JP2004078232A - Method and device for restoring wide-band voice, voice transmission system, and voice transmission method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To obtain a wide-band voice restoring device which restores a wide-band voice signal of high quality by estimating a stable wide-band sound source with a correct amplitude from narrow-band voices or narrow-band voice codes under less influence of differences of speakers and noise. <P>SOLUTION: The wide-band voice restoring device is provided with a narrow-band sound source decoding means which uses narrow-band voice codes to generate narrow-band synthesized sounds, a spectrum decoding means which uses narrow-band spectral codes separated from narrow-band voice codes to estimate wide-band spectral parameters, a wide-band sound source decoding means which uses narrow-band sound source codes separated from narrow-band voice codes to estimate a wide-band sound source signal, and a synthesizing means which generates a wide-band voice signal from the generated narrow-band synthesized sounds, the estimated wide-band spectral parameters, and the wide-band sound source signal. <P>COPYRIGHT: (C)2004,JPO

Description

The present invention relates to a wideband speech restoration apparatus for restoring a wideband speech signal from a narrowband speech signal whose band is limited or a narrowband speech code obtained by encoding the narrowband speech signal.

One example of a narrowband audio signal is the current telephone audio. In the telephone system, the audio signal is transmitted in a limited band of about 300 Hz to 3.4 KHz, and the sound quality is poor and crowded as compared with the case where there is no band limitation. To improve the quality, a telephone system capable of transmitting a wideband voice signal may be constructed, but much time and expense are required.

Japanese Patent Laid-Open No. 6-118995 discloses a conventional method for restoring a wideband speech signal from a narrowband speech limited to a telephone band.

Japanese Patent Application Laid-Open No. 6-118995 discloses an LPC analysis of a narrowband speech signal to calculate a spectrum parameter, and vector-quantizes the spectrum parameter using a narrowband codebook. Then, broadband spectral parameters are decoded using the wideband codebook learned in association with the narrowband codebook. An LPC synthesis process is performed using the spectrum parameters to obtain a temporary wideband audio signal. A final wideband audio signal is generated by extracting and adding a band component other than the narrowband audio signal from the temporary wideband audio signal to an upsampled narrowband audio signal. Note that when performing wideband LPC synthesis processing, a wideband sound source signal is required, but a method for generating the sound source signal is not specifically disclosed.

As a document having the same configuration as that of Japanese Patent Application Laid-Open No. 6-118995, and which discloses a wide-band sound source signal generation, reference 1 "Reconstruction of wide-band speech from narrow-band speech by codebook mapping" IEICE, IEICE There is SP93-61 (1993-08).

文献 In this document 1, two methods are disclosed as a wideband sound source generation method.

The first method uses a pitch and power obtained by analyzing a narrowband voice to generate a sound source by a general method among those skilled in the art. That is, an impulse train that repeats at a pitch cycle is generated for voiced sounds, and white noise is generated for unvoiced sounds, and the amplitude is determined by power.

In Reference 1, some post-processing is performed to improve sound quality. When restoring a low frequency of 300 Hz or less, the power of the low frequency restored sound is multiplied by a low multiple in order to compensate for the power shortage in the restored band. When restoring a high band from 3.4 Hz to 7.3 KHz, a cosine function is applied so as to crush the pulse in order to reduce a pulse-like sound generated by using an impulse train as a sound source.

{Circle around (2)} In the second method, the spectral parameters of the narrowband speech signal are vector-quantized, and a narrowband representative waveform element and a highband representative waveform element corresponding to the obtained code are selected. Then, the following processing is performed on these two waveform segments. The voiced and unvoiced waveform segments are determined, and in the case of voiced sounds, the waveform segments are superimposed in synchronization with a pitch obtained by analyzing a narrowband audio signal. In the case of an unvoiced sound, a signal of a required length is cut out from a random position of a waveform element. The power ratio between the synthesized sound and the narrow-band sound synthesized using the signal generated from the narrow-band waveform element by the above processing and the narrow-band spectral parameter is calculated. Then, a synthesized sound is generated from the high-band waveform element using the signal generated by the above processing and the broadband spectral parameters, and a high-frequency restored signal is obtained by multiplying the synthesized sound by the power ratio.

Although the field of use is different, as another method for expanding the band of the sound source signal, reference 2 `` A 2.4 Kbps High-Qaulity Speech Coder '' IEEE International Conference on Acoustics, Speech, and Signal Processing vol.1, S9.5, pp. 589-592 (1991.5).

Reference 2 relates to a method for encoding and decoding telephone band voice with high efficiency. In order to reduce the amount of information of a sound source at the time of encoding, a long-period prediction analysis of a sound source signal from 0 Hz to 3.4 KHz is performed. Then, the signal is separated into a long-period prediction coefficient and a long-period prediction residual signal. Encoding is performed by band-limiting the long-period prediction residual signal from 0 Hz to 3.4 kHz from 0 Hz to 1 kHz. Then, after decoding, a long-period prediction residual signal of the telephone band up to 3.4 KHz is generated from the band-limited long-period prediction residual signal, and a long-period synthesis process is performed to restore the sound source signal. It is. Restoration of a long-period prediction residual signal is performed by up-sampling a signal having a component of 0 Hz to 1 KHz to a sampling frequency of 8 KHz, leaving the signal at intervals of 4 samples, and setting the rest to zero.
JP-A-06-118995 JP-A-63-034600 JP 05-297988 A

従 来 The above conventional methods have the following problems.

Although it is a different document from Japanese Patent Application Laid-Open No. Hei 6-118995, Document 1 which discloses a concrete practical example thereof is roughly divided into the following four problems, namely, a sound source amplitude estimation, a sound source generation method, and a spectrum parameter estimation method. However, there is a problem regarding application to communication systems.

First, the first sound source amplitude estimation will be described.

When the first sound source generation method of Document 1 is used, the power value used for synthesizing the reconstructed sound uses the power value obtained by analyzing the narrow-band sound as it is or by multiplying it by a constant. Since the gain of the synthesis filter is different between the spectral parameter of (1) and the estimated broadband spectral parameter, the amplitude of the synthesized sound obtained differs even when the same sound source amplitude is given. Since this difference changes for each frame, there is a problem that broadband speech having a correct amplitude cannot be restored by multiplying the sound source amplitude, that is, the power value by a constant.

Further, when the second sound source generation method of Document 1 is used, a narrow-band synthesized sound is generated, a power ratio with respect to the narrow-band sound is calculated, and the high-band synthesized sound is multiplied. There is a problem that it is necessary to perform complicated processing for

Next, the second sound source generation method will be described.

場合 When the first sound source generation method of Document 1 is used, a wide band sound source signal is generated using only a small amount of information, such as pitch and power, and thus it is not possible to sufficiently estimate an original wide band sound source that changes in various ways. As a result, although the pulse-like sound is reduced by the cosine function, the pulse-like sound cannot be completely suppressed, and there is a problem that the sound quality becomes unnatural. In addition, since it is impossible to express a voiced sound source having different characteristics for each speaker with one fixed sound source, there is a problem that sound quality is degraded by each speaker.

When the second sound source generation method of Document 1 is used, a representative waveform segment corresponding to the sign of the vector quantization result of the spectrum parameter is used. However, the spectrum parameter originally depends on the shape of the vocal tract, and the sound source waveform is Since it depends on the manner of vocal cord vibration, there is no strong correspondence between the two. The sound source waveform depends largely on the speaker. Therefore, there is a problem that an appropriate sound source cannot be selected.

As described in Document 1, when the second sound source generation method is used, a waveform unit of an unvoiced sound is selected in spite of being a voiced sound. Regardless, a voiced waveform segment may be selected, and if synthesized as it is, there is a problem that quality degradation occurs. In order to avoid this, the power ratio at that portion is forcibly set to 0. As a result, however, the restored high-frequency amplitude becomes partially 0, which causes another quality deterioration. is there.

Furthermore, in either of the sound source generation methods, there is a problem that quality degradation due to voiced / unvoiced determination and pitch extraction error is inevitable. In particular, when applied to a narrow-band audio signal on which noise is superimposed, there is a problem that determination errors and extraction errors increase, and large degradation occurs.

Also, since there are only two modes, voiced and unvoiced, there is a problem that sound sources having intermediate properties cannot be sufficiently expressed, and quality degradation occurs at the boundary between voiced and unvoiced.

Next, the third spectral parameter estimation method will be described.

In Japanese Patent Application Laid-Open No. Hei 6-118995 and Reference 1, vector quantization and inverse quantization using two codebooks are performed, but a memory for storing codebooks is required, The problem is that a large amount of computation is required.

