JP2011033789A - Adaptive speech-rate conversion device and program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a speech-rate conversion device which performs adaptive speech-rate conversion of an input signal, and to provide a program. <P>SOLUTION: The speech-rate conversion device 1 includes: a physical index calculating part 2 which, for each segment obtained by dividing an input signal into unit time, calculates a physical index of the input signal; and a speech-rate conversion factor determination part 3 for performing speech-rate conversion by determining speech-rate to be designated to each segment of the input signal, according to the physical index calculated by the physical index calculation part 2. The speech-rate conversion device 1 performs speech-rate conversion by determining a speech-rate conversion factor αn to be designated to each segment of the input signal, by using one or more "physical indexes" among a voicing degree Un expressing a relative maximum value obtained by autocorrelation per unit time in the input signal, an irregularity degree Sn expressing change tendency of loci of fundamental frequency and pseudo-fundamental frequency per unit time in the input signal; and a division band power ratio En expressing a ratio between a low region power component and a high region power component, which are divided in bands per unit time in the input signal. <P>COPYRIGHT: (C)2011,JPO&INPIT

Description

  The present invention relates to a speech speed conversion technique for converting the speech speed of an input signal, and more particularly to a speech speed conversion apparatus and program for adaptively converting the speech speed of an input signal.

  Conventionally, several techniques for adaptively converting the speech speed of an input signal have been proposed (see, for example, Patent Documents 1 to 6).

  The objective common to adaptive speech speed conversion was given an arbitrary playback speed conversion factor α [double speed] such as 1x speed (playback in real time) and 2x speed (playback in half the real time). In some cases, the speed of the continuous input signal is partially changed with a magnification larger or smaller than the reproduction speed conversion magnification α, instead of changing the speed with a uniform reproduction speed conversion magnification α over the entire input signal. Thus, speech speed converted speech is generated. Therefore, the entire input signal can be adjusted so that it can be played back in the same time as the speech rate conversion at a uniform magnification α, and for those who hear continuous input signals, at a uniform magnification α. "Slower and easier to hear" than when the speech speed is converted.

  The technique of Patent Document 1 generally includes the following three components. (1) Of the input signal, the speaking speed is reduced at a high fundamental frequency, and the speaking speed is increased at a low fundamental frequency. (2) Using the section of the input signal uttered at a breath as a unit, the speech speed is slowed at the start point of the voice and gradually increased in accordance with the change of the fundamental frequency toward the end point of the voice. (3) A silent section between adjacent sections uttered with a breath of the input signal is shortened within a range that does not cause a sense of incongruity.

  In the technique of Patent Document 2, a silent period longer than a certain length of an input signal is set as a pause period, and the speech (phrase) period sandwiched between the pause periods is reduced at the start point and constant. The speech speed is increased over time based on a predetermined decrease function. When the speech speed is reduced after a certain period of time, the speech speed is adjusted in consideration of the magnitude relationship of the maximum fundamental frequency in each speech (phrase) section. It changes the rate of loosening.

  Further, the technology of Patent Document 3 allows the speech speed control to be shortened within a range in which there is no sense of incongruity even for a short silent section in a voice section sandwiched between pause sections, and is divided into blocks. When a certain block of speech speed converted speech matches the time assumed when the speech speed conversion is performed at a uniform playback speed conversion magnification α with respect to the entire input signal, or there is almost no delay In addition, the speed of the next block is set to be as slow as possible. In particular, the technique of Patent Document 3 is delayed with respect to the time that is assumed when a certain block of speech that has been subjected to speech speed conversion is subjected to speech speed conversion with a uniform reproduction speed conversion factor α for the entire input signal. The higher the degree, the higher the degree of slowing down the subsequent speech speed, and the time assumed when each speech speed converted speech block is converted at a uniform playback speed conversion factor α. Control is performed so as not to deviate as much as possible.

  The techniques of Patent Documents 4 and 5 are based on dividing an input signal into a voice section and a silent section, slowing down the speech speed of the voice section, and shortening the silent section. Here, since the output signal length is increased with respect to the input signal length per unit time by slowing down the speech speed of the speech section, it is necessary to temporarily store the speech after the speech speed conversion in the memory. Therefore, in the techniques of Patent Documents 4 and 5, the speech speed is gradually increased for each voice section or the deletion amount of the silent section is increased according to the remaining amount available for the upper limit amount of the memory. Adjust the overall audio duration.

  In the technique of Patent Document 6, the speech speed of an input signal divided every predetermined period is determined by a coefficient that is inversely proportional to the nth power of the numerical value of the audio data size (power) and pitch (speech height) for each predetermined period. To decide.

Japanese Patent No. 3249567 Japanese Patent No. 3219892 Japanese Patent No. 3220043 Japanese Patent No. 3357742 Japanese Patent No. 3373933 Specification Japanese Patent No. 36199946

  What is common to the techniques of Patent Documents 1 to 5 is that an input signal is divided into a voice section with voice and a silent section without voice, and the duration is partially expanded or contracted based on some information in the voice section. At the same time, the length of the silent section is shortened to comprehensively adjust the overall voice time length. However, these techniques function effectively in the case of an input signal only of a human voice, but in the case of an input signal in which background sound and sound such as a broadcast program are mixed, a section of only the background sound is present. Whether it is determined as a “silent section” or a “voice section” is not guaranteed. When an erroneous determination occurs, the effect of shortening the “silent period” cannot be obtained, and as a result, the expansion rate of the voice period cannot be increased, and the speech speed converted voice is not easy to hear.

  Regarding the technique of Patent Document 6, the magnitude (power) of the input signal can be obtained in all sections of the input signal, but the fundamental frequency of the input signal is the “voiced sound section” in which the human vocal cords vibrate. However, it can only be obtained correctly. Therefore, in the technique of Patent Document 6, in the case of an input signal in which background sound and sound are mixed, power is large in the section of only the background sound, and the fundamental frequency cannot be obtained correctly. Although it is desired to increase the speech speed in the section of only background sound that is not speech, since the power is large, the speech speed can be controlled to be relaxed.

  As described above, in the conventional technique, for an input signal in which the determination of a voice section with voice and a silent section without voice is not performed accurately, as in the case of an input signal in which background sound and voice are mixed. Therefore, there is a drawback that adaptive speech speed conversion does not work as expected.

  An object of the present invention is to provide a speech speed conversion apparatus and program capable of stably performing adaptive speech speed conversion even in the case of an input signal in which background sound and sound are mixed.

  In order to solve the above-described problem, the speech speed conversion apparatus of the present invention is a speech speed conversion apparatus that performs adaptive speech speed conversion of an input signal, and for each segment obtained by dividing the input signal for each unit time, A physical index calculation unit for calculating the physical index of the input signal, and a speech speed conversion by determining a rate of speech speed conversion to be specified for each segment of the input signal according to the physical index calculated by the physical index calculation unit And a fast conversion magnification determination unit.

