JP2012208177A - Band extension device and sound correction device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a band extension device capable of improving an auditory recognition degree.SOLUTION: A band extension device includes: time frame division means for dividing an input signal of a time area for every time frame; Fourier transformation means for performing Fourier transformation to time frames to generate an original sound spectrum of a frequency area; harmonic spectrum generation means for generating a harmonic spectrum based on the original sound spectrum; harmonic spectrum addition means for adding the harmonic spectrum to the original sound spectrum; Fourier inverse transformation means for performing Fourier inverse transformation to the original sound spectrum to which the harmonic spectrum is added to generate output signal components of the time areas; and output signal generation means for adding the output signal components respectively to generate an output signal whose frequency band is extended.

Description

  The present invention relates to a band extending apparatus for extending a frequency band of an input signal. The present invention also relates to an audio correction device for correcting an audio signal from a bone-conduction microphone.

  Conventionally, band extension devices for extending the frequency band of an input signal have been proposed (see, for example, Patent Document 1 and Patent Document 2). The conventional band extension device includes a low-frequency signal generating means for generating a low-frequency signal from an input signal, a high-frequency signal generating means for generating a high-frequency signal from the input signal, and the generated low-frequency signal and high-frequency signal, respectively. Adding means for adding to the input signal. The low-frequency signal generating means performs full-wave rectification (nonlinear processing) after oversampling the input signal, and generates a low-frequency signal by performing bandpass filtering on the input signal subjected to full-wave rectification in the frequency domain. The high frequency signal generating means generates a high frequency signal by oversampling the input signal and then copying a low frequency band of the input signal to the high frequency side in the frequency domain. The low-frequency signal and high-frequency signal thus generated are added to the input signal, thereby generating an output signal with an expanded frequency band. By expanding the frequency band in this way, it is possible to improve telephone call sound quality, playback sound quality of digital audio equipment, and the like.

  Conventionally, an audio correction device for correcting an audio signal from a bone-conduction microphone has been proposed (see, for example, Patent Document 3). Bone-conduction microphones are also called bone-conduction microphones. These are vibration pickups that record bone vibrations caused by vocal cord vibrations during voice production using the forehead, chin, cheeks, ear holes, etc., and use them as substitute signals for actual speech. One. Such a bone-conduction microphone can obtain a higher S / N ratio than an air-conduction microphone, and is effectively used as a voice input means under high noise. A conventional speech correction apparatus includes a bone conduction microphone, a time frame dividing unit that divides a speech signal from the bone conduction microphone into time frames, and an LPC that analyzes pitch frequency data and bone conduction speech feature parameters based on the time frames. LPC synthesis for generating a pseudo air conduction speech signal based on the pitch frequency data and the pseudo air conduction speech feature parameter, and an analysis means, a signal conversion means for converting the analyzed bone conduction speech feature parameter into a pseudo air conduction speech feature parameter Means and smoothing means for converting the pseudo air conduction sound signal into a long-time signal.

Japanese Patent No. 3301473 JP 2003-15695 A Japanese Patent No. 3306784

  First, the conventional band extending apparatus as described above has the following problems. In general, as an important factor representing the nature of an input signal, there is a frequency at the peak of the spectral envelope (hereinafter referred to as “peak frequency”). The peak frequency of the spectral envelope in speech is the formant peak frequency, and this formant peak frequency is important for identifying the phoneme of the vocal tract transmission system. Further, the peak frequency of the spectrum envelope in the musical sound is a resonance frequency of the resonance system based on the type of the musical instrument and the structure of each part thereof, and this resonance frequency is important as representing the characteristics of the musical sound. Since the peak frequency of such a spectral envelope is a resonance / resonance phenomenon, it is common to have a high-order resonance mode at a frequency that is multiplied by the primary resonance mode. Therefore, when generating a high-frequency signal, it is important to have a peak of a high-order resonance mode with a spectral envelope at an appropriate frequency, similar to harmonics of the fundamental frequency.

  However, in the conventional band extension device as described above, the shift amount when copying the low frequency band of the input signal to the high frequency side in the frequency domain is independent of the peak frequency of the spectrum envelope, The peak of the higher order resonance mode of the spectral envelope cannot be reproduced in the band signal. For this reason, it is difficult to obtain an output signal with natural characteristics, and there is a limit to improving the degree of hearing. In addition, a low-frequency signal is generated by full-wave rectification after oversampling the input signal. However, in such a configuration, when noise is mixed in the input signal, the rectification timing is near the zero crossing. It is greatly affected by noise. As a result, the output sound quality may be degraded.

  Secondly, the conventional sound correcting apparatus as described above has the following problems. The conventional speech correction apparatus is configured to analyze the pitch frequency data and the bone conduction speech feature parameters based on the speech signal. However, when the speech signal is an unvoiced sound due to the characteristics of the bone conduction microphone, the pitch frequency is determined. I can't get the data. Therefore, a pseudo air conduction sound signal cannot be generated for an unvoiced sound, and the output sound becomes unnatural and it is difficult to improve the degree of hearing.

  An object of the present invention is to provide a band extension device and a sound correction device capable of improving the degree of hearing.

The band extending apparatus according to claim 1 of the present invention is a band extending apparatus for extending a frequency band of an input signal,
Time frame dividing means for dividing the time domain input signal into time frames; Fourier transform means for generating a frequency domain original sound spectrum by Fourier transforming the time frame; and generating a harmonic spectrum based on the original sound spectrum A harmonic spectrum generating means for adding the harmonic spectrum to the original sound spectrum, and inverse Fourier transforming the original sound spectrum to which the harmonic spectrum has been added to obtain an output signal component in the time domain. An inverse Fourier transform means for generating, and an output signal generation means for adding an output signal component to generate an output signal with an expanded frequency band.
The harmonic spectrum generation means calculates the frequency of the original sound spectrum component included in the original sound spectrum, and sets the frequency multiplied by the calculated frequency as the frequency of the harmonic spectrum component included in the harmonic spectrum. Features.

  Further, in the band extending apparatus according to claim 2 of the present invention, the harmonic spectrum generating means calculates a phase angle of the original sound spectrum component included in the original sound spectrum, and a phase of multiplication of the calculated phase angle. An angle is set as a phase angle of the harmonic spectrum component included in the harmonic spectrum.

  Further, in the band extending apparatus according to claim 3 of the present invention, the harmonic spectrum generating means is based on the original sound spectrum analyzing means for analyzing the original sound spectrum and the analysis result of the original sound spectrum analyzing section. Weighting means for weighting the magnitude of the harmonic spectrum component contained in the wave spectrum with a predetermined weighting factor.

  Moreover, in the band extending apparatus according to claim 4 of the present invention, the original sound spectrum analyzing means is configured to analyze an inclination of an envelope of the original sound spectrum, and the weighting means The weighting coefficient is changed based on the slope of the envelope of the original sound spectrum.