雑 音 Also, noise, unvoiced sound, and voiced sound can be easily distinguished by power, and the correspondence between the narrow-band spectral parameter and the wide-band spectral parameter changes depending on the distinction. However, in each case, since the spectrum parameter and the power are treated independently, information on the power is not reflected in the estimation of the spectrum parameter in a wide band. For this reason, if the shapes of the narrow-band spectra are similar, there is a problem that a similar wide-band spectrum is estimated regardless of the magnitude of the power.

Lastly, application to the fourth communication system will be described.

When applying the method of Japanese Patent Application Laid-Open No. 6-118995 and Reference 1 to a communication system, it is necessary to decode a narrow-band synthesized sound from a received speech code and then re-analyze the narrow-band synthesized sound to restore a wide-band speech signal. However, when the spectrum parameters and the sound source information are transmitted after being separated / encoded, it is considered more efficient to directly use the speech code to restore the wideband speech signal. In other words, the method disclosed in Japanese Patent Application Laid-Open No. 6-118995 and Reference 1 is inefficient in that reanalysis is required. Also, distortions due to interpolation at the time of synthesis and windowing at the time of analysis are superimposed on parameters obtained by performing the synthesis and re-analysis, and there is a deterioration in quality of wideband speech.

In addition, in Japanese Patent Application Laid-Open No. Hei 6-118995 and Reference 1, since a signal processing process which is generally introduced to reduce noise in synthesized sounds and improve intelligibility is not added, the restored wideband audio signal There is a problem that cannot be improved when the sound quality is insufficient.

In addition, when applied to a communication system, signal processing may be applied to a narrow-band synthesized sound, and in order to superimpose a processed narrow-band audio signal and an unprocessed wide-band audio signal, both of the signals are processed. There is a problem that the continuity of sound quality deteriorates.

In the method of Document 2, if the frequency band from 0 Hz to 1 KHz is considered to be a narrow band and the frequency band from 0 Hz to 3.4 KHz is considered to be a wide band, a wideband sound source signal estimation is performed. The parameters obtained by analyzing this are encoded and decoded to obtain a wideband synthesized sound, and a wideband speech signal is restored from a narrowband speech signal or a parameter extracted from a narrowband speech signal. It does not disclose a method for doing so.

The embodiment described below has been made to solve such a problem, and has as its object to realize a wideband audio restoration apparatus for restoring a wideband audio signal having a more correct amplitude from narrowband audio.

It is another object of the present invention to realize a wideband sound restoration apparatus having a relatively simple process of estimating a wideband sound source amplitude.

Further, it aims at realizing a wideband speech restoration apparatus which has little dependence on a speaker, estimates a good wideband sound source even near a voiced / unvoiced boundary, and restores a wideband speech with stable and natural sound quality.

It is another object of the present invention to realize a wideband speech restoration apparatus which is less affected by voiced unvoiced decision errors and pitch extraction errors which are likely to occur in a narrowband speech signal on which noise is superimposed.

Furthermore, it is an object of the present invention to realize a wideband speech restoration apparatus that efficiently restores wideband speech without performing reanalysis when applied to a communication system.

Furthermore, when the sound quality of the reconstructed wideband audio signal is insufficient, it can be improved, and when the signal processing is applied to the narrowband synthesized sound, the processed wideband audio signal having good narrowband continuity. The purpose of the present invention is to realize a wideband audio restoration apparatus that can obtain the above.

A broadband speech restoration apparatus according to the present invention decodes a narrowband spectral parameter from a narrowband spectral code, extends a spectrum envelope of the decoded narrowband spectral parameter in the frequency axis direction, and outputs the result as a wideband spectral parameter. When,
It is characterized by comprising a synthesizing means for generating a wideband speech signal using the outputted wideband spectrum parameter.

Embodiment 1 FIG.
An embodiment of the present invention will be described with reference to the drawings.

This embodiment mainly describes a configuration and an operation for restoring the generation of a broadband sound source signal in a more correct manner.

FIG. 1 is a configuration diagram of a wideband audio restoration apparatus according to Embodiment 1 of the present invention. In the figure, 1 is an input narrowband speech signal, 2 is an analyzing means, 3 is a spectrum analyzing means, 4 is a narrowband spectral parameter, 5 is an inverse filter, 6 is a narrowband sound source signal, 7 is a wideband spectral estimating means, 8 Is a vector quantization means, 9 is a narrowband spectrum codebook, 10 is a spectrum code, 11 is an inverse quantization means, 12 is a wideband spectrum codebook, and 13 is a wideband spectrum parameter. 14 is a broadband sound source estimating means which is an important new component in the present embodiment, 15 is a zero filling means as a specific example thereof, 16 is a wideband sound source signal, 17 is a synthesizing filter as a synthesizing means, 18 is a bandpass filter, 19 is an upsampling means, and 20 is a wideband audio signal.

FIG. 2 is a signal explanatory diagram for explaining the process of the zero filling means 15.

The operation of the first embodiment of the present invention will be described below with reference to FIGS.

First, a narrow-band audio signal 1 sampled at, for example, 8 KHz and limited to a telephone band of 300 Hz to 3.4 KHz is input to the analysis unit 2 and the up-sampling unit 19. The spectrum analysis means 3 in the analysis means 2 analyzes the narrowband speech signal 1 to calculate a narrowband spectrum parameter 4 and outputs the same to the inverse filter 5 and the wideband spectrum estimation means 7 in the analysis means 2. As the narrow-band spectrum parameter 4, various parameters such as a linear prediction coefficient, an LSP, a PARCOR coefficient, and a cepstrum can be applied. The inverse filter 5 inversely filters the narrowband audio signal 1 using the narrowband spectral parameter 4 and outputs the obtained narrowband sound source signal 6 to the wideband sound source estimating means 14.

The vector quantization means 8 in the wideband spectrum estimating means 7 performs vector quantization of the narrowband spectral parameters 4 using the narrowband spectral codebook 9 and converts the obtained spectrum code 10 into the inverse quantization in the wideband spectrum estimating means 7. Output to the converting means 11. The inverse quantization means 11 inversely quantizes the spectrum code 10 using the wideband spectrum codebook 12 and outputs the obtained wideband spectrum parameters 13 to the synthesis filter 17.

The processing in the wideband spectrum estimating means 7 is the same as that in the literature 1, and detailed description on the method of generating the narrowband spectral codebook 9 and the wideband spectral codebook 12 and the method of vector quantization will be omitted.

The zero-filling means 15 in the wide-band sound source estimating means 14, which is an important part of the present embodiment, inserts M-1 samples between each sample value of the narrow-band sound source signal 6, and obtains M times the number of samples obtained. Is output to the synthesis filter 17 as the broadband sound source signal 16. Here, M is a value obtained by dividing the sampling frequency of the wideband audio signal to be restored by the sampling frequency of the narrowband audio signal. In this embodiment, the case where M is 2 will be described. FIG. 2A shows a narrowband excitation signal 6 of N samples. When this signal is subjected to zero padding processing by the zero padding means 15, M-1 samples, that is, zeros for each sample are inserted between each sample, and a 2N sample broadband sound source shown in FIG. A signal 16 is obtained. When the zero padding process is performed with M being 2, a spectrum symmetrical from 0 Hz to 4 KHz with half the frequency of the sampling frequency of the wideband audio signal, that is, 4 KHz, is restored from 4 KHz to 8 KHz.

The synthesis filter 17 performs synthesis filter processing on the broadband sound source signal 16 using the wideband spectral parameters 13 to generate a temporary wideband audio signal. The band-pass filter 18 performs band-pass filtering on the provisional wide-band audio signal, and extracts components other than the band in which the narrow-band audio component exists. When the band of the wideband audio signal is from 0 Hz to 7.3 kHz, components of 0 Hz to 300 Hz and components of 3.4 kHz to 7.3 kHz are extracted.

The up-sampling unit 19 up-samples the narrowband audio signal 1 by M times. The signal generated by the upsampling has the same sampling frequency as the wideband audio signal 20 and has the same narrowband component as the narrowband audio signal 1. Then, the output of the bandpass filter 18 and the output of the upsampling means 19 are added to generate a wideband audio signal 20.

Since the narrow-band sound source signal and the wide-band sound source signal originally reflect the characteristics of the sound source signal generated from the same vocal organ, the strength of the harmonic component of the pitch frequency and the strength of the noise component between the harmonic components are increased. There is a correlation in the characteristics of the sound source signal, such as In other words, if the narrowband sound source signal has a regular characteristic in which the harmonic component of the pitch frequency is strong, the broadband sound source signal also has a regular characteristic in which the harmonic component of the pitch frequency is strong. Conversely, when the narrowband sound source signal has a characteristic that the noise component is strong, the wideband sound source signal also has the characteristic that the noise component is strong.