  Further, in the speech rate conversion apparatus of the present invention, the physical index calculation unit includes a voiced degree calculation unit that calculates a voiced degree representing a relative maximum value obtained by autocorrelation per unit time in an input signal as the physical index. It is characterized by providing.

  Further, in the speech speed conversion device of the present invention, the physical index calculation unit calculates a degree of unevenness representing a change tendency of a trajectory of a fundamental frequency and a pseudo fundamental frequency per unit time in an input signal as the physical index. A calculation unit is provided.

  Further, in the speech speed conversion device of the present invention, the physical index calculation unit calculates a ratio between a low-frequency power component and a high-frequency power component obtained by dividing a band per unit time in the input signal as the physical index. A division band power ratio calculation unit is provided.

  Further, in the speech speed conversion device of the present invention, the speech speed conversion magnification determination unit determines the determined speech speed conversion magnification when the playback speed conversion magnification to be speed converted is given to the entire input signal. A speech speed conversion magnification fine adjustment unit that finely adjusts the reproduction speed conversion magnification to suit the reproduction speed conversion magnification is provided.

  Further, in the speech speed conversion device of the present invention, the speech speed conversion magnification determination unit uses one or more physical indices of the voiced degree, the unevenness degree, and the divided band power ratio of the input signal. A speech speed conversion magnification fine adjustment unit for determining a speech speed conversion magnification to be specified for each segment is provided.

  Further, in the speech speed conversion device according to the present invention, the speech speed conversion magnification fine adjustment unit may include one or more of the voicing degree, the unevenness degree, and the divided band power ratio according to the type of the input signal. It is characterized by performing allocation allocation of speech speed conversion magnification based on the index.

  In addition, the present invention can be obtained by autocorrelation per unit time in an input signal for each segment obtained by dividing the input signal for each unit time in a computer configured as a speech rate conversion apparatus that performs adaptive speech speed conversion of the input signal. The voicedness representing the relative maximum value, the unevenness representing the change tendency of the trajectory of the fundamental frequency and the pseudo fundamental frequency per unit time in the input signal, and the low frequency side power component and the high frequency band divided per unit time in the input signal A step of calculating one or more physical indices out of the divided band power ratios representing a ratio to the band-side power component, and a speech speed to be specified for each segment of the input signal according to the physical indices calculated by the steps It is also characterized as a program for executing the step of converting the speech speed by determining the conversion magnification.

  According to the present invention, since adaptive speech speed conversion is performed based on the physical index of the input signal, in the conventional technology, the determination of “speech interval” and “silent interval” is performed for an input signal in which background sound and sound are mixed. Even when it is not performed accurately, adaptive speech speed conversion can be performed, that is, the effect of giving a stable and slow feeling to an input signal in which background sound and sound are mixed, is enhanced, Adaptive speech speed conversion that makes natural sound is possible.

1 is a block diagram of a speech speed converting apparatus according to an embodiment of the present invention. It is a flowchart which shows operation | movement of the speech speed converter of one Example by this invention. It is operation | movement explanatory drawing of the voicing degree calculation part in the speech rate converter of one Example by this invention, (a) is a figure which shows the window function with respect to the speech waveform of an input signal, (b) is based on an autocorrelation function. It is a figure which shows calculation of voicedness. It is operation | movement explanatory drawing of the fundamental frequency and pseudo | simulation fundamental frequency unevenness | corrugation calculation part in the speech rate converter of one Example by this invention.

  A speech speed converting apparatus according to an embodiment of the present invention will be described below. The speech speed conversion apparatus according to the present embodiment is an embodiment that includes all the elements according to the present invention, but it will be apparent from the following description that there are some modifications.

[Device configuration]
FIG. 1 shows a block diagram of a speech speed converting apparatus according to an embodiment of the present invention. The speech speed converting apparatus 1 according to this embodiment includes a physical index calculation unit 2 that calculates a physical index of an input signal for each segment obtained by dividing the input signal for each unit time, and a physical index calculated by the physical index calculation unit 2. And a speech speed conversion magnification determining unit 3 that determines the speech speed conversion magnification αn to be specified for each segment of the input signal and performs speech speed conversion, thereby performing adaptive speech speed conversion of the input signal. . Note that n is an integer value indicating the position of the input signal when the input signal is divided from the beginning, for example, every 5 ms. Hereinafter, this segmentation interval will be described as 5 ms as a segment (section) per unit time.

  The physical index calculator 2 includes a voicedness calculator 100, a fundamental frequency / pseudo fundamental frequency unevenness calculator 200, an unevenness calculator 210, a frequency band / power calculator 300, and a divided band power ratio calculator 310. Is provided. The fundamental frequency / pseudo fundamental frequency unevenness calculation unit 200 includes a fundamental frequency extraction unit 202, a pseudo fundamental frequency calculation unit 204, and a fundamental frequency locus coupling unit 206. The frequency band / power calculation unit 300 includes a spectrum calculation unit 302, a band division unit 304, and a power calculation unit 306.

  The speech speed conversion magnification determination unit 3 includes a first speech speed conversion magnification designation unit (speech speed conversion magnification designation unit a) 120, a second speech speed conversion magnification designation unit (speech speed conversion magnification designation unit b) 220, A three-speech speed conversion magnification designation unit (speech speed conversion magnification designation unit c) 320 and a speech speed conversion magnification fine adjustment unit 400 are provided.

  The speech speed conversion apparatus 1 according to the present embodiment generally includes a “voicedness” Un representing a relative maximum value (relative maximum value) obtained by autocorrelation per unit time in the input signal, and a unit in the input signal. “Roughness degree” Sn representing the change tendency of the trajectory of the fundamental frequency and the pseudo fundamental frequency per time, and the ratio of the low-frequency side power component and the high-frequency side power component divided into bands per unit time in the input signal Using one or more “physical indices” of “divided band power ratio” En, the speech speed conversion magnification αn to be specified for each segment of the input signal is determined, the speech speed conversion is performed, and the speech speed converted output Generate and output a signal.

  Hereinafter, the determination of the speech rate conversion magnification of each section with respect to the input signal based on the “voiced” Un, “unevenness” Sn, and “divided band power ratio” En will be described in order. The “speech rate conversion magnification” described below corresponds to the reciprocal of the temporal expansion / contraction rate with respect to the speech section per unit time of the input signal.

  First, the speech speed conversion magnification determination based on the voiced degree Un will be described with reference to FIGS. FIG. 3 is an explanatory diagram of the operation of the voicing degree calculation unit 100 in the speech rate conversion apparatus 1 according to an embodiment of the present invention. FIG. 3A is a diagram showing a window function with respect to the waveform of the input signal, and FIG. It is a figure which shows calculation of the voicing degree by an autocorrelation function.

