

According to the present invention, the phoneme extracted by the labeling means from the input speech with reference to the rhythm perception point and expansion ratio table storing the rhythm perception point unique to the phoneme sequence and the expansion ratio specific to the phoneme sequence. The rhythm perception point of the input voice and the expansion / contraction ratio for each phoneme section are extracted from the label, and given the speech speed change rate, the waveform expansion / contraction rate for each phoneme section is determined so that the expansion / contraction rate of the rhythm perception point interval is constant. , The waveform expansion and contraction is performed, so that the waveform expansion and contraction can be automatically performed for the vowel section and the consonant section,
By performing waveform expansion and contraction in consideration of the rhythm perception point and phoneme type, which are the basics of the time structure of natural speech, there is an effect that it is possible to achieve a more natural change in speech rate than uniform conversion.

According to the present invention, the vowel center position of the vowel section voice extracted with reference to the VQ codebook is determined,
Given the speech rate change rate, the expansion and contraction rate of the interval between the vowel center positions was determined to be constant, according to the expansion and contraction ratio of the vowel section and consonant section, or the vowel section, consonant section and silent section. Since the waveform is expanded and contracted based on the waveform expansion and contraction rate for each section, it is possible to automatically perform waveform expansion and contraction, and consider the vowel center position so that the interval between vowels of natural speech can be used. By performing the waveform expansion / contraction, there is an effect that the utterance speed can be changed more naturally than the uniform conversion.

According to the present invention, the vowel center position of the vowel section voice extracted with reference to the VQ codebook is determined,
Given the speech rate change rate, based on the waveform expansion and contraction rate determined to approach the boundary with the consonant section at the maximum or minimum at the vowel center position within the vowel section and approach the waveform expansion and contraction rate of the consonant section Since the waveform is expanded and contracted in each section, the waveform can be automatically expanded and contracted, and the vowel center position is considered so that the interval between vowels of natural speech can be used. By performing weighted waveform expansion and contraction so as to expand and contract more strongly around the vowel center position than near the boundary, it is possible to obtain an effect that it is possible to achieve a more natural change in speech rate than with uniform conversion.

According to the present invention, the pitch outline reference position interval is determined from the maximum value and the minimum value of the pitch outline obtained by the pitch frequency analysis of the input voice, and when the speech speed change rate is given, the pitch is determined. The waveform expansion / contraction based on the waveform expansion / contraction rate for each section according to the expansion / contraction rate of the vowel section and consonant section, or the vowel section / consonant section / silent section determined so that the expansion / contraction rate of the outline reference position interval is constant. Is configured to perform
Waveform expansion and contraction can be performed automatically, and by performing waveform expansion and contraction based on the reference position obtained from the pitch frequency of natural voice, waveform expansion and contraction can be performed without destroying the approximate pitch. There is an effect that it is possible to realize a more natural change of the utterance speed than the transformation.

According to the present invention, a provisional consonant section voice is separated into a part including a vowel section and a consonant section by judging a pitch outline based on a result of pitch frequency analysis, and a vowel section and a consonant section are renewed. When the speech speed change rate is given, the waveform expansion / contraction is performed based on the waveform expansion / contraction rate for each section, which is obtained according to the expansion / contraction ratio uniquely determined for the vowel section and the consonant section. Therefore, waveform expansion and contraction can be automatically performed, and by using the pitch outline as a criterion for determining vowel consonant intervals, waveform expansion and contraction can be performed without destroying the pitch outline. There is an effect that it is possible to realize a more natural change of the utterance speed than the transformation.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration of a speech speed changing device according to a first embodiment of the present invention.

FIG. 2 is a block diagram showing a configuration of a speech speed changing device according to a second embodiment of the present invention.

FIG. 3 is a block diagram showing a configuration of a speech speed changing device according to a third embodiment of the present invention.

FIG. 4 is a block diagram showing a configuration of a speech speed changing device according to a fourth embodiment of the present invention.

FIG. 5 is a block diagram showing a configuration of a speech speed changing device according to a fifth embodiment of the present invention.

FIG. 6 is an explanatory diagram showing the contents of a rhythm perception table according to Embodiment 5 of the present invention.

FIG. 7 is a block diagram showing a configuration of a speech speed changing device according to a sixth embodiment of the present invention.

FIG. 8 is a block diagram showing a configuration of a speech speed changing device according to a seventh embodiment of the present invention.

FIG. 9 is a block diagram showing a configuration of a speech speed changing device according to an eighth embodiment of the present invention.

FIG. 10 is a block diagram showing a configuration of a speech speed changing device according to a ninth embodiment of the present invention.

FIG. 11 is a block diagram showing a configuration of a speech speed changing device according to a tenth embodiment of the present invention.

FIG. 12 is a block diagram showing a configuration of a speech speed changing device according to an eleventh embodiment of the present invention.

FIG. 13 is a block diagram showing a configuration of a conventional speech speed changing device.