By configuring the wideband sound source estimating means as in this embodiment, it is possible to generate a wideband sound source signal of 0 to 8 kHz having the same characteristics as the narrow band sound source signal of 0 to 4 kHz in the low frequency range. There is an effect that a wideband sound with stable and natural sound quality can be restored with little dependency on the sound.

Also, unlike the conventional example, voiced unvoiced judgment and pitch extraction are not required, and a sound source having an intermediate characteristic can be naturally expressed by this configuration. It is possible to estimate a good wideband sound source even near a voiced / unvoiced boundary without the influence of a pitch extraction error and a voiced unvoiced boundary.

Embodiment 2. FIG.
FIG. 3 is a configuration diagram of the sound source estimating means 14 in the wideband audio restoration apparatus according to the second embodiment of the present invention. In the figure, the new parts are 21 sound source analysis means, 22 narrow band adaptive codebooks, 23 distortion minimizing means, 24 narrow band drive excitation signals, 25 narrow band adaptive lag lengths, 26 narrow band adaptive gains. , 27 broadband driving excitation estimating means, 28 zero filling means, 29 wideband driving excitation signal, 30 wideband adaptive excitation estimating means, 31 wideband adaptive excitation codebook, 32 wideband adaptive excitation signal, 33 wideband adaptive lag Long, a wideband adaptive gain of 34. Since the entire configuration is the same as in FIG. 1, the description of the configuration and the operation of the parts other than FIG. 3 will be omitted.

According to this configuration, the broadband sound source signal can be restored even better.

Hereinafter, the operation of the embodiment of the present invention will be described with reference to FIG.

The narrow band sound source signal 6 is input to the sound source analyzing means 21 in the wide band sound source estimating means 14. The past narrowband excitation signal 6 is stored in the narrowband adaptive codebook 22 in the excitation analysis unit 21, and the lag length is an integer according to the lag length sequentially output by the distortion minimization unit 23 described later. In this case, a signal obtained by repeating the stored past narrowband sound source signal 6 with this lag length is output. If the lag length is a non-integer value, see Reference 3 “Pitch Predictors with High Temporal Resolution” IEEE International Conference on Acoustics, Speech, and Signal Processing vol.2, S12.6, pp.661-664 (1990.4) Generate and output a signal with the polyphase filter output as described. The length of the signal to be output is the same as that of the current narrowband sound source signal 6.

FIG. 4 shows an example of the past narrowband excitation signal 6 stored in the narrowband adaptive codebook 22 and a signal output according to the input lag length.

In the figure, the horizontal axis indicates time and elapse of time in the direction of the arrow. (A1) and (B1) thus indicate the temporal length of the sound source signal, (A2) and (B2) indicate the lag length normalized to the output time, such as 20 to 128, A3) and (B3) show examples of output sound source signals.

FIG. 4A shows a case where the length of the output signal is shorter than the lag length. In this case, the sound source signal (A3) having the length of the output signal time T1 from the beginning of the lag length follows the past sound source signal. Output. When the lag length is shorter than the length of the output signal, such as T2, the same sound source signal (B3) is repeated a plurality of times as shown in FIG.

The distortion minimizing means 23 sequentially outputs a plurality of lag length values to the narrowband adaptive codebook 22, and multiplies a signal output from the narrowband adaptive codebook 22 for each lag length by a gain. The gain is determined so that the distortion between the signal and the narrow-band sound source signal 6 is minimized. Then, a lag length that minimizes distortion among all lag lengths is selected and output to the wideband adaptive sound source estimating means 30 as a narrowband adaptive lag length 25. Further, the gain value at that time is output to the wideband adaptive excitation estimating means 30 as a narrowband adaptive gain 26, and a signal obtained by multiplying the signal output from the narrowband adaptive codebook 22 by the narrowband adaptive gain 26 and the narrowband excitation signal 6 Is output to the wide band drive sound source estimating means 27 as the narrow band drive sound source signal 24. As a method of determining the gain in the distortion minimizing means 23, a generally known Lagrange's undetermined coefficient method can be used.

That is, the distortion minimizing means 23 receives the narrow-band excitation signal 6 and the output of the narrow-band adaptive codebook 22 as inputs, and provides a minimum distortion lag length 25 and a gain 26, which are narrow-band adaptive excitation codes, and a narrow-band driving excitation of the error signal. The signal 24 is output.

The zero-filling means 28 in the wide-band driving sound source estimating means 27 inserts M-1 samples of zero between each sample value of the narrow-band driving sound source signal 24, and converts the obtained signal having M times the number of samples into the wide-band driving sound source signal 24. Output as signal 29. Here, M is a value obtained by dividing the sampling frequency of the wideband audio signal to be restored by the sampling frequency of the narrowband audio signal, and the operation of inserting zero is the same as that of the zero-filling means 15.

In the wideband adaptive sound source estimating means 30, first, the narrowband adaptive lag length 25 is multiplied by M to obtain a wideband adaptive lag length 33, and the narrowband adaptive gain 26 is multiplied by g to obtain a wideband adaptive gain 34. Assuming that g is 1, the pitch periodicity of the finally obtained broadband excitation signal 16 is equivalent to that of the narrowband excitation signal 6, and the pitch periodicity becomes weaker as compared with the narrowband excitation signal 6 as it is reduced from 1. Go. When observing the actual sound, the pitch periodicity is likely to be weaker as the frequency becomes higher. Therefore, when restoring the high frequency, setting g to a value smaller than 1 enables a higher-quality wideband sound to be restored. .

The wideband adaptive excitation codebook 31 in the wideband adaptive excitation estimation means 30 stores the past wideband excitation signal 16, and outputs a signal obtained by repeating this signal with the wideband adaptive lag length 33. The signal is multiplied by the wideband adaptive gain 34 in the wideband adaptive sound source estimating means 30 and output as a wideband adaptive sound source signal 32.

(4) Finally, the broadband drive excitation signal 29 and the broadband adaptive excitation signal 32 are added and output as the broadband excitation signal 16.

With such a configuration, the characteristics related to the strength and fluctuation of the pitch periodicity of the narrow-band excitation signal are well represented by the narrow-band adaptive lag length 25 and the narrow-band adaptive gain 26, and are reflected in the wide-band excitation signal. Therefore, there is an effect that it is possible to sufficiently estimate a sound source that changes in various ways and to restore a broadband sound having good sound quality without pulse-like sound. Further, there is an effect that an appropriate sound source can be estimated without depending on a speaker.

In the wideband adaptive excitation signal 32, the fundamental frequency determined by the wideband adaptive lag length 33 and the frequency of its harmonic component are correctly aligned at integer multiple positions, so that the narrowband component in the finally recovered wideband audio signal 20 There is an effect that the connection between the restored broadband components is good and high-quality wideband speech can be restored.

Furthermore, since a feature that the pitch periodicity becomes weaker as the frequency becomes higher can be introduced by the coefficient g, a more natural sound quality can be obtained.

In addition, since voiced unvoiced judgment and pitch extraction are not required and a sound source of an intermediate property can be expressed, there is no influence of voiced unvoiced judgment error or pitch extraction error which is likely to occur on a narrowband audio signal on which noise is superimposed, A good wideband sound source can be estimated even near a voiced / unvoiced boundary, and there is an effect that a wideband sound with stable and natural sound quality can be restored.

Embodiment 3 FIG.
FIG. 5 is a block diagram of the wideband driving sound source estimating means 27 in the wideband audio restoration apparatus according to the third embodiment of the present invention. In the figure, new parts are a power calculating means 35 and a noise generating means 36. Other configurations are the same as those in FIG. 1 and FIG.

Hereinafter, the operation of the portion shown in the drawing of the third embodiment of the present invention will be described with reference to FIG.

The narrow-band driving sound source signal 24 is input to the power calculating means 35 in the wide-band driving sound source estimating means 27. The power calculator 35 calculates and outputs the power of the narrow-band drive sound source signal 24. The noise generator 36 generates and outputs a power-normalized white noise signal. Then, the white noise signal is multiplied by the power output from the power calculator 35 in the broadband drive sound source estimator 27, and the obtained signal is output as the broadband drive sound source signal 29.

The pitch period and the strength of the periodicity change every moment. Since the fine fluctuation of the pitch period and the intensity of the periodicity in the narrow band excitation signal 6 cannot be expressed by the narrow band adaptive lag length 25 and the narrow band adaptive gain 26, the error is included in the narrow band driving excitation signal 24. . When the wide-band drive excitation signal 29 is generated using the narrow-band drive excitation signal 24 including this error component as in the second embodiment, unnecessary disturbance may occur in the wide-band drive excitation signal 29, and the power may be the same. It has been experimentally confirmed that a better restored sound may be obtained when white noise is generated and used as the broadband driving sound source signal 29.