  In the band extending apparatus according to claim 5 of the present invention, the harmonic spectrum generating means includes an interpolation means for interpolating the harmonic spectrum component, and the interpolation means adds the original sound spectrum to the original sound spectrum. The harmonic spectral component having a frequency between a frequency multiplied by the frequency of the first original sound spectral component included and a frequency multiplied by the frequency of the second original sound spectral component adjacent to the first original sound spectral component is interpolated. It is characterized by that.

  In the band extending apparatus according to claim 6 of the present invention, the input signal is a voice signal or a musical tone signal generated by a microphone, a vibration pickup, a bone conduction microphone, a digital acoustic device, a telephone system, an artificial vocal cord, or the like. It is characterized by being.

  In the sound correction device according to claim 7 of the present invention, the bone conduction microphone, the time frame dividing means for dividing the sound signal from the bone conduction microphone for each time frame, and the sound of the sound signal in the time frame. A voice characteristic analyzing means for analyzing the characteristics; a voice property determining means for determining whether the voice signal in the time frame is voiced or unvoiced based on the voice characteristic analysis result; and correcting the voice signal determined to be voiced. First signal correcting means for generating a pseudo air conduction sound signal, second signal correcting means for generating a pseudo air conduction sound signal by correcting the sound signal determined to be unvoiced sound, and existence by the first signal correcting means. A correction mode switching means for switching between the voice sound correction mode and the unvoiced sound correction mode by the second signal correction means, and the generated pseudo air conduction sound signal are respectively added and output. Characterized in that it comprises an output signal generating means for generating a signal.

The speech correction apparatus according to claim 8 of the present invention further includes parameter storage means for storing a bone-conducted vocal tract characteristic parameter, an air-conducted vocal tract characteristic parameter, and an air-conducted sound source characteristic parameter. The analysis means is configured to analyze the bone conduction sound source characteristics and the bone conduction vocal tract characteristics of the speech signal in the time frame,
The first signal correction means reads out the air conduction vocal tract characteristic parameter corresponding to the bone conduction vocal tract characteristic of the sound signal analyzed by the sound characteristic analysis means from the storage means, and the sound characteristic A pseudo air conduction sound signal is generated by synthesizing the bone conduction sound source characteristic of the sound signal analyzed by the analyzing means and the air conduction sound path characteristic parameter.

  Further, in the sound correction device according to claim 9 of the present invention, the second signal correction means is based on the bone-conducted vocal tract characteristic of the sound signal analyzed by the sound characteristic analysis means, and corresponds to the sound conduction characteristic. The air conduction sound path characteristic parameter and the air conduction sound source characteristic parameter are read from the storage means, and the read air conduction sound path characteristic parameter and the air conduction sound source characteristic parameter are synthesized to generate a pseudo air conduction sound signal. It is characterized by generating.

  According to the band extending apparatus of the first aspect of the present invention, the harmonic spectrum generating means calculates the frequency of the original sound spectrum component included in the original sound spectrum, and sets the frequency multiplied by the calculated frequency as the harmonic spectrum. Since it is set as the frequency of the included harmonic spectrum component, the peak of the higher-order resonance mode of the spectrum envelope can be reproduced in the output signal. Therefore, an output signal with natural characteristics can be obtained, and the degree of hearing can be improved. In addition, the S / N ratio of the original sound spectrum component can be restored with the harmonic spectrum component, and the influence of noise can be reduced.

  According to the band extending apparatus of the second aspect of the present invention, the harmonic spectrum generating means calculates a phase angle of the original sound spectrum component included in the original sound spectrum, and a phase angle obtained by multiplying the calculated phase angle. Is set as the phase angle of the harmonic spectrum component included in the harmonic spectrum, the time relationship between the original sound spectrum and the harmonic spectrum can be kept constant. Thereby, when the output signal components in the time domain are added, the harmonic components of the output signal components can be prevented from canceling each other, and the output signal can be generated with high accuracy.

  According to the band extending apparatus of the third aspect of the present invention, the weighting means is predetermined with respect to the magnitude of the harmonic spectrum component included in the harmonic spectrum based on the analysis result of the original sound spectrum analysis unit. Thus, for example, a harmonic spectrum considering the phoneme of the original sound spectrum can be generated, and an output signal with more natural characteristics can be obtained.

  According to the band extending apparatus of the fourth aspect of the present invention, the weighting means changes the weighting coefficient based on the slope of the envelope of the analyzed original sound spectrum. For example, when the slope of the envelope of the original sound spectrum is large, the weighting coefficient is set small. Thereby, the magnitude | size of a harmonic spectrum component becomes small. Generally, the spectrum of a vowel has a characteristic that the attenuation of the spectrum component on the high frequency side becomes large. Therefore, the vowel can be restored more realistically by setting the weighting coefficient small as described above. For example, when the slope of the envelope of the original sound spectrum is small, the weighting coefficient is set large. Thereby, the magnitude | size of a harmonic spectrum component becomes large. In general, the consonant spectrum has a characteristic that the attenuation of the spectral component on the high frequency side is reduced. Therefore, the consonant can be restored more realistically by setting the weighting factor large as described above.

  According to the band extending apparatus of the fifth aspect of the present invention, the harmonic spectrum generating means includes the interpolation means for interpolating the harmonic spectrum component. By interpolating the harmonic spectrum components in this way, the harmonic spectrum can be generated with higher accuracy.

  According to the band extending apparatus of the sixth aspect of the present invention, the input signal is a voice signal or musical tone signal generated by a microphone, vibration pickup, bone-conduction microphone, digital acoustic device, telephone system, artificial vocal cord, or the like. It is. Such microphones, vibration pickups, bone-conduction microphones, digital audio equipment, telephone systems, artificial vocal cords, etc. are difficult to output high frequencies due to their insufficient characteristics, but the band expansion device of the present invention is applied to them. Thus, a natural high frequency range can be reproduced, and the degree of hearing can be improved.

  According to the sound correction apparatus of the seventh aspect of the present invention, the sound correction apparatus includes the correction mode switching means for switching between the voiced sound correction mode by the first signal correction means and the unvoiced sound correction mode by the second signal correction means. Thus, it is possible to generate a pseudo air conduction sound signal not only for voiced sound but also for unvoiced sound as in the prior art. As a result, an output signal with natural characteristics can be obtained, and the degree of hearing can be improved.

  According to the speech correction apparatus of the present invention, the first signal correction means is adapted to the corresponding voice based on the bone conduction vocal tract characteristic of the voice signal analyzed by the voice characteristic analysis means. The derived vocal tract characteristic parameter is read from the storage means, and the bone conduction sound source characteristic and the air conduction vocal tract characteristic parameter of the voice signal analyzed by the voice characteristic analyzing means are synthesized to generate a pseudo air conduction voice signal. Thereby, a pseudo air conduction sound signal can be easily and accurately generated for voiced sound.