[Explanation of symbols]

1 input voice, 5 synthesized voice, 6 VQ codebook,
7 section discriminating means, 8 vowel section voices, 9 waveform expansion / contraction means, 10 speech rate change rate, 11 consonant section voices, 12 expansion / contraction ratio table, 13 VQ codebook with expansion / contraction ratio, 1
4 VQ waveform expansion / contraction means, 15 labeling means, 16
Phoneme label, 17 rhythm perception point table, 18 rhythm perception point extraction means, 19 rhythm perception point and expansion / contraction ratio table, 20 vowel center extraction means, 21 vowel center position, 22
Vowel center waveform expansion / contraction means, 23 pitch analysis means, 2
4 pitch outline, 25 pitch outline reference position determining means,
26 pitch outline reference position, 27 provisional vowel section voice,
28 provisional consonant section voice, 29 pitch approximate shape dividing means.

Claims (12)

[Claims] 1. A vector quantization codebook for recording a code vector for learning speech data together with an appearance probability of a vowel or a consonant of each code vector, and a vector quantization codebook for input speech by referring to the vector quantization codebook. Segmentation means for selecting a code vector by performing conversion, based on the appearance probability of a vowel or consonant in the selected code vector, a section discriminating means for discriminating whether the section is a vowel section or a consonant section, A speech expander which performs waveform compression or expansion in accordance with a speech speed change rate given as an expansion / contraction ratio with respect to the original voice when the speech speed is changed in a section determined as a vowel section by the discriminator; Speed change device. 2. A waveform expansion / contraction means for defining a specific expansion / contraction ratio for each of a vowel section and a consonant section, or for each of a vowel section, a consonant section and a silent section, and using an expansion / contraction ratio with respect to an original voice for changing a speech speed. When a speech rate change rate is given, a waveform expansion / contraction rate for each section is obtained according to the expansion / contraction ratio between a vowel section and a consonant section, or a vowel section, a consonant section, and a silent section, and based on the waveform expansion / contraction rate. 2. The speech speed changing device according to claim 1, wherein the speech speed is changed by compressing or expanding the waveform. 3. An expansion / contraction ratio table for storing a specific expansion / contraction ratio defined for each type of phoneme and each category of phoneme, wherein the waveform expansion / contraction means uses an expansion / contraction ratio with respect to the original voice for changing the speech speed. When the speech rate change rate is given, the waveform expansion / contraction ratio is obtained by referring to the expansion / contraction ratio table, for each phoneme or each phoneme category of the input voice, for each section corresponding to the expansion / contraction ratio. 2. The speech speed changing device according to claim 1, wherein waveform compression or decompression is performed based on the following. 4. A vector quantization codebook with an expansion / contraction ratio for recording a code vector for learning speech data together with an expansion / contraction ratio specific to each of the code vectors; and a vector quantization codebook with an expansion / contraction ratio,
When a code rate is selected by performing vector quantization on an input voice and a speech speed change rate is given by an expansion / contraction ratio with respect to an original voice to change a speech speed, an expansion / contraction ratio for each of the selected code vectors is given. A speech rate changing device comprising: a vector quantization waveform expansion / contraction unit that obtains a waveform expansion / contraction rate for each section according to the waveform expansion / contraction based on the waveform expansion / contraction rate. 5. A labeling means for extracting a phoneme label of an input voice, a rhythm perception point table storing a rhythm perception point unique to a phoneme sequence of the input voice, and a phoneme with reference to the rhythm perception point table. A rhythm perception point extracting means for extracting a rhythm perception point of the input voice from a label; and a rhythm perception point extraction when a speech speed change rate based on an expansion / contraction ratio with respect to an original voice for changing a speech speed is given. Waveform expansion / contraction means for obtaining a waveform expansion / contraction rate for each section so that the expansion / contraction rate of the rhythm perception point interval at the rhythm perception point extracted by the means is constant, and performing waveform compression or expansion based on the waveform expansion / contraction rate. Speed change device. 6. A waveform expansion / contraction means for defining a specific expansion / contraction ratio for each of a vowel section and a consonant section, or for each of a vowel section, a consonant section and a silent section, and using an expansion / contraction ratio with respect to an original voice for changing a speech speed. When the speech rate change rate is given, the vowel section and the consonant section are set so that the expansion rate of the rhythm perception point interval becomes constant.
Alternatively, according to the expansion / contraction ratio of the vowel section, the consonant section, and the silent section, a waveform expansion / contraction rate is obtained for each section, and waveform compression or expansion is performed based on the waveform expansion / contraction rate. The speech speed changing device according to 5. 7. An expansion / contraction ratio table for storing a specific expansion / contraction ratio defined for each type of phoneme or each category of phoneme, wherein the waveform expansion / contraction means uses an expansion / contraction ratio with respect to the original voice for changing the speech speed. When the speech speed change rate is given, the expansion and contraction is performed for each phoneme type or phoneme category of the input voice so that the expansion and contraction rate of the rhythm perception point interval is constant with reference to the expansion and contraction ratio table. 