With the configuration as in the third embodiment, the white noise having the same power as that of the narrow-band drive excitation signal 24 is generated and used as the broad-band drive excitation signal 29. In addition to the effects of the second embodiment, the pitch period There is an effect that a good restored sound with little disturbance due to fluctuations in the intensity of the periodicity is obtained.

と Further, when the zero padding process is performed, a spectrum symmetric about 4 KHz is generated. Therefore, if zero-padding is performed on the narrow-band drive sound source signal 24 having no components of 0 Hz to 300 Hz and 3.4 kHz to 4.0 kHz, 0 Hz to 300 Hz, 3.4 kHz to 4.6 kHz, and 7.7 kHz to 8 kHz. A signal having no component is obtained. On the other hand, in this configuration using white noise, the broadband driving sound source signal 29 having all components from 0 Hz to 8 KHz is obtained, so that there is an effect that a good restored sound having a band over the entire region can be obtained. In particular, the effect is great when restoring from 0 Hz to 300 Hz.

Embodiment 4. FIG.
FIG. 6 is a configuration diagram of the wideband driving sound source estimating means 27 in the wideband audio restoration apparatus according to the fourth embodiment of the present invention. In the figure, 28 zero-filling means, 35 power calculating means, and 36 noise generating means are the same as those of the second and third embodiments. Other configurations are the same as those in FIG. 1 and FIG. 3, and the description of the operation of the parts other than those illustrated is omitted.

Hereinafter, the operation of the embodiment of the present invention will be described with reference to FIG.

The narrow-band driving sound source signal 24 is input to the zero-filling means 28 and the power calculating means 35 in the wide-band driving sound source estimating means 27. The zero-filling means 28 in the wide-band driving sound source estimating means 27 inserts zeros by M-1 samples between each sample value of the narrow-band driving sound source signal 24, and outputs a signal of M times the obtained number of samples. Here, M is a value obtained by dividing the sampling frequency of the wideband audio signal to be restored by the sampling frequency of the narrowband audio signal, and the operation of inserting zero is the same as that of the zero-filling means 15.

The power calculation means 35 calculates and outputs the power of the narrow band drive excitation signal 24. The noise generator 36 generates and outputs a power-normalized white noise signal. Then, a signal obtained by multiplying the signal output by the zero-filling means 28 by the gain gr1 is added to a white noise signal output by the noise generating means 36 by the power output by the power calculating means 35, and further a signal obtained by multiplying the gain gr2. And outputs it as a broadband drive sound source signal 29.

When the restored sounds according to the second and third embodiments have respective advantages and disadvantages, by configuring in this way and setting gr1 and gr2 appropriately, it is possible to restore a broadband sound having a quality exceeding both of them. effective. The second embodiment has the same effect as the second and third embodiments.

Embodiment 5 FIG.
Another configuration capable of favorably restoring a broadband sound source signal will be described.

FIG. 7 is a configuration diagram of the wideband sound source estimating means 14 in the wideband speech restoration apparatus according to the fifth embodiment of the present invention. In the figure, a new portion is a narrow-band long-period prediction analysis means 37, a narrow-band long-period delay 38, a narrow-band long-period prediction coefficient 39, a long-period inverse filter 40, and a narrow-band long-period prediction residual 41. Signal, wideband long-period prediction residual estimating means 42, zero-filling means 43, wideband long-period prediction parameter (code) estimating means 44, wideband long-period delay 45, wideband long-period prediction coefficient 46, length 47 This is a wideband long-period prediction residual signal of the period synthesis filter 48. Since the entire configuration is the same as that of FIG. 1, the description is omitted.

Hereinafter, the operation of the embodiment of the present invention will be described with reference to FIG.

The narrow band sound source signal 6 is input to the sound source analyzing means 21 in the wide band sound source estimating means 14. The narrow-band long-period prediction analysis unit 37 in the sound source analysis unit 21 performs a long-period prediction analysis on the narrow-band excitation signal 6, and outputs a narrow-band long-period delay 38, which is a narrow-band long-period prediction code, and a narrow-band long-period. The prediction coefficient 39 is output. Note that the long-period prediction analysis is a method often used in the CELP coding method, and thus the description is omitted.

The long-period inverse filter 40 in the sound source analyzer 21 performs inverse filtering of the narrow-band sound source signal 6 using the narrow-band long-period delay 38 and the narrow-band long-period prediction coefficient 39, and predicts the obtained signal in the narrow-band long-period prediction. The residual signal 41 is output to the wideband long-period prediction residual estimating means 42.

The zero-filling means 43 in the wide-band long-period prediction residual estimating means 42 inserts zero by M-1 samples between each sample value of the narrow-band long-period prediction residual signal 41, and obtains M times the number of samples obtained. The signal is output as a wideband long-period prediction residual signal 48. Here, M is a value obtained by dividing the sampling frequency of the wideband audio signal to be restored by the sampling frequency of the narrowband audio signal, and the operation of inserting zero is the same as that of the zero-filling means 15.

The wide-band long-period prediction parameter (code) estimating means 44 multiplies the narrow-band long-period delay 38 by M to output a wide-band long-period delay 45, which is one of the prediction codes. The multiplication factor is multiplied to output a wideband long-period prediction coefficient 46 which is another prediction code. Assuming that g is 1, the pitch periodicity of the finally obtained broadband excitation signal 16 is equivalent to that of the narrowband excitation signal 6, and the pitch periodicity becomes weaker as compared with the narrowband excitation signal 6 as it is reduced from 1. Go. As in the second embodiment, when restoring a high frequency range, setting g to a value smaller than 1 results in higher quality.

Finally, the long-period synthesis filter 47 performs long-period synthesis filtering on the wide-band long-period prediction residual signal 48 using the wide-band long-period delay 45 and the wide-band long-period prediction coefficient 46, and It is output as a sound source signal 16.

With such a configuration, the characteristics related to the strength and fluctuation of the pitch periodicity of the narrow-band excitation signal are well represented by the narrow-band long-period delay 38 and the narrow-band long-period prediction coefficient 39, and are reflected in the wide-band excitation signal. Therefore, there is an effect that it is possible to sufficiently estimate a sound source that changes in various ways and to restore a broadband sound having good sound quality without pulse-like sound. Further, there is an effect that an appropriate sound source can be estimated without depending on a speaker.

In the broadband sound source signal 16, the fundamental frequency determined by the wideband long-period delay 45 and the frequency of its harmonic component are correctly aligned at integer multiple positions. The connection of the wideband components is good, and there is an effect that high-quality wideband speech can be restored.

Furthermore, since a feature that the pitch periodicity becomes weaker as the frequency becomes higher can be introduced by the coefficient g, a more natural sound quality can be obtained.

In addition, since voiced unvoiced judgment and pitch extraction are not required and a sound source of an intermediate property can be expressed, there is no influence of voiced unvoiced judgment error or pitch extraction error which is likely to occur on a narrowband audio signal on which noise is superimposed, A good wideband sound source can be estimated even near a voiced / unvoiced boundary, and there is an effect that a wideband sound with stable and natural sound quality can be restored.

Embodiment 6 FIG.
FIG. 8 is a configuration diagram of the wideband long-period prediction residual estimating means 42 in the wideband speech restoration apparatus according to the sixth embodiment of the present invention. In the figure, the power calculation means 35 and the noise generation means 36 are the same as those in the third embodiment. Other configurations are the same as those in FIG. 1 and FIG.

Hereinafter, the operation of the embodiment of the present invention will be described with reference to FIG.

The narrow-band long-period prediction residual signal 41 is input to the power calculator 35 in the wide-band long-period prediction residual estimator 42. The power calculator 35 calculates and outputs the power of the narrow band long cycle prediction residual signal 41. The noise generator 36 generates and outputs a power-normalized white noise signal. Then, the white noise signal is multiplied by the power output from the power calculation unit 35 in the wideband long-period prediction residual estimation unit 42, and the obtained signal is output as a wideband long-period prediction residual signal 48.

As described in the third embodiment, fine fluctuations in the pitch period and the strength of the periodicity in the narrow-band excitation signal 6 cannot be expressed by the narrow-band long-period delay 38 and the narrow-band long-period prediction coefficient 39. Are included in the narrow-band long-period prediction residual signal 41. When the wideband long-period prediction residual signal 48 is generated using the narrowband long-period prediction residual signal 41 including the error component as in the fifth embodiment, unnecessary disturbance occurs in the wideband long-period prediction residual signal 48. In some cases, better restored sound may be obtained by generating white noise having the same power and using it as the wideband long-period prediction residual signal 48.