  According to the sound correcting device of the present invention, the second signal correcting means is based on the bone-conducted vocal tract characteristic of the sound signal analyzed by the sound characteristic analyzing means, and is adapted to this. The guided sound path characteristic parameter and the air conduction sound source characteristic parameter are read from the storage means, and the read air conduction sound path characteristic parameter and the air conduction sound source characteristic parameter are synthesized to generate a pseudo air conduction sound signal. Thereby, a pseudo air conduction sound signal can be easily and accurately generated with respect to an unvoiced sound.

It is a block diagram which shows the structure of the band expansion apparatus by one Embodiment of this invention. It is a figure for demonstrating the state which divided | segmented the input signal for every time frame. It is a figure for demonstrating the production | generation process of an output signal. (A) is a spectrum diagram showing an original sound spectrum to which a harmonic spectrum is added, (b) is a diagram showing, for example, one original sound spectrum component of the original sound spectrum on a complex plane, and (c) These are the figures which represented one harmonic spectrum component corresponding to the original sound spectrum component of FIG.4 (b) on a complex plane. It is a block diagram which shows the structure of the band expansion apparatus by other embodiment of this invention. It is a figure for demonstrating the method to analyze the inclination of the envelope of an original sound spectrum. (A) is a spectrum diagram showing an original sound spectrum whose phoneme is a vowel, and (b) is a spectrum diagram showing an original sound spectrum to which a harmonic spectrum is added. (A) is a spectrum figure which shows the original sound spectrum whose phoneme is a consonant, (b) is a spectrum figure which shows the original sound spectrum to which the harmonic spectrum was added. It is a block diagram which shows the structure of the audio | voice correction apparatus by one Embodiment of this invention. It is a figure which shows the correspondence of each parameter memorize | stored in the parameter memory | storage part. It is a block diagram which shows the structure of the parameter production apparatus for producing an air conduction vocal tract characteristic parameter. It is a block diagram which shows the parameter production apparatus for producing a bone-conduction sound path characteristic parameter and an air conduction sound source characteristic parameter.

Hereinafter, various embodiments of a bandwidth expansion device and a sound correction device according to the present invention will be described with reference to the accompanying drawings.
[Embodiment of Bandwidth Expansion Device]
First, an embodiment of a bandwidth expansion device will be described with reference to FIGS. FIG. 1 is a block diagram showing a configuration of a bandwidth extension apparatus according to an embodiment of the present invention, FIG. 2 is a diagram for explaining a state in which an input signal is divided for each time frame, and FIG. FIG. 4A is a diagram for explaining a generation process of an output signal, FIG. 4A is a spectrum diagram showing an original sound spectrum to which a harmonic spectrum is added, and FIG. 4B is an example of one of the original sound spectra. FIG. 4C is a diagram showing one harmonic spectrum component corresponding to the original sound spectrum component of FIG. 4B on the complex plane. .

  Referring to FIG. 1, the illustrated bandwidth extension device 2 includes an input terminal 4, an oversampling low-pass filter (LPF) 6, a time frame division unit 8 (which constitutes a time frame division unit), and a Fourier transform unit 10 (Fourier). Conversion means), harmonic spectrum generation unit 12 (configures harmonic spectrum generation means), harmonic spectrum addition unit 14 (configures harmonic spectrum addition means), inverse Fourier transform unit 16 (inverse Fourier transform) An output signal generator 18 (which constitutes an output signal generator) and an output terminal 20.

  A time domain input signal sampled at a sampling frequency of 8 kHz is input to the input terminal 4. The input signal is, for example, an audio signal in a telephone system or a musical sound signal in a digital acoustic device.

  The oversampling low-pass filter 6 oversamples the sampling frequency of the input signal from the input terminal 4 from 8 kHz to 16 kHz, and attenuates the frequency band of 4 kHz or more of the input signal.

  The time frame dividing unit 8 divides the time domain input signal output from the oversampling low-pass filter 6 into time frames having a predetermined time length (for example, 16 msec) using the Hanning window as a window function (see FIG. 2). ). Each time frame is divided so that both end portions thereof overlap with time frames adjacent to both sides. From each time frame divided in this way, 256 samples are taken out.

  The Fourier transform unit 10 generates an original sound spectrum in the frequency domain by performing a Fourier transform (short-time Fourier transform) on each of the time frames divided by the time frame dividing unit 8. In this embodiment, the frequency band of each original sound spectrum is 0 to 4 kHz.

  The harmonic spectrum generation unit 12 includes an original sound spectrum component extraction unit 22, a harmonic spectrum calculation unit 24, and an interpolation calculation unit 26 (which constitutes an interpolation unit). The original sound spectrum component extraction unit 22 extracts a plurality of original sound spectrum components from the frequency band of 2 to 4 kHz of the original sound spectrum. The harmonic spectrum calculation unit 24 calculates a harmonic spectrum component based on the extracted original sound spectrum component as described later, synthesizes the calculated harmonic spectrum component, and has a frequency band of 4 to 8 kHz. Generate a wave spectrum. In addition, the interpolation calculation unit 26 performs an interpolation calculation of harmonic spectrum components as described later.

  The harmonic spectrum adding unit 14 adds the harmonic spectrum generated as described above to the original sound spectrum. As a result, an original sound spectrum in which the frequency band of 4 to 8 kHz is expanded is generated.

  The inverse Fourier transform unit 16 generates an output signal component in the time domain by performing inverse Fourier transform on the original sound spectrum to which the harmonic spectrum is added.

  The output signal generation means 18 generates a time domain output signal by adding the output signal components generated as described above. Each output signal component is added such that both end portions thereof overlap with output signal components adjacent to both sides. The output signal generated in this way is output from the output terminal 20 to the outside.

Next, with reference to FIG. 3 and FIG. 4 as well, the flow of generating an output signal by the band extending device 2 described above will be described. First, the time domain input signal T _in (analog signal) input from the input terminal 4 is converted into a digital signal by an A / D converter (not shown), and then the sampling frequency is converted by an oversampling low-pass filter 6. Oversampling is performed from 8 kHz to 16 kHz, and a frequency band of 4 kHz or more is attenuated (see FIG. 3A). Next, the input signal is divided for each time frame T _n (n = 1, 2, 3,...) By the time frame dividing unit 8 (see FIG. 3B). Each time frame T _n is Fourier transformed by the Fourier transform unit 10, thereby generating an original sound spectrum X _{n in the} frequency domain (see FIG. 3C). Each original sound spectrum _Xn has a frequency band of 0 to 4 kHz.