(Determination of speech rate conversion magnification based on voicedness)
The voicing degree calculation unit 100, for each segment divided every predetermined unit time from the waveform of the input signal, the time delay amount τ from the start of each segment and the autocorrelation function R (τ) at τ = 0. , A maximum value R (τ) max of the autocorrelation function R (τ) when τ> 0, and a weight W (τ determined in advance according to the value of τ giving R (τ) max. ) Is calculated as “physical index”. More specifically, the voicing degree calculation unit 100, for example, the waveform of the input signal for each segment (n-th section) obtained by dividing an input signal in which broadcast sound and background sound are mixed for each predetermined unit time. Then, the autocorrelation function R (τ) is obtained from the above, then the maximum value R (τ) max of the autocorrelation function R (τ) when τ> 0 is detected, and further the autocorrelation function R (τ) when τ = 0. The reference value R (0) is calculated, and the voicedness Un = W (τ) · R (τ) max is calculated using a weight W (τ) predetermined according to the value of τ giving R (τ) max. Find / R (0). Here, τ is the amount of time delay from the start of the nth section.

  For example, as shown in FIG. 3A, the input signal waveform x (k) is weighted by a window function (Humming window h (k)), and the input signal weighted waveform x ′ (k) is obtained. Extract. Next, as shown in FIG. 3B, an autocorrelation function R (τ) with respect to the waveform x ′ (k) of the input signal in the section τ is calculated. Thus, using the reference value R (0) and the weight W (τ) of the autocorrelation function R (τ) at τ = 0, the voicedness Un = W (τ) · R (τ) max / R (0) Can be requested.

  Therefore, the voiced degree Un represents a relative maximum value obtained by autocorrelation per unit time in the input signal. Therefore, as another calculation method, the input signal in the unit time (5 ms in this example) in the input signal is used. It is also possible to count the number of zero crossings of the waveform of, and the reciprocal of this count value is the voicedness Un.

Next, the first speech speed conversion magnification designation unit (speech speed conversion magnification designation unit a) 120 determines the speech speed when the value of the voiced degree Un is greater than a predetermined threshold according to the value of the voiced degree Un. Loosen, when the value of Yukoedo Un is equal to or lower than a predetermined threshold value so as to speed up the speech speed, determines the speech speed conversion ratio .alpha.a _n defining a scaling factor for the speech section per unit time of the input signal.

For example, when the autocorrelation function R (τ) is calculated as the value of the voiced degree Un, the voiced degree Un is in the range of about −0.2 to 1.2 with respect to most of the input signals. It turns out that it takes a value. Therefore, a half-value Ub (for example, Ub = 0.7) of a fluctuation range that is assumed to be a value of the voiced degree Un is defined, and a reference value Ua (the value of the voiced degree Un corresponds to the median value of this range) For example, if Ua = 0.5), the speech speed is slowed down (αa _n <1.0), and if the value of voicedness Un is equal to or less than a predetermined threshold Ua (for example, Ua = 0.5), the speech speed is reduced. (Αa _n ≧ 1.0), it can be expressed as equation (1).

Here, K is a constant that serves as a reference value for determining a width for slowing or speeding up the speech speed. For example, K = 1.4 as a constant that gives a scaling rate corresponding to a predetermined slowest speech speed conversion magnification. It can be. Also, Ra is a contribution to the speech speed conversion ratio .alpha.a _n specified by Yukoedo Un "Yukoedo" Un physical indicators, "asperity" Sn, and, based on the "split-band power ratio" En Represents the degree of assignment when determining the assignment of the speech rate conversion magnification.

As described above, by "Yukoedo" Un physical indicators, it is possible to determine the speech rate conversion ratio .alpha.a _n for each unit of the input signal time.

  Next, determination of the speech speed conversion magnification based on the degree of unevenness will be described with reference to FIGS. FIG. 4 is a diagram for explaining the operation of the fundamental frequency / pseudo fundamental frequency unevenness calculating unit 200 in the speech speed converting apparatus according to the embodiment of the present invention.

(Determination of speech speed conversion magnification by unevenness)
The fundamental frequency extraction unit 202 is a region where the fundamental frequency value extracted every unit time (5 ms in this example) is stably and substantially continuously changed within a predetermined change range with respect to the input signal. In addition to determining as “stable section”, the area between each stable section is determined as “unstable section”, and the fundamental frequency in each stable section is specified, and the fundamental frequency in each stable section is even smoother. The trajectory consisting of the fundamental frequency of each stable section is smoothed so that For this smoothing, it is preferable to use a low-pass filter having a cutoff frequency of about 3 to 6 Hz. It should be noted that any known technique can be used as the technique for extracting the fundamental frequency per unit time (in this example, 5 ms) (see, for example, Japanese Patent No. 3219868).

  Further, the fundamental frequency extraction unit 202 outputs the information on the stable interval and / or the unstable interval and the value of the fundamental frequency of the smoothed locus in the stable interval to the pseudo fundamental frequency calculation unit 204 and the fundamental frequency locus coupling unit 206. To do.

  Note that the fundamental frequency extraction unit 202 rejects all the fundamental frequency values in the “unstable section”, which means that the extracted fundamental frequency values are not stable, discontinuous, and change rapidly.

  The pseudo fundamental frequency calculation unit 204 uses each value of the smoothed trajectory fundamental frequency of the stable interval supplied from the fundamental frequency extraction unit 202 and interpolates with an interpolation function such as a spline function to simulate the pseudo frequency in the unstable interval. A basic fundamental frequency (pseudo fundamental frequency) value is determined and output to the fundamental frequency locus coupling unit 206. Since the start and end portions of the input signal to be processed are often not speech intervals, they become unstable intervals for obtaining a pseudo fundamental frequency. In that case, when interpolating these sections with a spline function, a prescribed value (for example, 30 Hz, which is almost impossible as the fundamental frequency of speech) is set at the start point or end point, and one stable section is set. Is interpolated with a spline function using each value of the fundamental frequency.

  The fundamental frequency trajectory linking unit 206 is a value of the fundamental frequency of a smoothed trajectory supplied from the fundamental frequency extraction unit 202 and a value of the pseudo fundamental frequency of the unstable interval supplied from the pseudo fundamental frequency calculation unit 204. To obtain a continuous trajectory (hereinafter referred to as “fundamental frequency trajectory”) consisting of a fundamental frequency and a pseudo fundamental frequency in all sections (in this example, every 5 ms) of the input signal to be processed. Each value of the basic frequency for each unit time constituting the frequency locus is sent to the unevenness degree calculation unit 210.