By configuring as in the sixth embodiment, the white noise having the same power as the narrow-band long-period prediction residual signal 41 is generated and used as the wide-band long-period prediction residual signal 48. In addition to this, there is an effect that a good restored sound with little disturbance due to the fluctuation of the pitch period or the intensity of the periodicity can be obtained.

Further, when the zero padding processing is performed, a spectrum symmetrical about 4 KHz is generated. Therefore, the processing is performed on the narrow-band long-period prediction residual signal 41 having no components from 0 Hz to 300 Hz and 3.4 KHz to 4.0 KHz. Then, a signal having no components of 0 Hz to 300 Hz, 3.4 KHz to 4.6 KHz, and 7.7 KHz to 8 KHz is obtained. On the other hand, in this configuration using white noise, the wideband long-period prediction residual signal 48 having all components from 0 Hz to 8 KHz is obtained, so that there is an effect that a good restored sound without a lacking band is obtained. In particular, the effect is great when restoring from 0 Hz to 300 Hz.

Embodiment 7 FIG.
FIG. 9 is a block diagram of the wideband long-period prediction residual estimating means 42 in the wideband speech restoration apparatus according to the seventh embodiment of the present invention. In the figure, 43 zero filling means, 35 power calculation means, and 36 noise generation means are the same as those of the fifth and sixth embodiments. Other configurations are the same as those in FIG. 1 and FIG.

Hereinafter, the operation of the embodiment of the present invention will be described with reference to FIG.

The narrow-band long-period prediction residual signal 41 is input to the zero-filling unit 43 and the power calculating unit 35 in the wide-band long-period prediction residual estimating unit 42. The zero-filling means 43 in the wide-band long-period prediction residual estimating means 42 inserts zeros by M-1 samples between each sample value of the narrow-band long-period prediction residual signal 41 and obtains M times the number of samples obtained. The signal of is output. Here, M is a value obtained by dividing the sampling frequency of the wideband audio signal to be restored by the sampling frequency of the narrowband audio signal, and the operation of inserting zero is the same as that of the zero-filling means 15.

The power calculation means 35 calculates and outputs the power of the narrow band long cycle prediction residual signal 41. The noise generator 36 generates and outputs a power-normalized white noise signal. Then, a signal obtained by multiplying the signal output from the zero-filling means 43 by the gain gr1 is multiplied by the white noise signal output from the noise generating means 36 by the power output from the power calculating means 35, and further a signal obtained by multiplying the gain gr2. And outputs it as a wideband long-period prediction residual signal 48.

In the case where the restored sounds according to the fifth and sixth embodiments have respective advantages and disadvantages, by configuring in this way and setting gr1 and gr2 appropriately, it is possible to restore a broadband sound having a quality exceeding both of them. effective. Note that the same effects as those of the fifth and sixth embodiments are obtained.

Embodiment 8 FIG.
FIG. 10 is a block diagram of the wideband sound source estimating means 14 in the wideband speech restoration apparatus according to the eighth embodiment of the present invention. The new parts in the figure are 49 upsampling means and 50 nulling means. Since the entire configuration is the same as that of FIG. 1, the description is omitted.

Hereinafter, the operation of the embodiment of the present invention will be described with reference to FIG.

The narrow-band sound source signal 6 is input to the up-sampling means 49. The up-sampling unit 49 up-samples the narrow-band sound source signal 6 by M times and outputs the obtained signal to the sound source analyzing unit 21.

The narrow-band long-period prediction analysis unit 37 in the sound source analysis unit 21 performs a long-period prediction analysis on the output signal of the up-sampling unit 49, and outputs a narrow-band long-period delay 38 and a narrow-band long-period prediction coefficient 39. . Note that the delay search range in the long-period prediction analysis is M times that in the fifth embodiment.

The long-period inverse filter 40 in the sound source analyzing unit 21 performs inverse filtering on the output signal of the up-sampling unit 49 using the narrow-band long-period delay 38 and the narrow-band long-period prediction coefficient 39, and converts the obtained signal into a narrow-band signal. The long-period prediction residual signal 41 is output to the wide-band long-period prediction residual estimating means 42.

The zeroing means 50 in the wideband long-period prediction residual estimating means 42 leaves only the signal every M samples of the narrowband long-period prediction residual signal 41, and sets the value of the remaining signal to zero. Then, the obtained signal is output as a wideband long-period prediction residual signal 48.

The wideband long-period prediction parameter estimating means 44 outputs the narrow-band long-period delay 38 as it is as the wide-band long-period delay 45, multiplies the narrow-band long-period prediction coefficient 39 by g, and outputs it as the wide-band long-period prediction coefficient 46. g is the same as in Example 5.

Finally, the long-period synthesis filter 47 performs long-period synthesis filtering on the wide-band long-period prediction residual signal 48 using the wide-band long-period delay 45 and the wide-band long-period prediction coefficient 46, and It is output as a sound source signal 16.

With this configuration, long-period analysis can be performed on a signal with a high sampling frequency, so that a more accurate delay can be analyzed, and characteristics relating to the strength and fluctuation of the pitch periodicity of the narrow-band sound source signal. Can be reflected more finely in a wideband sound source signal, and there can be obtained an effect that a sound source that changes variously can be sufficiently estimated, and a wideband sound with good sound quality can be restored. Note that the same effect as in the fifth embodiment is also obtained.

Embodiment 9 FIG.
FIG. 11 is a configuration diagram of a wideband audio restoration apparatus according to Embodiment 9 of the present invention. The new parts in the figure are a narrow-band power calculating means 51, a narrow-band excitation power 52, and a narrow-band power-inclusive spectrum codebook 53. Others are the same as those described above, and therefore only those having a slight difference in operation will be described.

Hereinafter, the operation of the embodiment of the present invention will be described with reference to FIG.

The narrow-band power calculating means 51 in the analyzing means 2 calculates the power included in the amplitude information of the narrow-band sound source signal 6 and outputs the calculated power as the narrow-band sound source power 52. In addition, it also outputs a spectrum parameter 4 and a narrow-band sound source signal 6.

The vector quantization means 8 in the wideband spectrum estimating means 7 collectively vector-quantizes the narrowband spectral parameters 4 and the narrowband excitation power 52 using the narrowband power-added spectral codebook 53, and obtains the obtained spectral code. 10 is output to the inverse quantization means 11 in the wideband spectrum estimation means 7.

Here, the narrow-band power-added spectrum codebook 53 is created in the same manner as in Reference 1, using a pair of a narrow-band spectral parameter and a narrow-band sound source power obtained by analyzing many narrow-band speech signals as learning data. I do. As a distance measure at the time of learning the narrow-band power-added spectrum codebook 53 and at the vector quantization means 8, a value obtained by adding the Euclidean distance of the logarithmic value of power to w times and the Euclidean distance of the spectral parameter can be used. .

Note that the narrow-band power calculating means 51 may calculate the power of the narrow-band sound signal 1 instead of the narrow-band sound source signal 6 and use the calculated power instead of the narrow-band sound source power 52. In this case, learning of the narrow-band power-added spectrum codebook 53 is performed using the pair of the narrow-band spectral parameter and the power of the narrow-band audio signal as learning data.

With such a configuration, in addition to the effects of the first embodiment, information relating to power is reflected in estimation of broadband spectrum parameters, and there is an effect that a good spectrum can be more stably estimated.

Embodiment 10 FIG.
FIG. 12 is a configuration diagram of a wideband audio restoration apparatus according to Embodiment 10 of the present invention. In the figure, the new parts are included in the sound source normalization means 54, the narrow band normalized excitation signal 55, the wideband normalized excitation signal 56, the wideband power codebook 57, the wideband excitation power 58, and the wideband spectrum estimation means. 59 broadband sound source power estimating means. Others are the same as those described above, and the description is omitted.

Hereinafter, the operation of the embodiment of the present invention will be described with reference to FIG.

The sound source normalizing means 54 in the analyzing means 2 calculates the power included in the amplitude information of the narrow band sound source signal 6 and outputs it as the narrow band sound source power 52 to the wide band sound source power estimating means 59. Is output to the broadband sound source estimating means 14 as a narrow band normalized sound source signal 55.

Actually, the vector quantization means 8 in the wideband excitation power estimation means 59 in the wideband spectrum estimation means 7 uses the narrowband power-inclusive spectrum codebook 53 to convert the narrowband spectral parameters 4 and the narrowband excitation power 52. The spectral code 10 obtained by the vector quantization is output to the inverse quantization means 11 in the wideband excitation power estimation means 59. The inverse quantization means 11 decodes the spectrum code 10 using the wideband power codebook 57 and outputs the obtained wideband excitation power 58.