Thereafter, the harmonic spectrum generator 12 generates a harmonic spectrum X _n ′ based on the original sound spectrum X _n as follows. First, the original sound spectrum component extracting unit 22 extracts a plurality of original sound spectrum components X (f _n ) from the frequency band of 2 to 4 kHz of the original sound spectrum X _n . When the original sound spectrum component X (f _n ) is expressed by a mathematical expression, the following expression (1) is obtained.

次いで、高調波スペクトル演算部２４によって、抽出された原音スペクトル成分Ｘ（ｆ_ｎ）に基づいて高調波スペクトル成分Ｘ’（２ｆ_ｎ）が演算される。具体的には、図４（ａ）〜（ｃ）に示すように、高調波スペクトル演算部２４は、原音スペクトル成分Ｘ（ｆ_ｎ）の周波数ｆ_ｎを算出し、この算出した周波数ｆ_ｎの２倍の周波数２ｆ_ｎを高調波スペクトル成分Ｘ’（２ｆ_ｎ）の周波数として設定する。また、高調波スペクトル演算部２４は、原音スペクトル成分Ｘ（ｆ_ｎ）の位相角θ_ｎを算出し、この算出した位相角θ_ｎの２倍の位相角２θ_ｎを高調波スペクトル成分Ｘ’（２ｆ_ｎ）の位相角として設定する。更に、高調波スペクトル演算部２４は、原音スペクトル成分Ｘ（ｆ_ｎ）の大きさ（レベル）｜Ｘ（ｆ_ｎ）｜を算出し、この算出した大きさ｜Ｘ（ｆ_ｎ）｜の１／２倍の大きさ１／２｜Ｘ（ｆ_ｎ）｜を高調波スペクトル成分Ｘ’（２ｆ_ｎ）の大きさとして設定する。なお、図４では、原音スペクトルをＸ（ｆ）、高調波スペクトルをＸ’（ｆ）と表している。また、本実施形態では、高調波スペクトル成分Ｘ’（２ｆ_ｎ）の大きさに対して加重を施す加重係数は、１／２（一定）である。高調波スペクトル成分Ｘ’（２ｆ_ｎ）を数式で表現すると、次式（２）で示すようになる。

Next, a harmonic spectrum component X ′ (2f _n ) is calculated by the harmonic spectrum calculation unit 24 based on the extracted original sound spectrum component X (f _n ). Specifically, as shown in FIGS. 4A to 4C, the harmonic spectrum calculation unit 24 calculates the frequency f _n of the original sound spectrum component X (f _n ), and calculates the frequency f _n of the calculated frequency f _n . The double frequency 2f _n is set as the frequency of the harmonic spectrum component X ′ (2f _n ). Further, the harmonic spectrum calculator 24 calculates a phase angle theta _n of the original spectral components X (f _n), twice the phase angle 2 [Theta] _n of the calculated phase angle theta _n harmonic spectral components X '( 2f _n ). Further, the harmonic spectrum calculation unit 24 calculates the magnitude (level) | X (f _n ) | of the original sound spectrum component X (f _n ), and 1 / of the calculated magnitude | X (f _n ) | The double magnitude 1/2 | X (f _n ) | is set as the magnitude of the harmonic spectrum component X ′ (2f _n ). In FIG. 4, the original sound spectrum is represented as X (f), and the harmonic spectrum is represented as X ′ (f). In the present embodiment, the weighting coefficient for applying weight to the magnitude of the harmonic spectrum component X ′ (2f _n ) is ½ (constant). When the harmonic spectrum component X ′ (2f _n ) is expressed by a mathematical expression, the following expression (2) is obtained.

上述した演算と同時に、補間演算部２６は、第１原音スペクトル成分Ｘ（ｆ_ｎ）の周波数ｆ_ｎの２倍の周波数２ｆ_ｎと、第２原音スペクトル成分Ｘ（ｆ_ｎ＋１）の周波数ｆ_ｎ＋１の２倍の周波数２ｆ_ｎ＋２との間の周波数２ｆ_ｎ＋１を有する高調波スペクトル成分Ｘ’（２ｆ_ｎ＋１）を例えば次式（３）の補間演算により求める。次式（３）において、高調波スペクトル成分Ｘ’（２ｆ_ｎ＋１）の位相角については、第１原音スペクトル成分Ｘ（ｆ_ｎ）及び第２原音スペクトル成分Ｘ（ｆ_ｎ＋１）の各位相角にそれぞれ周波数比を乗じることによって算出している。

At the same time the operation described above, the interpolation operation unit 26, the frequency _{f n} of twice the frequency 2f _n of frequency _{f n} of the first original spectral component X _{(f n),} second original spectral component _{X (f} n +1) A harmonic spectrum component X ′ (2f _n +1) having a frequency 2f _n +1 between a frequency 2f _n +2 that is twice +1 is obtained by, for example, an interpolation operation of the following equation (3). In the following equation (3), for the phase angle of the harmonic spectrum component X ′ (2f _n +1), each phase angle of the first original sound spectrum component X (f _n ) and the second original sound spectrum component X (f _n +1). Is calculated by multiplying each by the frequency ratio.

このようにして演算された高調波スペクトル成分Ｘ’（２ｆ_ｎ）、Ｘ’（２ｆ_ｎ＋１）はそれぞれ、高調波スペクトル演算部２４によって合成され、４〜８ｋＨｚの周波数帯域を有する高調波スペクトルＸ’（ｆ）が生成される（図４（ａ）参照）。

The harmonic spectrum components X ′ (2f _n ) and X ′ (2f _n +1) calculated in this way are synthesized by the harmonic spectrum calculation unit 24, and the harmonic spectrum X having a frequency band of 4 to 8 kHz. '(F) is generated (see FIG. 4A).

The harmonic spectrum X ′ _n generated as described above is added to the original sound spectrum X _n by the harmonic spectrum adding unit 14 (see FIG. 3D). Thereby, the original sound spectrum Xn in which the frequency band of 4 to 8 kHz is expanded is generated.

Harmonic spectrum X 'each _n is the addition the original spectrum X _n is the inverse Fourier transform by the inverse Fourier transform unit 16, thereby the output signal component T _n in the time _domain' is generated (see FIG. 3 (e) ). Each output signal component T _n ′ is added by the output signal generation means 18 to generate an output signal T _out (see FIG. 3F). The generated output signal T _out is converted to an analog signal by a D / A converter (not shown) and then output to the outside from the output terminal. The frequency band of the output signal T _out is 0 to 8 kHz, and the high frequency side of the frequency band 0 to 4 kHz of the input signal T _in is expanded.