  The unevenness degree calculation unit 210 has a fundamental frequency value in a stable section of a region where a fundamental frequency value extracted every unit time stably changes within a predetermined variation range with respect to an input signal. For the smoothed fundamental frequency trajectory defined by the value of the pseudo fundamental frequency of the unstable section in the area between the stable sections, the degree of unevenness representing the change tendency of the fundamental frequency trace is defined as a “physical index”. calculate. More specifically, the unevenness degree calculation unit 210 sets a value (for example, a value before a predetermined time) with respect to a certain value Pn of the fundamental frequency for each unit time (in this example, every 5 ms) constituting the fundamental frequency locus. , The value Pn at time −30 ms) P1 and the value after a predetermined time (for example, the value Pn at time +30 ms) P2 are sampled, and the front difference value (Pn−P1) and the rear difference value (Pn) are sampled. -P2) is averaged over all intervals of the input signal to be processed, and each of the average values in all intervals is normalized by dividing by the maximum value of these average values. Each of the average values is calculated as “concave / convex degree” Sn representing the change tendency of the fundamental frequency locus, and the calculated uneven degree Sn is sent to the second speech speed conversion magnification designation unit (speech speed conversion magnification designation unit b) 220.

  For example, the unevenness level Sn is a value close to 0 in a section where the fundamental frequency locus is flat or monotonically increasing or decreasing. Since all the irregularities Sn are normalized using a value having the largest absolute value, each value of the irregularity Sn representing the change tendency of the fundamental frequency locus is −1 to 1.

The second speech speed conversion magnification designation unit (speech speed conversion magnification designation unit b) 220 corresponds to each value of the unevenness degree Sn of unit time (in this example, every 5 ms) supplied from the unevenness degree calculation unit 210. Talk that regulates the rate of expansion / contraction of the input signal per unit time so that the speech speed is slowed when the unevenness Sn value is positive and the speech speed is increased when the unevenness Sn value is negative The fast conversion magnification αb _n is determined. That is, in this fundamental frequency locus, the speaking speed is slowed down in a mountain-shaped convex portion (maximum), and the speaking speed is increased in a valley-shaped portion (minimum).

For example, the speech rate conversion magnification αb _n can be expressed as in Expression (2).

Here, K is the same as in equation (1), Rb is the contribution rate to the speech rate conversion magnification αb _n specified by the degree of unevenness Sn, and the physical index “voiced” Un, “unevenness” Sn, Also, it represents the degree of assignment when determining the assignment of the speech rate conversion magnification based on the “divided band power ratio” En.

As described above, the speech rate conversion magnification αb _n can be determined for each unit time of the input signal based on the “concave / convex degree” Sn of the physical index.

  Next, determination of the speech rate conversion magnification based on the divided band power ratio En in the frequency band division / power calculation unit 300 will be described.

(Determination of speech rate conversion ratio based on divided band power ratio)
The spectrum calculation unit 302 converts the time domain waveform into the frequency domain by FFT (Fast Fourier transform) or the like for each unit time (5 ms in this example) with respect to the input signal, and converts the logarithmized power spectrum of each frequency. Obtained by the dB value and sends it to the band dividing unit 304.

  The band division unit 304 divides the logarithmic power spectrum supplied from the spectrum calculation unit 302 into a plurality of predetermined frequency bands, and sends the logarithmized power spectrum value of each frequency band divided to the power calculation unit 306. To do. For example, when dividing into five, it can divide into B1: 0-300 Hz, B2: 300-1500 Hz, B3: 1500-3000 Hz, B4: 3000-8000 Hz, B5: 8000 Hz or more. Note that it may be simply divided into two.

  The power calculation unit 306 normalizes each band in the lower band and the higher band arbitrarily selected in advance for the logarithmized power spectrum value of each frequency band divided from the band supplied from the band dividing unit 304 The power component is obtained, and the normalized low frequency side power component and high frequency side power component are sent to the divided band power ratio calculation unit 310. The normalized power component can be obtained by summing the values of each power spectrum by the number of power spectra included in each band on the low frequency side and high frequency side, and then dividing by the number. For example, when dividing into five as described above, it is assumed that B2 is selected in advance as the lower band and B4 is selected as the higher band. In this case, the normalized power components of the lower band B 2 and the higher band B 4 are output to the divided band power ratio calculation unit 310.

  The divided band power ratio calculation unit 310 is configured to arbitrarily select a ratio between a low band side power component and a high band side power component for a logarithmized power spectrum value of each frequency band divided into bands for the input signal ( The divided band power ratio En) is calculated as a “physical index”. More specifically, the divided band power ratio calculation unit 310 calculates the ratio of the normalized low frequency side power component and high frequency side power component supplied from the power calculation unit 306 (division band power ratio En). Then, it is sent to the third speech speed conversion magnification designation section (speech speed conversion magnification designation section c) 320. When the normalized low frequency side power component and high frequency side power component supplied from the power calculation unit 306 are already represented by logarithmic values (dB), the difference between these logarithmic values is obtained. The subband power ratio En can be obtained by subtracting the normalized power on the high frequency side from the normalized power on the low frequency side.

The third speech speed conversion magnification designation unit (speech speed conversion magnification designation unit c) 320 determines the speech speed when the value of the divided band power ratio En is larger than a predetermined threshold according to the value of the divided band power ratio En. And the speech rate conversion magnification αc _n that defines the expansion / contraction rate for the speech interval per unit time of the input signal is determined so that the speech rate is increased when the value of the divided band power ratio En becomes a predetermined threshold value or less. To do.

  For example, it has been found that the divided band power ratio En takes a value in the range of about 10 [dB] to 40 [dB] with respect to most of the input signals. Therefore, the third speech speed conversion magnification designation unit (speech speed conversion magnification designation unit c) 320 defines a half-value Eb (for example, Eb = 15) of the fluctuation range that can be assumed to be taken by the input signal. By defining a reference value Ea (for example, Ea = 25) corresponding to the median value of the range, when the divided band power ratio En is larger than Ea = 25 [dB], the speech speed is reduced, and the divided band power ratio En If Ea = 25 [dB] or less, it is defined as in equation (3) so as to increase the speech speed.

Here, K is the same as equation (1), Rc is the contribution to the speech speed conversion ratio .alpha.c _n specified by the split-band power ratio En, "Yukoedo" Un of the above-mentioned physical indicators, "uneven The degree of assignment when the assignment of the speech rate conversion magnification is determined based on the degree “Sn” and the “divided band power ratio” En.

As described above, the "split-band power ratio" En physical indicators, it is possible to determine the speech rate conversion ratio .alpha.c _n for each unit of the input signal time.

  As described above, the speech speed converting apparatus 1 according to the present embodiment uses one or more “physical indices” of “voicedness” Un, “unevenness” Sn, and “divided band power ratio” En. Then, the speech speed conversion is performed by determining the speech speed conversion magnification of each section with respect to the input signal.

(Advanced adaptive speech speed conversion)
In order to realize adaptive speech speed conversion, the speech speed converting apparatus 1 according to the present embodiment has one or more of “voiced degree” Un, “unevenness degree” Sn, and “divided band power ratio” En. The speech speed conversion magnification fine adjustment unit 400 for determining the speech speed conversion magnification of each section with respect to the input signal using the “physical index” of FIG.