The wideband sound source estimating means 14 estimates the wideband normalized sound source signal 56 using the narrowband normalized sound source signal 54. The estimation in the broadband spectrum estimating means 7 and the wideband sound source estimating means 14 can be performed in the same manner as in the first to eighth embodiments. Then, the broadband normalized excitation signal 56 is multiplied by the broadband excitation power 58 to generate the broadband excitation signal 16.

With such a configuration, in addition to the effect of the first embodiment, the estimation of the broadband sound source power can reflect the difference in the spectrum parameter, and thus, there is an effect that a wideband voice having a more correct amplitude can be restored. .

Embodiment 11 FIG.
FIG. 13 is a configuration diagram of a wideband audio restoration apparatus according to Embodiment 11 of the present invention. The new part in the figure is the 60 wideband power-inclusive spectral codebook. The other parts are the same as those in FIGS. 11 and 12, and only those having a slight difference in the operation will be described.

Hereinafter, the operation of the embodiment of the present invention will be described with reference to FIG.

The inverse quantization means 11 in the wideband spectrum estimating means 7 decodes the spectrum code 10 using the wideband power-added spectrum codebook 60, and outputs the obtained wideband spectrum parameters 13 and wideband excitation power 58.

Here, the spectrum codebook 60 with wideband power is created by a method similar to that of Reference 1 using, as learning data, a pair of a wideband spectrum parameter and a wideband excitation power obtained by analyzing many wideband speech signals. As the distance scale, the same scale as that used to create the spectrum codebook 53 including the narrow band power is used.

With such a configuration, the effects of the ninth and tenth embodiments can be obtained.

Embodiment 12 FIG.
FIG. 14 is a configuration diagram of a wideband audio restoration apparatus according to Embodiment 12 of the present invention. The new part in the figure is 61 post-filter means. Other configurations are the same as those in the first to eleventh embodiments, and a description thereof will be omitted.

Hereinafter, the operation of the embodiment of the present invention will be described with reference to FIG.

The post-filter unit 61 performs post-filtering processing on the temporary wideband audio signal output from the synthesis filter 17, and outputs the obtained signal to the band filter 18. The band-pass filter 18 performs band-pass filter processing on the signal output from the post-filter unit 61, and extracts components other than the band including the narrow-band sound component.

The post-filtering process is a signal processing process that improves perceived quality. It emphasizes pitch periodicity and spectral poles, enhances high frequencies to improve clarity, and is used when passing through a transmission path. In this case, a band in which a large amount of distortion occurs is suppressed to reduce the sense of distortion.

As a process of enhancing the pitch periodicity, a method is generally used in which a temporary wideband audio signal before the pitch period is multiplied by a coefficient smaller than 1 and added to the current temporary wideband audio signal.

As the pole enhancement processing, the wideband spectrum parameter 13 is modified to have a large gain in a frequency band near the pole frequency of the wideband spectrum parameter 13 and a small gain in a frequency band other than the pole band of the wideband spectrum parameter 13. Various methods for calculating the filter coefficient of the pole-zero filter have been proposed, and can be realized by applying this filter to a temporary wideband audio signal. Further, since distortion generated when passing through the transmission path is large in a frequency band having a small amplitude, that is, in a frequency band other than a very close vicinity, a band with a large amount of distortion can be suppressed by this pole enhancement processing.

As the high-frequency emphasis processing, a method called pre-emphasis, that is, a method of multiplying a temporary wideband audio signal one point before by a coefficient of 1 or less and subtracting it from the current temporary wideband audio signal is generally used.

In FIG. 14, the post filter means 61 and the band-pass filter 18 may be in opposite positions, or the post-filter means 61 may be applied to the wideband audio signal 20.

With such a configuration, in addition to the effects of the first embodiment, when the sound quality of the restored wideband audio signal is insufficient, the pitch periodicity of the wideband audio signal or the pole of the spectrum is emphasized, Is emphasized to improve the clarity, and there is an effect that a band with much distortion generated when passing through a transmission path is suppressed to reduce a feeling of distortion.

In addition, a configuration in which the inverse filter 5 and the wideband sound source estimating means 14 are removed in FIG. 14 is also possible. This configuration corresponds to the application of the present invention to Document 1, and has the same effects as described above.

Embodiment 13 FIG.
A configuration is also possible in which the wideband spectrum estimating means 7 in the first to twelfth embodiments outputs the narrowband spectrum parameter 4 as it is as the wideband spectrum parameter 13.

FIG. 15 is an explanatory diagram for explaining the general relationship between the narrowband spectrum and the wideband spectrum in this case. In the case where the spectrum envelope represented by the narrowband spectrum parameter 4 is as shown in FIG. 15A, if this is used as it is as the broadband spectrum parameter 13, the width is consequently expanded, and the spectrum envelope represented by the wideband spectrum parameter 13 is as shown in FIG. FIG. 15B is a diagram in which FIG. 15A is expanded M times in the frequency axis direction, and when M is 2, FIG. Therefore, when the narrow band spectral envelope is higher from 2 KHz to 3.4 KHz, the restored high band above 3.4 KHz also becomes higher, and conversely, when the lower 2 KHz to 3.4 KHz is lower, the higher band becomes lower. As a result, the rough slope of the narrow band spectral envelope is directly reflected in the high band.

With such a configuration, in addition to the effects of the first embodiment, there is an effect that the broadband spectrum can be restored extremely easily, though roughly. Compared with the first embodiment, there is no need for a memory for storing a codebook, and there is an effect that the amount of calculation is reduced.

Embodiment 14 FIG.
In the first to twelfth embodiments, a configuration is also possible in which the wideband spectrum estimating means 7 outputs from the lowest order to the predetermined order of the narrowband spectrum parameter 4 as the wideband spectrum parameter 13. However, the narrow band spectral parameter 4 output from the spectrum analyzing means 3 is always stable even when the one extracted from the lowest order to the predetermined order such as a PARCOR coefficient or an autocorrelation coefficient is used as the wide band spectral parameter 13. Only when the parameters are appropriate.

FIG. 16 is an explanatory diagram for explaining the general relationship between the narrowband spectrum and the wideband spectrum in this case. In the case where the spectrum envelope represented by the narrowband spectrum parameter 4 is as shown in FIG. 15A, if the lowest to the predetermined order of the spectrum envelope is used as the wideband spectrum parameter 13, the spectrum envelope represented by the wideband spectrum parameter 13 is as shown in FIG. Is expanded M times in the frequency axis direction to further smooth the pole structure, and when M is 2, it becomes as shown in FIG. As a result, the rough slope of the narrow-band spectrum envelope is directly reflected in the high band, and a strong non-existent pole is generated in the high band, so that generation of an unnatural restoration sound can be suppressed.

With this configuration, in addition to the effect of the thirteenth embodiment, it is possible to suppress the occurrence of an unnatural restoration sound, which is rarely generated in the thirteenth embodiment and has a strong non-existent pole in the high range. There is an effect that can be done.

Embodiment 15 FIG.
FIG. 17 is a configuration diagram of the wideband spectrum estimating means 7 of the wideband speech restoration apparatus according to Embodiment 15 of the present invention. In the figure, new parts are a spectrum parameter conversion means 62, an order reduction means 63, and a spectrum parameter inverse conversion means 64. Other configurations are the same as those of the first to twelfth embodiments, and a description thereof will be omitted.

Hereinafter, the operation of the embodiment of the present invention will be described with reference to FIG.

The spectrum parameter converting means 62 in the wide band spectrum estimating means 7 converts the narrow band spectral parameter 4 into a parameter whose synthesis is always stable when the narrow band spectral parameter 4 is extracted from the lowest order to a predetermined order such as a PARCOR coefficient or an autocorrelation coefficient. I do. The order reduction unit 63 outputs a parameter extracted from the lowest order to a predetermined order of the parameters output by the spectrum parameter conversion unit 62 to the spectrum parameter inverse conversion unit 64. The spectrum parameter inverse transform unit 64 returns the parameter output from the order reduction unit 63 to the same region as the narrowband spectral parameter 4 and outputs the same as the wideband spectral parameter 13.

構成 With such a configuration, the same effect as in the fourteenth embodiment can be obtained even when the narrowband spectral parameter 4 is a parameter in which the synthesis becomes unstable when extracting from the lowest order to the predetermined order.

Embodiment 16 FIG.
In the fourteenth and fifteenth embodiments, the strong poles are suppressed by the order reduction. However, a method that gives a similar effect, such as applying a lag window to the autocorrelation coefficient as a spectral parameter, can be used.

構成 With this configuration, the same effect as that of the fourteenth embodiment can be obtained by another means.