  In the band extending apparatus 2 of the present embodiment, as described above, the frequency multiplied by the frequency of the original sound spectral component (twice in the present embodiment) is set as the frequency of the harmonic spectral component. The peak of the secondary resonance mode (secondary resonance mode in this embodiment) can be reproduced. Therefore, an output signal with natural characteristics can be obtained, and the degree of hearing can be improved.

  Note that the bandwidth expansion device 2 of the present embodiment can be applied to various uses. For example, when the bandwidth extension device 2 is applied to a telephone system, it becomes possible to obtain a natural call voice, thereby improving the degree of hearing when an elderly person makes a call, In addition, criminal acts (so-called transfer fraud, etc.) caused by making a call while impersonating another person can be prevented. In addition, when the band extending device 2 is applied to an artificial vocal cord for laryngectomies, a natural voice can be reproduced. In addition, when the band extending device 2 is applied to a digital audio device, natural reproduction sound quality can be obtained. Furthermore, for example, microphones, vibration pickups, bone-conduction microphones, and the like that generate input signals are difficult to output a high sound range due to insufficient characteristics, but by applying the band extending device 2 of the present embodiment to these, It becomes easy to reproduce a natural high range.

  Next, another embodiment of the band extending apparatus will be described with reference to FIGS. FIG. 5 is a block diagram showing a configuration of a band extending apparatus according to another embodiment of the present invention, and FIG. 6 is a diagram for explaining a method of analyzing the slope of the envelope of the original sound spectrum. (A) is a spectrum diagram showing an original sound spectrum whose phoneme is a vowel, FIG. 7 (b) is a spectrum diagram showing an original sound spectrum to which a harmonic spectrum is added, and FIG. 8 (a) is a phoneme. FIG. 8B is a spectrum diagram showing an original sound spectrum to which a harmonic spectrum is added. In the present embodiment, components that are substantially the same as those in the above embodiment are denoted by the same reference numerals, and the description thereof is omitted.

  Referring to FIG. 5, in the band extending apparatus 2A of the present embodiment, the harmonic spectrum generation unit 12A includes an original sound spectrum analysis unit 28 (which constitutes an original sound spectrum analysis unit) and a weighting unit 30 (which constitutes a weighting unit). Is further included. The original sound spectrum analysis unit 28 analyzes the slope of the envelope of the original sound spectrum. The weighting unit 30 applies weighting to the magnitude of the harmonic spectrum component included in the harmonic spectrum with a weighting coefficient described later, based on the slope of the envelope of the analyzed original sound spectrum. Hereinafter, the original sound spectrum analysis unit 28 and the weighting unit 30 will be described.

  The original sound spectrum analyzing unit 28 divides the frequency of the original sound spectrum component included in the original sound spectrum generated by the Fourier transform in an octave and logarithmizes the magnitude of the original sound spectrum component (see FIG. 6). Then, the slope α of the envelope of the original sound spectrum is analyzed by an approximate straight line represented by the following equation (4) by performing an operation using the least square method.

加重部３０は、上述のようにして分析された傾斜αより、次式（５）で示す加重係数Ｚ（α）を算出する。更に、加重部３０は、高調波スペクトル演算部２４による高調波スペクトル成分Ｘ’（２ｆ_ｎ）の演算の際に、次式（６）で示すように、算出された加重係数Ｚ（α）によって高調波スペクトル成分Ｘ’（２ｆ_ｎ）の大きさに重みを付ける。

The weighting unit 30 calculates a weighting coefficient Z (α) represented by the following equation (5) from the slope α analyzed as described above. Further, when the harmonic spectrum component X ′ (2f _n ) is calculated by the harmonic spectrum calculation unit 24, the weighting unit 30 uses the calculated weighting coefficient Z (α) as shown in the following equation (6). A weight is applied to the magnitude of the harmonic spectral component X ′ (2f _n ).

例えば入力信号が音声信号である場合には、原音スペクトルの音韻の種類によって、原音スペクトルの包絡線の傾斜αが変化する。図７（ａ）に示すように、原音スペクトルの音韻が母音であるときには、母音の特性上、原音スペクトルの包絡線の傾斜αはマイナス側に大きくなる。これにより、加重係数Ｚ（α）は小さくなり、図７（ｂ）に示すように、高調波スペクトル成分の大きさは小さくなる。一般に、母音のスペクトルでは、高域側におけるスペクトル成分の減衰が大きくなる特性がある。それ故に、上述のように加重係数Ｚ（α）を小さく設定することによって、母音をよりリアルに復元することができる。

For example, when the input signal is an audio signal, the slope α of the envelope of the original sound spectrum changes depending on the type of phoneme of the original sound spectrum. As shown in FIG. 7A, when the phoneme of the original sound spectrum is a vowel, the slope α of the envelope of the original sound spectrum increases to the negative side due to the characteristics of the vowel. As a result, the weighting coefficient Z (α) is reduced, and the magnitude of the harmonic spectrum component is reduced as shown in FIG. 7B. In general, the spectrum of vowels has a characteristic that the attenuation of the spectrum component on the high frequency side is increased. Therefore, by setting the weighting coefficient Z (α) to be small as described above, the vowel can be restored more realistically.

  Further, as shown in FIG. 8A, when the phoneme of the original sound spectrum is a consonant, the slope α of the envelope of the original sound spectrum is small due to the characteristics of the consonant. As a result, the weighting coefficient Z (α) decreases, and the harmonic spectrum component increases in size as shown in FIG. In general, the consonant spectrum has a characteristic that the attenuation of the spectrum component on the high frequency side is small. Therefore, the consonant can be restored more realistically by setting the weighting coefficient Z (α) large as described above.

In each of the above embodiments, the harmonic spectrum calculation unit 24 sets a frequency 2f _n twice the frequency f _n of the original sound spectrum component X (f _n ) as the frequency of the harmonic spectrum component X ′ (2f _n ). However, the frequency may be 3 times the frequency 3f _n or 4 times the frequency 4f _n , and the frequency mf _n multiplied by an arbitrary integer m is set as the frequency of the harmonic spectrum component X ′ (mf _n ). Can do. Correspondingly, the harmonic spectrum calculation unit 24, original spectral components X (f _n) harmonic spectral components X phase angle m.theta _n multiplied by any integer m of the phase angle theta _n of '(mf _n) Can be set as the phase angle.

In each of the above embodiments, for example, the first harmonic spectral component X having twice the frequency 2f _n frequency _{f n} of the original spectral components _{X (f n) '(2f} n), original spectral components X ( and a second harmonic spectral component X ′ (3f _n ) having a frequency 3f _n that is three times the frequency f _n of f _n ), and these first and second harmonic spectral components X ′ (2f _n ) , X ′ (3f _n ) may be configured to add the first and second harmonic spectra obtained by synthesis to the original sound spectrum.
[Sound correction device embodiment]
Next, an embodiment of a sound correction apparatus will be described with reference to FIGS. FIG. 9 is a block diagram showing the configuration of the sound correction apparatus according to an embodiment of the present invention, FIG. 10 is a diagram showing the correspondence between parameters stored in the parameter storage unit, and FIG. FIG. 12 is a block diagram showing a configuration of a parameter creation apparatus for creating a guided sound path characteristic parameter, and FIG. 12 is a block diagram showing a parameter creation apparatus for creating a bone-conducted sound path characteristic parameter and an air-conducted sound source characteristic parameter. FIG.