Speech speed conversion ratio fine adjustment section 400, a speech speed conversion ratio .alpha.a _n determined by the first episode speed conversion magnification specifying unit (speech speed conversion ratio specifying section a) 120, Episode 2 speed conversion magnification specifying unit (speech rate a speech speed conversion ratio .alpha.b _n determined by the conversion magnification specifying unit b) 220, the determined speech speed conversion ratio .alpha.c _n and were inputted by the third episode speed conversion magnification specifying unit (speech speed conversion ratio specifying section c) 320 , the type of input signal to be preset (e.g., the genre of the program) by adding the speech speed conversion ratio was distributed allocation in the allocation ratio corresponding to .alpha.a _n, .alpha.b _n, the value of .alpha.c _n, obtained by adding performing speech speed conversion per unit time in the input signal by the speech speed conversion ratio _{αn (αn = αa n + αb} n + αc n) to be.

For example, each value of the contribution ratios Ra, Rb, and Rc in the equations (1) to (3) is changed in order to allocate and allocate the values of the respective speech speed conversion magnifications αa _n , αb _n , and αc _n . For example, when the input signal is a broadcast program audio, the allocation allocation is performed by changing the distribution ratio of the contribution ratios Ra, Rb, Rc according to the genre of the broadcast program (news, documentary, drama, variety, rakugo, comic story, etc.). It can be carried out. This makes it possible to perform adaptive speech speed conversion with higher quality in terms of ease of listening and naturalness according to the genre of the broadcast program. For example, when the input signal is news speech, the contribution rate Ra = 0.5, Rb = 0.3, Rc = 0.2, and when the input signal is rakugo or comic, the contribution rate Ra = 0.2. Rb = 0.6, Rc = 0.2, etc. can be assigned.

In addition, when the speech speed is determined based on the “concavo-convex degree” Sn, that is, based on “the uneven state of the trajectory of the fundamental frequency or the pseudo fundamental frequency smoothed over the entire voice”, it is recorded once on a hard disk recorder or the like. Although it can be applied when viewing the sound of a broadcast program, it is not preferable to apply it in the case of viewing a real-time broadcast program. Therefore, in the case of real-time viewing, for example, when there is a requirement that the delay from input to output in the speech speed conversion device 1 is less than 100 ms, the contribution rate Rb of the speech speed αb _n determined by the “degree of unevenness” Sn. Can be Rb = 0. As described above, the speech speed conversion apparatus 1 according to the present embodiment can perform advanced adaptive speech speed conversion by allocating the speech speed at a distribution rate according to the type of the input signal.

Also, the speech speed conversion magnification fine adjustment unit 400 gives a unit time (5 ms in this example) from the beginning of the input signal when a reproduction speed conversion magnification α [double speed] to be speed converted is given to the entire input signal. ) has a function of finely adjusting the fit signal length to the speech speed conversion ratio .alpha.n given to n-th _{(αn = αa n + αb n} + αc n) the reproduction speed conversion ratio alpha [speed] counted every other.

  For example, when the length of the entire input signal is set to L [seconds], the speech speed fine adjustment unit 400 performs the post-speech conversion when an arbitrary reproduction speed conversion magnification α [double speed] is given to the entire signal waveform. In order to make the length of the entire signal L / α [second], the length L0 [second] of the signal after speech speed conversion with the speech speed conversion magnification αn concatenated is calculated, and the following equation (4) To finely adjust the signal length to the playback speed conversion magnification α [double speed].

αn = (αa _n + αb _n + αc _n ) × L0 / (L / α) (4)

  Note that an arbitrary value such as 0.5 to 5.0 can be set as the reproduction speed conversion magnification α.

  That is, the speech speed conversion magnification fine adjustment unit 400 re-determines the speech speed conversion magnification αn per unit time according to the equation (4), finely adjusts by converting the speech speed, and sets the length of the converted signal waveform. It can be adjusted to a predetermined length.

Furthermore, when it is desired to match the same timing as that of the voice converted uniformly at α [double speed] as frequently as possible, not the length L of the entire input signal but the length of the signal obtained by dividing this by a shorter unit. It is also possible to correct αn so that fine adjustment is performed. For example, divided into M as the length _{L = L 1 + L 2 +} ··· + L M of the entire input _signal, L _1, L 2, · · ·, an input signal waveform in each section of _{L M} divided, in each divided section, the m th interval, first the speech speed of the m th interval conversion using a speech speed conversion ratio _{αn = αa n + αb n +} αc n of each part of each 5ms of the section First, the partial length L _m0 of the connected signal waveform after conversion is calculated. Thus, in the formula (4), instead of the length L of the entire input signal, applying a partial length L _m of the signal waveform, the signal waveform after conversion concatenated instead of the signal waveform length L0 after conversion concatenated by applying a partial length L _m0, performs speech speed conversion performed a fine adjustment again seeking the speech speed conversion ratio αn again.

  Various methods have already been proposed for speaking speed conversion (waveform expansion / contraction) when the speaking speed conversion magnification αn is given. For example, there are PICOLA (Pointer Interval Controlled OverLap and Add) method, TDHS (Time Domain Harmonic Scaling) method, PSOLA (Pitch Synchronous OverLap Add) method, etc. In addition, there are waveform expansion / contraction methods disclosed in Japanese Patent Nos. 261868, 3083830, and 2955247, and any of the waveform expansion / contraction methods may be used.

  The advanced adaptive speech speed conversion operation of the speech speed conversion apparatus 1 of the present embodiment will be described with reference to FIG. FIG. 2 is a flowchart showing the operation of the speech speed converting apparatus according to an embodiment of the present invention.

  In step S1, the speech speed converting apparatus 1 inputs a signal for adjusting the speech speed, and parameters necessary for the speech speed adjustment (contribution ratios Ra, Rb, Rc that can be defined by the type of the input signal, reproduction speed). Enter the conversion magnification α). The input signal is input to the voicing degree calculation unit 100, the fundamental frequency extraction unit 202, and the spectrum calculation unit 302. The contribution rates Ra, Rb, and Rc are respectively a first speech speed conversion magnification designation unit (speech speed conversion magnification designation unit a) 120, a second speech speed conversion magnification designation unit (speech speed conversion magnification designation unit b) 220, and The third speech speed conversion magnification designation unit (speech speed conversion magnification designation unit c) 320 is set. The reproduction speed conversion magnification α is set in the speech speed conversion magnification fine adjustment unit 400.

  First, the speech rate conversion apparatus 1 obtains the voicing degree Un as described above for the nth section divided for each predetermined unit time of the input signal by the voicing degree calculation unit 100 (step S2). The speech speed conversion magnification designation unit (speech speed conversion magnification designation unit a) 120 reduces the speech speed when the value of the voiced degree Un is larger than a predetermined threshold according to the value of the voiced degree Un. When the value of is lower than a predetermined threshold value, the speech speed conversion magnification αan that defines the expansion / contraction rate for the signal waveform per unit time of the input signal is determined so as to increase the speech speed (step S3).