It is also possible to apply the wideband spectrum estimating means 7 of the thirteenth to sixteenth embodiments to a conventional configuration such as that of Reference 1. For example, the overall configuration when applied to Document 1 is such that the inverse filter 5, the broadband sound source estimating means 14, and the post-filtering means 61 are removed from FIG. In such a configuration, the effects newly generated in the thirteenth to sixteenth embodiments can be added to the conventional technology.

Embodiment 17 FIG.
In the following embodiment, an example will be described in which the present invention is applied to an apparatus for restoring wideband speech based on coded information by transmission or the like.

FIG. 18 is a configuration diagram of a wideband audio restoration apparatus according to Embodiment 17 of the present invention. In the figure, 101 is a narrowband speech code, 102 is a separating means, 103 is a narrowband spectral code, 104 is a narrowband excitation code, 105 is a wideband spectral decoding means, 106 is a wideband excitation decoding means, and 107 is a narrowband spectral decoding means , 108 are narrowband sound source decoding means, and 109 is a narrowband speech decoding means. Other configurations are the same as those in the first to sixteenth embodiments, and a description thereof will not be repeated.

に お い て This embodiment is also configured to obtain a good wideband sound source signal without performing reanalysis.

Hereinafter, the operation of the embodiment of the present invention will be described with reference to FIG.

{First, the narrowband speech code 101 is input to the separating unit 102 and the narrowband speech decoding unit 109. The narrow-band speech code 101 is, for example, sampled at 8 KHz, is a separately encoded narrow-band speech signal limited to a telephone band of 300 Hz to 3.4 KHz, and is input from a storage medium or a communication path. Things.

The separation means 102 separates the narrowband speech code 101 into a narrowband spectrum code 103 and a narrowband excitation code 104, and converts the narrowband spectrum code 103 into a wideband spectrum decoding means 105 and the narrowband excitation code 104 into a wideband excitation codec. Output to 106.

狭 The narrow band spectrum decoding unit 107 in the wide band spectrum decoding unit 105 decodes the narrow band spectrum code 103 and outputs the obtained narrow band spectrum parameter 4. It should be noted that the narrowband spectrum decoding means 107 only needs to perform a process opposite to the process of encoding the narrowband spectrum parameters used when the narrowband speech code 101 was encoded.

{Circle around (2)} The wideband spectrum estimating means 7 in the wideband spectrum decoding means 105 estimates the wideband spectrum parameters 13 using the narrowband spectrum parameters 4. As the broadband spectrum estimating means 7, the method described in the embodiment described above can be used.

狭 The narrow band excitation decoding unit 108 in the wide band excitation decoding unit 106 decodes the narrow band excitation code 104 and outputs the obtained narrow band excitation signal 6. Then, the wideband excitation estimating means 14 in the wideband excitation decoding means 106 estimates the wideband excitation signal 16 using the narrowband excitation signal 6.

Note that the wideband sound source estimating means 14 may use a zero-filling means or the like. The narrowband excitation decoding means 108 may perform the reverse of the encoding of the narrowband excitation signal used when the narrowband speech code 101 was encoded.

The synthesis filter 17 performs synthesis filter processing on the broadband sound source signal 16 using the wideband spectral parameters 13 to generate a temporary wideband audio signal. The band-pass filter 18 performs a band-pass filter process on the provisional wide-band audio signal, and extracts components other than the band having the narrow-band audio component. When the band of the wideband audio signal is from 0 Hz to 7.3 kHz, components of 0 Hz to 300 Hz and components of 3.4 kHz to 7.3 kHz are extracted.

On the other hand, the narrowband speech decoding means 109 decodes the input narrowband speech code 101 and outputs the obtained narrowband speech signal 1 to the upsampling means 19. This decoding process may be the reverse of the encoding process used when the narrowband speech code 101 was encoded.

Next, the up-sampling means 19 up-samples the narrowband audio signal 1 by M times. The signal generated by the upsampling has the same sampling frequency as the wideband audio signal 20 and has the same narrowband component as the narrowband audio signal 1. Then, the output of the bandpass filter 18 and the output of the upsampling means 19 are added to generate a wideband audio signal 20.

With this configuration, when a narrowband speech code is received from a storage medium or a communication path, there is no need to re-analyze the narrowband speech, so that there is an effect that restoration can be performed with a small processing amount. In addition, since distortion due to interpolation at the time of synthesis or windowing at the time of analysis is not superimposed, there is an effect that better quality wideband speech can be restored. Note that the same effect as in the first embodiment is also obtained.

Note that the narrow-band speech decoding means 109 may be configured to input the narrow-band spectrum parameter 4 and the narrow-band sound source signal 6 and synthesize the narrow-band speech signal 1, or conversely, to decode the narrow-band speech decoding means 109. A configuration is also possible in which the narrowband spectral parameter 4 and the narrowband excitation signal 6 calculated as intermediate parameters in the process are input to the wideband spectrum decoding means 105 and the wideband excitation decoding means 106. In this case, there is an effect that the overlapping processing can be omitted and the wideband sound can be restored with a smaller processing amount.

When the pitch cycle code and the power code can be separated from the narrowband speech code 101, the pitch cycle and the power information are decoded from these codes, and the broadband spectrum parameter 13 and the pitch cycle and the power information are used. Thus, a configuration in which a temporary broadband synthesized sound is generated in the same manner as in Reference 1 is also possible.

Embodiment 18 FIG.
FIG. 19 is a configuration diagram of the wideband sound source decoding means 106 of the wideband speech restoration apparatus according to the eighteenth embodiment of the present invention. In the figure, the novel parts are a narrow band pitch code 110, a narrow band power code 111, a wide band pitch decoding unit 112, a wide band pitch period 113, a wide band power decoding unit 114, a wide band power 115, and a sound source generation 116. Means. Other points are the same as those of the seventeenth embodiment, and the description is omitted.

This embodiment is limited to the case where the narrow band speech code 101 is input so that the narrow band pitch code 110 and the narrow band power code 111 can be easily separated by the separating means 102. In this case, the configuration of FIG. 19 is significant.

Hereinafter, the operation of the embodiment of the present invention will be described with reference to FIG.

As the narrow band excitation code 104, the narrow band pitch code 110 and the narrow band power code 111 are input to the wide band excitation decoding means 106.

広 帯 域 The wideband pitch decoding unit 112 in the wideband excitation decoding unit 106 estimates the wideband pitch period 113 using the narrowband pitch code 110. As an estimation method, a narrow band pitch period may be decoded from the narrow band pitch code 110 and its value may be multiplied by M, but the result is stored as a table and a table component corresponding to the narrow band pitch code 110 is stored. You may ask by reading out.

Next, the wideband power decoding unit 114 in the wideband excitation decoding unit 106 estimates the wideband power 115 using the narrowband power code 111. As an estimation method, the narrow-band power code 111 may be decoded from the narrow-band power code 111 and its value may be multiplied by g, but the result is stored as a table and the table component corresponding to the narrow-band power code 111 is stored. You may ask for it by reading it out.

The sound source generating means 116 outputs a signal in which fixed sound sources are arranged side by side with the broadband pitch period 113 as a repetition period. Finally, the output signal of the sound source generating means 116 is multiplied by a wideband power 115 and output as a wideband sound source signal 16. I do.

With such a configuration, in addition to the effects of the seventeenth embodiment, the wideband excitation signal 16 is directly generated without decoding the narrowband excitation signal, so that there is an effect that restoration can be performed with a small amount of processing.

Embodiment 19 FIG.
FIG. 20 is a configuration diagram of the wideband sound source decoding means 106 of the wideband speech restoration apparatus according to the nineteenth embodiment of the present invention. In the figure, a new portion is a narrow band adaptive excitation code 117, a narrow band excitation code 118, a wide band adaptive excitation decoding device 119, a wide band driving excitation decoding device 120, a narrow band adaptive excitation decoding device 121 and a narrow band excitation code 122. This is a narrow-band drive sound source decoding means. Others are the same as those described above, and the description is omitted.

This embodiment is limited to a case where a narrow-band speech code 101 is input so that the narrow-band adaptive excitation code 117 and the narrow-band excitation code 118 can be easily separated from the input narrow-band speech code by the separating means 102. Can be In this case, the configuration of FIG. 20 is significant.

Hereinafter, the operation of the embodiment of the present invention will be described with reference to FIG.

As the narrow-band excitation code 104, the narrow-band adaptive excitation code 117 and the narrow-band excitation code 118 are input to the wide-band excitation decoding means 106.