  Referring to FIG. 9, the sound correction device 32 of this embodiment includes a bone-conduction microphone 34, a low-pass filter (LPF) 36, an A / D conversion unit 38, and a time frame division unit 40 (which constitutes a time frame division unit). , LPC analysis unit 42 (which constitutes speech characteristic analysis means and speech property determination means), correction mode switching unit 44 (which constitutes correction mode switching means), and first LPC synthesis unit 46 (which constitutes first signal correction means) A second LPC synthesis unit 48 (which constitutes a first signal correction unit), a parameter storage unit 50 (which constitutes a parameter storage unit), a smoothing unit 52 (which constitutes an output signal generation unit), and a D / A conversion unit 54 The low-pass filter 56 and the output terminal 60 are included.

  The bone-conduction microphone 34 is mounted on a face part, for example, the forehead, chin, cheek, ear hole or the like, and records vocal cord vibrations of a speaker who is transmitted to bones and skin. The low-pass filter 36 attenuates a frequency band equal to or higher than a predetermined frequency (for example, 8 kHz) of the audio signal from the bone conduction microphone 34. The A / D converter 38 converts the audio signal from an analog signal to a digital signal.

  The time frame dividing unit 40 divides the time domain audio signal into time frames having a predetermined time length (for example, 16 msec) using the Hanning window as a window function. Each time frame is divided so that both end portions thereof overlap with time frames adjacent to both sides.

  The LPC analysis unit 42 performs linear prediction analysis (LPC) on the audio signal in each time frame, and analyzes the audio characteristics of the audio signal. By this analysis, the audio signal is separated into the bone-conducted sound source characteristic and the bone-conducted vocal tract characteristic. The analyzed bone conduction sound source characteristic is output to the first LPC synthesis unit 46, and the analyzed bone conduction vocal tract characteristic is sent to the first LPC synthesis unit 46 or the second LPC synthesis unit 48 via the correction mode switching unit 44. Is output.

  Moreover, the LPC analysis part 42 discriminate | determines the audio | voice property of an audio | voice signal based on the analyzed bone-conduction sound source characteristic. When a pitch component (that is, a pulse train indicating the voiced sound source characteristic) is detected in the bone-conducted sound source characteristic, the LPC analysis unit 42 determines that the audio signal is a voiced sound, and also determines the bone-conducted sound source characteristic. When the pitch component is not detected, it is determined that the audio signal is an unvoiced sound.

  The correction mode switching unit 44 is a switch for switching the output destination of the bone conduction vocal tract characteristics based on the discrimination result of the voice property of the voice signal by the LPC analysis unit 42. When the LPC analysis unit 42 determines that the voice signal is voiced, the correction mode switching unit 44 switches the output destination of the bone conduction vocal tract characteristic to the first LPC synthesis unit 46. Thus, the voiced sound correction mode in which the first LPC synthesis unit 46 corrects the voice signal is set. In addition, when the LPC analysis unit 42 determines that the audio signal is an unvoiced sound, the correction mode switching unit 44 switches the output destination of the bone-conducted vocal tract characteristic to the second LPC synthesis unit 48. As a result, an unvoiced sound correction mode in which the audio signal is corrected by the second LPC synthesis unit 48 is set.

  The parameter storage unit 50 stores a plurality of parameter groups (for example, 40 sets) with the bone conduction vocal tract characteristic parameter, the air conduction vocal tract characteristic parameter, and the air conduction sound source characteristic parameter as a set of parameter groups. (See FIG. 10). These parameters are created in advance by parameter creation devices 62 and 76, which will be described later, and stored in the parameter storage unit 50. A method for creating each parameter will be described later.

  The first LPC synthesis unit 46 performs LPC synthesis (that is, synthesis by a linear prediction method) between the bone conduction sound source characteristics output from the LPC analysis unit 42 and the air conduction vocal tract characteristic parameters stored in the parameter storage unit 50. Thus, a pseudo air conduction sound signal is generated. In addition, the second LPC synthesis unit 48 generates a pseudo air conduction sound signal by performing LPC synthesis of the air conduction sound source characteristic parameter and the air conduction sound path characteristic parameter stored in the parameter storage unit 50. A method of generating the pseudo air conduction speech signal by the first LPC synthesis unit 46 and the second LPC synthesis unit 48 will be described later.

  The smoothing unit 52 generates an output signal by adding the pseudo air conduction sound signals generated by the first LPC synthesis unit 46 or the second LPC synthesis unit 48 and performing a smoothing process. As the smoothing process, for example, a method is used in which the Hanning window is used as a window function and the amplitude value of the signal junction is approximated to zero. The D / A converter 54 converts the generated output signal from a digital signal to an analog signal. The low-pass filter 56 attenuates a frequency band of a predetermined frequency (for example, 8 kHz) or more of the output signal converted into the analog signal. The output signal generated in this way is output from the output terminal 60 to the outside.

  Next, the flow of audio signal correction by the audio correction device 32 described above will be described. First, when a speaker who wears the bone conduction microphone 34 speaks, an audio signal is output from the bone conduction microphone 34. The sound signal (analog signal) from the bone-conduction microphone 34 is attenuated in a frequency band equal to or higher than a predetermined frequency by the low-pass filter 36. Next, the audio signal is converted into a digital signal by the A / D conversion unit 38 and then divided by the time frame division unit 40 for each time frame.

  The audio signal in each time frame is separated into a bone-conducted sound source characteristic and a bone-conducted vocal tract characteristic by the LPC analyzer 42. Further, the LPC analysis unit 42 determines whether the sound signal in each time frame is a voiced sound or an unvoiced sound based on the analyzed bone conduction sound source characteristics.

When it is determined that the audio signal in the time frame is a voiced sound, the correction mode switching unit 44 switches to the voiced sound correction mode. In the voiced sound correction mode, the bone conduction sound source characteristic and the bone conduction vocal tract characteristic analyzed by the LPC analysis unit 42 are output to the first LPC synthesis unit 46, respectively. In the first LPC synthesis unit 46, a bone conduction vocal tract characteristic parameter (for example, b ₁ ) having a characteristic closest to the bone conduction vocal tract characteristic output from the LPC analysis unit 42 is selected from the parameter storage unit 50 and selected. The air conduction vocal tract characteristic parameter (for example, a ₁ ) corresponding to the bone conduction vocal tract characteristic parameter is read from the parameter storage unit 50. The read air conduction vocal tract characteristic parameter and the bone conduction sound source characteristic are LPC synthesized to generate a pseudo air conduction sound signal.