  Furthermore, in the speech speed converting apparatus 1, the fundamental frequency value extracted at every unit time by the fundamental frequency extracting unit 202 stably changes within a predetermined variation range with respect to the input signal. Segment the area between the “stable section” and each stable section as an “unstable section”, specify the fundamental frequency in each stable section, and smooth the trajectory of the fundamental frequency in each stable section Thus, the fundamental frequency of the “stable section” is determined, and all the values of the fundamental frequencies of the “unstable section” are rejected (steps S4 and S5).

  Subsequently, the speech rate conversion apparatus 1 uses the values of the fundamental frequency of the smoothed trajectory of the stable section supplied from the fundamental frequency extraction unit 202 by the pseudo fundamental frequency calculation unit 204 to use an interpolation function such as a spline function. Are interpolated to determine the value of the pseudo fundamental frequency in the unstable interval, and the original fundamental frequency value in the unstable interval is converted (replaced) to the value of the pseudo fundamental frequency (step S6).

  Subsequently, the speech speed converting apparatus 1 is supplied from the fundamental frequency locus coupling unit 206 by the fundamental frequency value of the smoothed locus of the stable section supplied from the fundamental frequency extraction unit 202 and the pseudo fundamental frequency calculation unit 204. The fundamental frequency trajectory that is a continuous trajectory composed of the fundamental frequency and the pseudo fundamental frequency of all the segments of the input signal to be processed is obtained (step S7).

  Subsequently, the speech speed conversion apparatus 1 uses the unevenness degree calculation unit 210 to set a value P1 before a predetermined time and a value P1 after a predetermined time with respect to a certain value Pn of the basic frequency for each unit time constituting the basic frequency locus. Value P2 is sampled, and an average value of the front side difference value (Pn−P1) and the rear side difference value (Pn−P2) is obtained over all the intervals of the input signal to be processed, and this average in all intervals Each of the values is normalized by dividing by the maximum value among these average values, and each of the normalized average values is calculated as the “concave / convex degree” Sn representing the change tendency of the fundamental frequency locus (step S8).

  Subsequently, the speech speed conversion apparatus 1 responds to each value of the unevenness level Sn of the unit time supplied from the unevenness level calculation unit 210 by the second speech speed conversion rate specifying unit (speech speed conversion rate specifying unit b) 220. Thus, the expansion / contraction rate for the signal waveform per unit time of the input signal is specified so that the speaking speed is slowed when the unevenness Sn value is positive and the speaking speed is increased when the unevenness Sn value is negative. The speech speed conversion magnification αbn to be determined is determined (step S9).

  Furthermore, the speech speed conversion apparatus 1 uses the spectrum calculation unit 302 to convert a time domain waveform into a frequency domain for each unit time with respect to the input signal, and calculate a spectrum distribution including logarithmic power spectra of each frequency ( Step S10).

  Subsequently, the speech speed conversion apparatus 1 divides the logarithmic power spectrum supplied from the spectrum calculation unit 302 into a predetermined number of frequency bands by the band dividing unit 304 (step S11).

  Subsequently, the speech speed conversion apparatus 1 uses the power calculation unit 306 to arbitrarily select the low band side and the high band for the logarithmized power spectrum value of each frequency band divided from the band division unit 304. Each normalized power component on the side is obtained (step S12).

  Subsequently, the speech speed converting apparatus 1 uses the divided band power ratio calculation unit 310 to divide the normalized low-frequency power component supplied from the power calculation unit 306 by the normalized high-frequency power component. The “divided band power ratio” En, which is the ratio, is calculated (step S13).

  Subsequently, in the speech speed converting apparatus 1, the third speech speed conversion magnification specifying unit (speech speed conversion magnification specifying unit c) 320 sets the value of the divided band power ratio En according to the value of the divided band power ratio En. The rate of expansion / contraction with respect to the signal waveform per unit time of the input signal is increased so that the speech speed is reduced when the threshold value is greater than the threshold value, and the speech speed is increased when the value of the divided band power ratio En is equal to or less than the predetermined threshold value. The prescribed speech speed conversion magnification αcn is determined (step S14).

  Finally, the speech speed conversion apparatus 1 uses the speech speed conversion magnification fine adjustment unit 400 to at least one of the “voiced” Un, “unevenness” Sn, and “divided band power ratio” En of the physical indices. With respect to the determination of the speech speed conversion magnification of each section with respect to the input signal based on one or more, the allocation allocation for the final speech speed conversion magnification is determined according to the reproduction speed conversion magnification α, and the speech speed conversion is performed on the input signal (step) S15).

  Therefore, according to the speech speed conversion apparatus 1 of the present embodiment, the speech speed conversion magnification αn can be adaptively controlled according to the “voicedness” Un indicating the strength of the periodicity with respect to the input signal waveform. it can. The physical index of the voiced degree Un can be obtained at all positions of the input signal. Further, even when background sounds are mixed, a physical index of voicing level Un can be obtained, and stable speech speed conversion can be realized.

  Usually, the vowel part of speech has a high voiced degree Un. In addition, a completely silent portion or a background sound such as music or noise in which frequency components of various sounds are generally mixed has low voicedness. Therefore, according to the speech speed conversion apparatus 1 of the present embodiment, the speech speed can be reduced when the voiced degree Un is high, and the speech speed can be increased where the voiced degree Un is low. Even in the case of a signal, the speaking speed is slowed down in the vowel part, which is an important part for listening to the voice, and the speaking speed is speeded up in the complete silent part and only the background sound. Furthermore, adaptive speech speed conversion is possible while matching the target time length as a whole. In particular, it is known that the vowel part mainly expands and contracts when the speaking speed is slow and fast in an actual person's utterance (for example, the IEICE Transactions (A), Vol. J67- A, No. 7, issued in July 1984, pp. 629-636). Therefore, according to the speech speed conversion apparatus 1 of the present embodiment, since the speech speed is converted according to the voiced degree Un, natural speech adaptive speech speed conversion can be expected.

  Furthermore, according to the speech speed conversion apparatus 1 of the present embodiment, “the state of irregularities in the fundamental frequency locus of the fundamental frequency or the pseudo fundamental frequency smoothed over the entire speech” is applied to the input speech waveform. The speech speed conversion magnification αn can be adaptively controlled according to the degree of unevenness “Sn”. Therefore, according to the speech speed converting apparatus 1 of the present embodiment, the speech speed is basically reduced in the section of the fundamental frequency locus that is convex in a mountain shape, and the fundamental frequency locus that is concave in a valley shape. Since the speech speed can be increased in the section, adaptive speech speed conversion can be performed while stably adjusting to the target time length as a whole.