6. The speech speed changing device according to claim 5, wherein a waveform expansion / contraction ratio for each section according to the ratio is obtained, and the waveform is compressed or expanded based on the waveform expansion / contraction ratio. 8. A labeling means for extracting a phoneme label of the input speech, a rhythm perception point unique to the phoneme sequence of the input speech, and a rhythm perception point and extension ratio table storing the extension ratio unique to the phoneme sequence. Rhythm perception point extracting means for extracting a rhythm perception point of the input voice and the expansion / contraction ratio of each of the phoneme sequences from the phoneme label with reference to the rhythm perception point and expansion / contraction ratio table; For each phoneme section corresponding to the expansion / contraction ratio of each of the phoneme sequences, when the speech speed change rate based on the expansion / contraction ratio with respect to the original voice is given, the expansion / contraction ratio of the interval between the rhythm perception points is constant. A speech speed changing device comprising: a waveform expansion / contraction unit that obtains a waveform expansion / contraction ratio and performs waveform compression or expansion based on the waveform expansion / contraction ratio. 9. A vector quantization codebook for recording a code vector of learning speech data together with an appearance probability of a vowel or a consonant of each code vector; and a vector quantization codebook for input speech by referring to the vector quantization codebook. Segmentation means for selecting a code vector by performing the conversion, and based on the appearance probability of a vowel or consonant in the selected code vector, a section discriminating means for determining whether the section is a vowel section or a consonant section, A vowel center extracting means for determining a vowel center position of the section, and a vowel section and a consonant section, or a vowel section, a consonant section, and an original expansion / contraction ratio for each silent section, for changing the speech rate, When a speech speed change rate based on the expansion / contraction ratio for voice is given, the expansion / contraction ratio at the interval between the vowel center positions becomes constant. In accordance with the expansion / contraction ratio of a vowel section and a consonant section, or a vowel section, a consonant section and a silent section, a waveform expansion / contraction rate for each section is obtained, and waveform compression or expansion is performed based on the waveform expansion / contraction rate. A speech speed changing device comprising a waveform expanding / contracting means. 10. A vector quantization codebook for recording a code vector for learning speech data together with an appearance probability of a vowel or a consonant of each code vector, and a vector quantization codebook for input speech by referring to the vector quantization codebook. Segmentation means for selecting a code vector by performing conversion, and based on the appearance probability of a vowel or consonant in the selected code vector, a section discriminating means for determining whether the section is a vowel section or a consonant section, A vowel center extracting means for determining a vowel center position of the vowel section determined by the discriminating means; and a vowel when a speech speed change rate based on an expansion / contraction ratio with respect to the original voice for changing a speech speed is given. The waveform expansion / contraction ratio in the section becomes maximum or minimum at the vowel center position, and approaches the boundary with the consonant section. A speech speed changing device comprising: a vowel center waveform expanding / contracting means for determining a waveform expansion / contraction rate of a consonant section to be close to the waveform expansion / contraction section and compressing or expanding the waveform of each section based on the waveform expansion / contraction rate. 11. A pitch analysis means for analyzing a pitch frequency of an input voice to obtain a pitch outline, and obtaining a maximum value and a minimum value of the pitch outline obtained by the pitch analysis means to obtain a pitch outline reference position. The pitch expansion and contraction reference position determination means is provided, and the waveform expansion / contraction means defines a specific expansion / contraction ratio for each vowel section and consonant section, or each vowel section, consonant section, and silent section, and changes the speech rate. Of the vowel section and the consonant section, or a vowel, so that when a speech speed change rate based on the expansion / contraction ratio with respect to the original voice is given, the expansion / contraction rate of the interval between the pitch rough reference positions is constant. 2. The speech speed according to claim 1, wherein a waveform expansion / contraction ratio is determined for each section according to the expansion / contraction ratio of the section, the consonant section, and the silent section, and the waveform is compressed or expanded based on the waveform expansion / contraction rate. Change device. 12. A pitch analyzing means for analyzing a pitch frequency of an input voice to obtain a pitch outline, and a provisional consonant section voice determined as a consonant section by a section determining means is converted into a pitch rough shape by said pitch analyzing means. According to the determination result, a part included in the vowel section and a consonant section are separated, and only the separated consonant section is output as a consonant section voice. A pitch rough dividing means for outputting the provisional vowel section voice determined as the section as a vowel section voice, wherein the waveform expanding / contracting means is specific to the section which is newly determined as a vowel section and a consonant section by the pitch rough dividing section. In order to determine the expansion / contraction ratio and change the utterance speed, when a speech speed change ratio based on the expansion / contraction ratio with respect to the original voice is given, a waveform for each section corresponding to the expansion / contraction ratio is given. Seeking shrinkage, speech speed changing device according to claim 1, characterized in that for performing waveform compression or decompression on the basis of the waveform scaling factor.