When it is determined that the audio signal in the time frame is an unvoiced sound, the correction mode switching unit 44 switches to the unvoiced sound correction mode. In this unvoiced sound correction mode, the bone conduction vocal tract characteristics analyzed by the LPC analysis unit 42 are output to the second LPC synthesis unit 48 via the correction mode switching unit 44. In the second LPC synthesis unit 48, a bone conduction vocal tract characteristic parameter (for example, b ₂ ) having a characteristic closest to the bone conduction vocal tract characteristic output from the LPC analysis unit 42 is selected from the parameter storage unit 50 and selected. The air conduction sound source characteristic parameter (for example, av ₂ ) and the air conduction sound path characteristic parameter (for example, a ₂ ) corresponding to the bone conduction vocal tract characteristic parameter are read from the parameter storage unit 50. The read air-conducted sound source characteristic parameter and air-conducted vocal tract characteristic are LPC synthesized to generate a pseudo air-conducted speech signal.

  Each of the pseudo air conduction sound signals generated as described above is added and smoothed by the smoothing unit 52, thereby generating an output signal. The generated output signal is converted into an analog signal by the D / A converter 54, and then the frequency band of a predetermined frequency or more is attenuated by the low-pass filter 56 and output to the outside from the output terminal 60.

  Note that the bone-conducted vocal tract characteristic parameter, the air-conducted vocal tract characteristic parameter, and the air-conducted sound source characteristic parameter stored in the parameter storage unit 50 are created as follows, for example. First, a method for creating an air conduction vocal tract characteristic parameter will be described. A parameter creating device 62 as shown in FIG. 11 is used to create the air conduction vocal tract characteristic parameters. The parameter creation device 62 includes an air conduction microphone 64, a low-pass filter 66, an A / D conversion unit 68, a time frame division unit 70, an LPC analysis unit 72, and a representative value selection unit 74.

  The air conduction microphone 64 is a so-called general microphone that records a voice signal of a voice of a speaker who propagates in the air. Each of the low-pass filter 66, the A / D conversion unit 68, and the time frame division unit 70 has substantially the same function as that of the audio correction device 32 described above.

  The LPC analysis unit 72 performs linear prediction analysis on the audio signal in each time frame divided by the time frame dividing unit 70, and analyzes the air conduction vocal tract characteristics of the audio signal. The representative value selection unit 74 selects a representative value as described later from the plurality of air conduction vocal tract characteristics analyzed by the LPC analysis unit 72, and sets the representative value as an air conduction vocal tract characteristic parameter.

  The flow of creating the air conduction vocal tract characteristic parameters is as follows. First, a speaker speaks a vocabulary or a sentence in which all features are expressed as an audio signal, such as 100 Japanese city names. The uttered voice is input to the air conduction microphone 64. The audio signal from the air conduction microphone 64 is converted into a digital signal via the low pass filter 66 and the A / D conversion unit 68. Thereafter, the audio signal is divided for each time frame by the time frame dividing unit 70, and is output to the LPC analysis unit 72. The LPC analysis unit 72 performs linear prediction analysis on the speech signal in each time frame, and analyzes the air conduction vocal tract characteristics of the speech signal. This analyzed air conduction vocal tract characteristic is classified into one of a plurality of characteristic groups. The plurality of characteristic groups are for classifying the air conduction vocal tract characteristics having similar properties, and a plurality of air conduction vocal tract characteristics having similar properties belong to each characteristic group. The representative value selection unit selects one air conduction vocal tract characteristic as a representative value from each characteristic group. The selected representative value is set as an air conduction vocal tract characteristic parameter and stored in the parameter storage unit 50.

  Next, a method for creating a bone-conducted vocal tract characteristic parameter and an air-conducted sound source characteristic parameter will be described. A parameter creation device 76 as shown in FIG. 12 is used to create the bone-conducted vocal tract characteristic parameter and the air-conducted sound source characteristic parameter. The parameter creation device 76 includes a bone conduction microphone 78, a low pass filter 80, an A / D conversion unit 82, a time frame division unit 84, an LPC analysis unit 86, an air conduction microphone 88, a low pass filter 90, an A / D conversion unit 92, A time frame division unit 94, an LPC analysis unit 96, a parameter assignment unit 98 and an averaging unit 100 are included.

  The low-pass filters 80 and 90, the A / D conversion units 82 and 92, and the time frame division units 84 and 94 each have substantially the same function as that of the audio correction device 32 described above. The LPC analysis unit 86 performs linear prediction analysis on the audio signal in each time frame divided by the time frame dividing unit 84, and analyzes the bone-conducted vocal tract characteristics of the audio signal. The LPC analysis unit 96 performs linear prediction analysis on the audio signal in each time frame divided by the time frame dividing unit 94, and analyzes the air conduction sound path characteristics and the air conduction sound source characteristics of the sound signals. The parameter allocating unit 98 creates a plurality of parameter groups including bone-conducted vocal tract characteristic parameters, air-conducted vocal tract characteristic parameters, and air-conducted sound source characteristic parameters as will be described later. The averaging unit 100 performs an averaging process when there are overlapping parameters among a plurality of sets of parameter groups.

  The flow of creating the bone-conducted vocal tract characteristic parameter and the air-conducted sound source characteristic parameter will be described as follows. First, in the same manner as described above, a speaker speaks a vocabulary or a sentence in which all features are expressed as an audio signal, for example, 100 Japanese city names. The uttered voice is input to the bone conduction microphone 78 and the air conduction microphone 88 at the same time. The audio signal from the bone-conduction microphone 78 is converted into a digital signal via the low-pass filter 80 and the A / D conversion unit 82. Thereafter, the audio signal is divided for each time frame by the time frame dividing unit 84 and output to the LPC analyzing unit 86. The LPC analysis unit 86 performs linear prediction analysis on the speech signal in each time frame, and analyzes the bone-conducted vocal tract characteristics of the speech signal.

  The audio signal from the air conduction microphone 88 is converted into a digital signal via the low-pass filter 90 and the A / D converter 92. Thereafter, the audio signal is divided for each time frame by the time frame dividing unit 94 and output to the LPC analyzing unit 96. The LPC analysis unit 96 performs linear prediction analysis on the speech signal in each time frame, and analyzes the air conduction vocal tract characteristics and the air conduction sound source characteristics of the speech signal.