  This is different from the method of “slowing the speaking speed when the fundamental frequency is high and increasing the speaking speed when it is low” as in the technique of Patent Document 1. As described above, the fundamental frequency can be accurately obtained in the vowel part of voice (voiced sound part), but the fundamental frequency cannot be stably obtained in other parts such as background sound. Therefore, the speech speed conversion apparatus 1 of the present embodiment uses the extracted fundamental frequency as a fundamental frequency used for determining the speech speed conversion magnification in a stable section where the value of the extracted fundamental frequency is stable and changes substantially continuously. The trajectory of the fundamental frequency is smoothed so that the trajectory of the fundamental frequency used for determining the speech rate conversion magnification becomes smoother.

  Further, the speech speed converting apparatus 1 of the present embodiment rejects all the values of the fundamental frequency in the unstable section in the unstable section where the value of the extracted fundamental frequency is not stable and is discontinuous and changes rapidly. The pseudo fundamental frequency is obtained by interpolating with a spline function or the like using the fundamental frequency value in the stable section. Thereby, according to the speech rate conversion apparatus 1 of a present Example, the continuous fundamental frequency locus | trajectory which consists of a fundamental frequency or a pseudo fundamental frequency can be acquired in the area of all the input signals. According to the speech speed converting apparatus 1 of the present embodiment, in this fundamental frequency locus, the speech speed is slowed at a portion that is convex (maximum) in a mountain shape, and is spoken at a portion that is valley shape (minimum). Therefore, adaptive speech speed conversion can be performed stably and in accordance with the target time length as a whole.

  Furthermore, the speech speed conversion apparatus 1 of the present embodiment has an advantage over the method of “relaxing the speech speed when the fundamental frequency is high and increasing the speech speed when the fundamental frequency is low” as in the technique of Patent Document 1. is there. For example, an input signal such as a comic dialogue by a male / female combination consists of a mixed voice section in which the male and female voice sections are violently switched with almost no pause. For such an input signal, the method of “low speed of speech at a high fundamental frequency and high speed of speech at a low frequency” as in Patent Document 1 always has a voice of a female voice because a female voice is high. However, since the male voice is low, there is a tendency for the male voice to always increase the speaking speed. On the other hand, the speech rate conversion apparatus 1 according to the present embodiment, in the fundamental frequency trajectory of the smoothed fundamental frequency or pseudo fundamental frequency, accompanies utterance accents, etc. Since irregularities always occur in the voice part, the convex part of the fundamental frequency trajectory can slow down the speech speed, and the concave part of the fundamental frequency trajectory can speed up the speech speed, regardless of the difference in gender utterance. It is possible to adaptively control speech speed with fair distribution to both parties.

  Furthermore, the speech speed converting apparatus 1 of the present embodiment is configured to output power components obtained by dividing the low band side of two bands, which may be obtained by dividing the frequency spectrum into a plurality of bands, by the high band side. The speech rate conversion magnification can be adaptively controlled according to the “divided band power ratio” En, which is the ratio. As in the technologies of Patent Document 4 and Patent Document 5, “the input signal is a“ voice section ”by comparing the power of each of the bands of the frequency spectrum in the steady state and the corresponding bands of the frequency spectrum of the input signal. Unlike the case of “determining whether it is a silence period”, the speech speed converting apparatus 1 of the present embodiment is capable of “reducing two bands in which the frequency spectrum is divided into a plurality of bands. A “divided band power ratio” En, which is a ratio of power components obtained by dividing the band side by the high band side, is used. This “divided band power ratio” En is not compared with the power of the spectrum in the steady state, but only the frequency spectrum at a certain moment of the input signal, and the frequency spectrum at a certain moment is divided into bands. The power ratio obtained by dividing the low band side of two bands in each frequency spectrum by the high band side is obtained.

  The speech speed conversion apparatus 1 according to the present embodiment is similar to the techniques disclosed in Patent Document 4 and Patent Document 5, “divides an input signal into a voice section and a silent section, slows down the speech speed of the voice section, and shortens the silent section”. There is an advantage over the method.

  For example, with the techniques of Patent Document 4 and Patent Document 5, it is determined whether the sound is a “sound section” or “silent section” with respect to an input signal in which music of a certain volume is mixed as background sound. When it is performed, as described above, it is difficult to correctly distinguish between “voice section” and “silent section”, and adaptive speech speed conversion cannot be performed. On the other hand, the speech speed converting apparatus 1 of the present embodiment is intended only for the frequency spectrum at a certain moment of the input signal, and the frequency spectrum at a certain moment is band-divided, and two bands of each frequency spectrum obtained by the band division are obtained. Since the speech speed conversion magnification is determined based on the “divided band power ratio” En obtained by dividing the low bandwidth side by the high bandwidth side, there is essentially no judgment error, and the speech speed is stable. Control can be performed. For example, when the power component on the high band side is small relative to the power component on the low band side, the speech speed is slowed down, and when the power component on the high band side is large compared to the power component on the low band side, the speech speed is increased. Can do. That is, since this “divided band power ratio” En has a different value depending on the type of voice section, music, noise, silence, etc. in the input signal, according to the speech rate conversion apparatus 1 of the present embodiment, By performing the speech speed control based on the value of the band power ratio “En”, the speech speed can be reduced in the speech section, and the speech speed can be increased in the sections that are not speech such as music, noise, and silence.

  Furthermore, the speech speed conversion apparatus 1 according to the present embodiment uses one or more “physical indicators” of “voicedness” Un, “unevenness” Sn, and “divided band power ratio” En. It is possible to perform more advanced adaptive speech speed conversion in which speech speed conversion is performed by determining the speech speed conversion magnification of each section with respect to. For example, the voice rate conversion magnification specified by “Voice” Un is 0.5, the voice rate conversion magnification specified by “Roughness Level” Sn is 0.3, and “Divided Band Power Ratio” En is specified. Giving a contribution rate (allocation allocation) of 0.2 to the speech rate conversion magnification, and adding each speech rate component based on "voicedness" Un, "unevenness" Sn, and "divided band power ratio" En Thus, the final speech rate conversion magnification can be determined. For example, when the target input signal for speech speed conversion is a broadcast sound, the program genre (news, documentary, drama, variety, rakugo, comic story, etc.) is given as meta information that has been actively developed in recent years. In the case of broadcasting sound, the contribution rate (allocation allocation) can be changed according to this genre, and adaptive speech speed conversion with higher ease of listening and higher naturalness is possible.

  Furthermore, when a predetermined time length is set for the entire input signal or each part divided according to a predetermined rule, the speech speed converting apparatus 1 of the present embodiment expands and contracts in time so as to match this time length. It is possible to perform adaptive speech speed conversion that adjusts the magnification to match a predetermined time length. This is because each divided part of the input signal is given an arbitrary playback speed conversion magnification α [double speed] such as 1 × speed (playback in real time) or 2 × speed (playback in half the real time). When it is required to change the speech speed at a magnification larger or smaller than the playback speed conversion magnification α [double speed], the overall playback time will be the same as the speech speed conversion at a uniform magnification α. As a result, it is possible to generate the speech speed converted voice with the same time length as when the speech speed is converted with the uniform playback speed conversion magnification α. .

  Furthermore, as one aspect of the present invention, the speech rate conversion apparatus 1 of the present embodiment can be configured as a computer. A program for causing a computer to realize each of the above-described components is stored in a storage unit provided inside or outside the computer. Such a storage unit can be realized by an external storage device such as an external hard disk or an internal storage device such as ROM or RAM. The control unit provided in the computer can be realized by controlling a central processing unit (CPU) or the like. In other words, the CPU can appropriately read from the storage unit a program in which the processing content for realizing the function of each component is described, and realize the function of each component on the computer. Here, you may implement | achieve the function of each component by all or a part of hardware.

  In addition, the program describing the processing contents can be distributed by selling, transferring, or lending a portable recording medium such as a DVD or CD-ROM, and such a program can be distributed on a network such as an IP. The program can be distributed by storing the program in the storage unit of the server and transferring the program from the server to another computer via the network.

  In addition, a computer that executes such a program can temporarily store, for example, a program recorded on a portable recording medium or a program transferred from a server in its own storage unit. As another embodiment of the program, the computer may directly read the program from a portable recording medium and execute processing according to the program, and each time the program is transferred from the server to the computer. In addition, the processing according to the received program may be executed sequentially.

  While the embodiments of the present invention have been described in detail with specific examples, it will be apparent to those skilled in the art that various modifications and changes can be made without departing from the scope of the claims of the present invention. For example, in the speech speed conversion apparatus 1 of the present embodiment, a “physical index” composed of “voicedness” Un, “concave / convex degree” Sn, and “divided band power ratio” En is calculated, and the allocation allocation of the input signal is performed. In the above description, the speech speed conversion is performed to calculate the “physical index” of only “voiced” Un and to convert the speech speed of the input signal. A speech speed conversion apparatus that calculates the "physical index" and converts the speech speed of the input signal, a speech speed conversion apparatus that calculates the "physical index" of only the "divided band power ratio" En and performs the speech speed conversion of the input signal, etc. , Calculating at least one of “voiced” Un, “unevenness” Sn, and “divided band power ratio” En, and converting the input signal to a speech rate converter. it can. Therefore, the present invention is not limited to the above embodiments.

  INDUSTRIAL APPLICABILITY According to the present invention, the present invention can be applied to all uses of speech speed conversion technology that listens slowly to a television or radio sound in real time, or records it on a hard disk recorder once and views it slowly or quickly. For example, there is a request from a visually impaired person to efficiently listen to audio information. According to the present invention, a recorded book for visually impaired persons can be reproduced and heard at high speed. Furthermore, according to the present invention, a language learning or utterance training system that is used when creating the teaching material or for converting the speech speed of the voice in accordance with the learner's progress at the time of learning to the learner is provided. It can be applied and is useful for any application that requires speech speed conversion.

DESCRIPTION OF SYMBOLS 1 Speech speed converter 2 Physical index calculation part 3 Speech speed conversion magnification determination part 100 Voicedness calculation part 200 Fundamental frequency / pseudo fundamental frequency unevenness calculation part 210 Concavity and convexity calculation part 300 Frequency band / power calculation part 310 Divided band power ratio calculation Unit 202 Fundamental frequency extraction unit 204 Pseudo fundamental frequency calculation unit 206 Fundamental frequency locus connection unit 300 Frequency band / power calculation unit 302 Spectrum calculation unit 304 Band division unit 306 Power calculation unit 120 First speech speed conversion magnification designation unit (speech speed conversion) Magnification designation part a)
220 Second speech speed conversion magnification designation section (speech speed conversion magnification designation section b)
320 Third speech speed conversion magnification designation section (speech speed conversion magnification designation section c)
400 Speaking speed conversion magnification fine adjustment section

Claims (8)

A speech speed conversion device that performs adaptive speech speed conversion of an input signal,
For each segment obtained by dividing the input signal every unit time, a physical index calculation unit that calculates a physical index of the input signal,
According to the physical index calculated by the physical index calculation unit, the speech speed conversion magnification determination unit that determines the speech speed conversion magnification to be specified for each segment of the input signal and performs the speech speed conversion;
A speech speed conversion device comprising: The physical index calculation unit
The speech rate conversion apparatus according to claim 1, further comprising: a voicing degree calculating unit that calculates a voicing degree representing a relative maximum value obtained by autocorrelation per unit time in the input signal as the physical index. The physical index calculation unit
The unevenness degree calculation part which calculates the unevenness degree showing the change tendency of the locus of the fundamental frequency per unit time in the input signal and the pseudo fundamental frequency as the physical index is provided, The unevenness degree calculation part according to claim 1 or 2 characterized by things. Speaking speed converter. The physical index calculation unit
2. A division band power ratio calculation unit that calculates a ratio between a low frequency side power component and a high frequency side power component obtained by dividing a band per unit time in an input signal as the physical index is provided. 4. The speech rate conversion device according to any one of 3 above. The speech speed conversion magnification determination unit
A speech speed conversion magnification fine adjustment unit that finely adjusts the determined speech speed conversion magnification to match the reproduction speed conversion magnification when a playback speed conversion magnification to be speed converted is given to the entire input signal; The speech rate conversion apparatus according to any one of claims 1 to 4, further comprising: The speech speed conversion magnification determination unit
A speech speed conversion magnification fine adjustment unit that determines a speech speed conversion magnification to be specified for each segment of the input signal using one or more physical indices of the voiced degree, the unevenness degree, and the divided band power ratio. The speech speed conversion device according to claim 1, comprising:   The speech speed conversion magnification fine adjustment unit allocates the speech speed conversion ratio based on one or more physical indices of the voiced degree, the unevenness degree, and the divided band power ratio according to the type of the input signal. The speech rate conversion device according to claim 6, wherein the speech rate conversion device is performed. In a computer configured as a speech rate conversion device that performs adaptive speech rate conversion of an input signal,
For each segment obtained by dividing the input signal per unit time, the voicing level representing the relative maximum value obtained by autocorrelation per unit time in the input signal, the change in the trajectory of the fundamental frequency and pseudo fundamental frequency per unit time in the input signal One or more physical indices are calculated from the degree of unevenness representing the trend and the divided band power ratio representing the ratio between the low-frequency power component and the high-frequency power component divided into bands per unit time in the input signal. Steps,
In accordance with the physical index calculated in the step, determining a speech speed conversion magnification to be specified for each segment of the input signal and performing speech speed conversion;
A program for running

Images