In the parameter assigning unit 98, the air conduction vocal tract characteristic parameter created as described above is collated with the air conduction vocal tract characteristic in each time frame, and the air conduction vocal tract characteristic in the time frame (for example, T ₁ ) The air conduction vocal tract characteristic parameter (for example, a ₁ ) having the closest characteristic is selected. Next, the parameter assigning unit 98 selects the air conduction vocal tract characteristic parameters selected as described above for the bone conduction vocal tract characteristic and the air conduction sound source characteristic analyzed in the same time frame (for example, T ₁ ). (e.g., a ₁₎ the same parameter number (i.e., for example, a number "n" in a _n) bone-conducted vocal tract characteristic parameter with (e.g., b ₁₎ and air conduction sound source characteristic parameters (e.g. av ₁₎ is allocated respectively . Thus, by assigning the parameter numbers of the air conduction vocal tract characteristic parameters to the bone conduction vocal tract characteristics and the air conduction sound source characteristics for each time frame, the bone conduction vocal tract characteristic parameters and the air conduction sound source characteristic parameters are assigned. A plurality of parameter groups are created.

If such processing is performed over the entire time frame, for example, a plurality of bone-conducted vocal tract characteristic parameters b ₁ and air-conducted sound source characteristic parameters av ₁ may appear. In such a case, a plurality of bone-conducted vocal tract characteristic parameters b ₁ and a plurality of air-conducted sound source characteristic parameters av ₁ are averaged by the averaging unit 100, respectively. One characteristic parameter b ₁ and _one air conduction sound source characteristic parameter av ₁ are created. As described above, the bone-conducted vocal tract characteristic parameter and the air-conducted sound source characteristic parameter associated with each air-conducted vocal tract characteristic parameter are stored in the parameter storage unit 50 (see FIG. 10).

  Although various embodiments of the band extending apparatus and the sound correcting apparatus according to the present invention have been described above, the present invention is not limited to such embodiments, and various modifications or corrections can be made without departing from the scope of the present invention. Is possible.

2,2A Band Expansion Device 8,40 Time Frame Dividing Unit 10 Fourier Transform Unit 12, 12A Harmonic Spectrum Generation Unit 14 Harmonic Spectrum Addition Unit 16 Inverse Fourier Transform Unit 18 Output Signal Generation Unit 26 Interpolation Operation Unit 28 Original Sound Spectrum Analysis Unit DESCRIPTION OF SYMBOLS 30 Weighting part 32 Audio | voice correction | amendment apparatus 34 Bone-conduction microphone 42 LPC analysis part 44 Correction mode switching part 46 1st LPC synthetic | combination part 48 2nd LPC synthetic | combination part 50 Parameter memory | storage part 52 Smoothing part

Claims (9)

A band extending device for extending the frequency band of an input signal,
Time frame dividing means for dividing the time domain input signal into time frames; Fourier transform means for generating a frequency domain original sound spectrum by Fourier transforming the time frame; and generating a harmonic spectrum based on the original sound spectrum A harmonic spectrum generating means for adding the harmonic spectrum to the original sound spectrum, and inverse Fourier transforming the original sound spectrum to which the harmonic spectrum has been added to obtain an output signal component in the time domain. An inverse Fourier transform means for generating, and an output signal generation means for adding an output signal component to generate an output signal with an expanded frequency band.
The harmonic spectrum generation means calculates the frequency of the original sound spectrum component included in the original sound spectrum, and sets the frequency multiplied by the calculated frequency as the frequency of the harmonic spectrum component included in the harmonic spectrum. A bandwidth extension device.   The harmonic spectrum generating means calculates a phase angle of the original sound spectrum component included in the original sound spectrum, and a phase angle obtained by multiplying the calculated phase angle is a phase of the harmonic spectrum component included in the harmonic spectrum. The band extending apparatus according to claim 1, wherein the band extending apparatus is set as a corner.   The harmonic spectrum generation unit is configured to analyze the original sound spectrum with respect to the magnitude of the harmonic spectrum component included in the harmonic spectrum based on the analysis result of the original sound spectrum analysis unit. The band extending apparatus according to claim 1, further comprising a weighting unit that applies weighting with a predetermined weighting factor.   The original sound spectrum analyzing means is configured to analyze an inclination of an envelope of the original sound spectrum, and the weighting means changes the weighting factor based on the analyzed inclination of the envelope of the original sound spectrum. The band extending apparatus according to claim 3.   The harmonic spectrum generation means includes interpolation means for interpolating the harmonic spectrum component, and the interpolation means has a frequency multiplied by a frequency of a first original sound spectrum component included in the original sound spectrum, and 5. The harmonic spectrum component having a frequency between a frequency multiplied by a frequency of a second original sound spectrum component adjacent to the first original sound spectrum component is interpolated. Bandwidth expansion device.   6. The input signal according to claim 1, wherein the input signal is a voice signal or a musical tone signal generated by a microphone, a vibration pickup, a bone conduction microphone, a digital acoustic device, a telephone system, an artificial vocal cord, or the like. Bandwidth expansion device.   The bone conduction microphone, the time frame dividing means for dividing the sound signal from the bone conduction microphone into time frames, the sound characteristic analyzing means for analyzing the sound characteristics of the sound signal in the time frame, and the analysis result of the sound characteristics A sound property determining means for determining whether the sound signal in the time frame is voiced sound or unvoiced sound, first signal correcting means for correcting the sound signal determined to be voiced sound and generating a pseudo air conduction sound signal; A second signal correcting means for correcting a voice signal determined to be an unvoiced sound to generate a pseudo air conduction voice signal; a voiced sound correcting mode by the first signal correcting means; and an unvoiced sound correcting mode by the second signal correcting means. Correction mode switching means for switching, and output signal generation means for adding the generated pseudo air conduction sound signals to generate an output signal. Voice correction apparatus characterized. The apparatus further comprises parameter storage means for storing a bone-conducted vocal tract characteristic parameter, an air-conducted vocal tract characteristic parameter, and an air-conducted sound source characteristic parameter. And configured to analyze bone-conducted vocal tract characteristics,
The first signal correction means reads out the air conduction vocal tract characteristic parameter corresponding to the bone conduction vocal tract characteristic of the sound signal analyzed by the sound characteristic analysis means from the storage means, and the sound characteristic 8. The speech correction apparatus according to claim 7, wherein a pseudo air conduction sound signal is generated by synthesizing the bone conduction sound source characteristic of the sound signal analyzed by the analysis means and the air conduction sound path characteristic parameter.   The second signal correction unit is configured to determine the air conduction sound source characteristic parameter and the air conduction sound source characteristic parameter corresponding thereto based on the bone conduction vocal tract characteristic of the sound signal analyzed by the sound characteristic analysis unit. 9. The speech correction apparatus according to claim 8, wherein the speech correction apparatus generates a pseudo air conduction sound signal by reading from the storage means and combining the read air conduction sound path characteristic parameter and the air conduction sound source characteristic parameter. .

Images