The narrow band adaptive excitation decoding unit 121 in the wide band adaptive excitation decoding unit 119 decodes the narrow band adaptive excitation code 117, and outputs the obtained narrow band adaptive lag length 25 and narrow band adaptive gain 26. The wideband adaptive excitation estimating means 30 in the wideband adaptive excitation decoding means 119 generates and outputs a wideband adaptive excitation signal 32 from the narrowband adaptive lag length 25 and the narrowband adaptive gain 26. The operation of the wideband adaptive sound source estimating means 30 is the same as in the second embodiment.

狭 The narrow band drive excitation decoding unit 122 in the wide band drive excitation decoding unit 120 decodes the narrow band drive excitation code 118 and outputs the obtained narrow band drive excitation signal 24. The broadband drive excitation estimating means 27 in the wideband drive excitation decoding means 120 estimates the wideband drive excitation signal 29 from the narrowband drive excitation signal 24 and outputs it. The operation of the broadband driving sound source estimating means 27 is the same as in the second to fourth embodiments.

Finally, the broadband adaptive excitation signal 32 and the broadband drive excitation signal 29 are added and output as the broadband excitation signal 16.

With this configuration, in addition to the effects of the second to fourth embodiments and the seventeenth embodiment, since the wideband excitation signal 16 is directly generated without decoding the narrowband excitation signal, a small amount of processing is performed. Has the effect of being able to restore.

Furthermore, since the fundamental frequency and the frequency of its harmonic component are correctly aligned at integer multiple positions, the connection between the narrowband component and the restored wideband component in the finally restored broadband audio signal is good, and high-quality wideband speech is reproduced. There is an effect that can be restored.

In addition, since voiced unvoiced information and pitch period information are not used, sound sources of intermediate characteristics can be expressed, so that the effects of voiced unvoiced decision errors and pitch extraction errors that tend to occur on narrowband speech signals with superimposed noise are reduced. In addition, a good wideband sound source can be estimated even near a voiced / unvoiced boundary, and there is an effect that a wideband sound with stable and natural sound quality can be restored.

Embodiment 20 FIG.
FIG. 21 is a configuration diagram of the wideband sound source decoding means 106 of the wideband sound restoration apparatus according to the twentieth embodiment of the present invention. In the figure, new parts are a narrow band long cycle prediction code 123, a wide band long cycle prediction parameter (code) decoding unit 124, a narrow band long cycle prediction parameter (code) decoding unit 125, and a narrow band long cycle prediction 126. Residual codes, 127 are wideband long-period prediction residual decoding means, and 128 are narrowband long-period prediction residual decoding means. Others are the same as those described above, and the description is omitted.

In this embodiment, a narrow-band speech code 101 is input so that the separation means 102 can easily separate a narrow-band long-period prediction code 123 and a narrow-band long-period prediction residual code 126 from an input narrow-band speech code. Limited to In this case, the configuration of FIG. 21 is significant.

Hereinafter, the operation of the embodiment of the present invention will be described with reference to FIG.

As the narrow-band excitation code 104, the narrow-band long-period prediction code 123 and the narrow-band long-period prediction residual code 126 are input to the wideband excitation decoding means 106.

A narrow-band long-period prediction parameter decoding unit 125 in the wide-band long-period prediction parameter (code) decoding unit 124 decodes the narrow-band long-period prediction code 123 and obtains a narrow-band length, which is one of the obtained prediction codes. A period delay 38 and a narrow band long period prediction coefficient 39 which is another prediction code are output. The wide-band long-period prediction parameter estimating unit 44 in the wide-band long-period prediction parameter decoding unit 124 calculates a wide-band long-period delay, which is one of the long-period prediction codes, from the narrow-band long-period delay 38 and the narrow-band long-period prediction coefficient 39. 45 and a wide-band long-period prediction coefficient 46, which is one of the other long-period prediction codes, are estimated and output. The operation of the wideband long-period prediction parameter estimating means 44 is the same as in the fifth embodiment.

The narrow-band long-period prediction residual decoding unit 128 in the wide-band long-period prediction residual decoding unit 127 decodes the narrow-band long-period prediction residual code 126 and obtains the obtained narrow-band long-period prediction residual signal 41. Is output. The wide-band long-period prediction residual estimating unit 42 in the wide-band long-period prediction residual decoding unit 127 estimates the wide-band long-period prediction residual signal 48 from the narrow-band long-period prediction residual signal 41 and outputs it. The operation of the wideband long-period prediction residual estimation unit 42 is the same as in the fifth to seventh embodiments.

Finally, the long-period synthesis filter 47 performs long-period synthesis filtering on the wide-band long-period prediction residual signal 48 using the wide-band long-period delay 45 and the wide-band long-period prediction coefficient 46, and It is output as a sound source signal 16.

With this configuration, in addition to the effects of the fifth to seventh and seventeenth embodiments, since the wideband excitation signal 16 is directly generated without decoding the narrowband excitation signal, a small amount of processing is required. Has the effect of being able to restore.

Embodiment 21 FIG.
In the seventeenth to twentieth embodiments, the wideband spectral parameter 13 is estimated after decoding the narrowband spectral parameter 4 from the narrowband spectral code 103. However, the narrowband spectral code 103 refers to the wideband spectral codebook. It is also possible to calculate the broadband spectrum parameter 13 directly by using.

With this configuration, in addition to the effects of the seventeenth to twentieth embodiments, there is an effect that restoration can be performed with a smaller processing amount.

Embodiment 22 FIG.
FIG. 22 is a configuration diagram of a wideband audio restoration apparatus according to an embodiment of the present invention. The new parts in the figure are the narrowband power decoding means 129 and the wideband normalized excitation decoding means 130.

The wideband spectrum estimating means 7 is the same as that of the eleventh embodiment, and the other parts are the same as those described above.

The operation of the embodiment of the present invention will be described below with reference to FIG.

The narrow-band power decoding unit 129 decodes a portion related to power from the narrow-band amplitude information included in the narrow-band excitation code 104 and outputs the obtained narrow-band excitation power 52 to the wide-band spectrum estimation unit 7. . The wideband spectrum estimating means 7 estimates the wideband spectrum parameter 13 and the wideband excitation power 58 using the narrowband spectrum parameter 4 and the narrowband excitation power 52.

The wideband normalized excitation decoding means 130 estimates a wideband excitation signal whose power has been normalized by using a portion other than the portion related to the narrowband power included in the narrowband excitation code 104. Output. For the processing in the wideband normalized excitation decoding means 130, the same processing as in the eighteenth to twentieth embodiments can be used. Then, the broadband normalized excitation signal 56 is multiplied by the broadband excitation power 58 to generate the broadband excitation signal 16.

With such a configuration, it is possible to combine the effects of the eleventh embodiment and the eighteenth and twentieth embodiments. Note that a configuration in which the wideband spectrum estimating means 7 estimates only one of the wideband spectrum parameter 13 and the wideband sound source power 58 as in the ninth and tenth embodiments is also possible.

Embodiment 23 FIG.
In the seventeenth to twenty-second embodiments, a configuration in which post filter means 61 is inserted between the synthesis filter 17 and the bandpass filter 18 is also possible. Further, a configuration in which the post-filter means 61 and the band-pass filter 18 are in opposite positions is possible, and a configuration in which the post-filter means 61 is applied to the wideband audio signal 20 is also possible.

With this configuration, when post-filter processing is performed in the narrow-band speech decoding unit 109, continuity between the narrow-band portion and the restored band can be improved. In addition, the effects of the twelfth embodiment and the seventeenth to twenty-second embodiments can be combined.

Embodiment 24 FIG.
In a configuration in which the wideband excitation decoding means 106 is removed from FIG. 18, a configuration in which a post filter means 61 is inserted between the synthesis filter 17 and the bandpass filter 18 is also possible. Further, a configuration in which the post-filter means 61 and the band-pass filter 18 are in opposite positions is possible, and a configuration in which the post-filter means 61 is applied to the wideband audio signal 20 is also possible.

This configuration corresponds to the configuration in which claims 9 and 15 of the present invention are applied to Document 1, and when post-filter processing is performed in narrow-band speech decoding means 109, the narrow-band portion and the restored band There is an effect that continuity can be improved.

(4) The features of each embodiment are described below.

The above-described wideband audio restoration apparatus includes an analyzing unit that analyzes a narrowband audio signal to obtain a narrowband spectral parameter and a narrowband sound source signal; a spectrum estimating unit that estimates a wideband spectral parameter using the narrowband spectral parameter; Broadband sound source estimating means for estimating a wideband sound source signal using a narrowband sound source signal, and synthesizing means for generating a wideband speech signal from the estimated wideband spectral parameters and the wideband sound source signal are provided.

{Circle around (4)} As the wideband sound source estimating means, a zero filling means for inserting a predetermined zero value into each sample interval of the input narrowband sound source signal is used.

Further, the wideband excitation estimating means analyzes the input narrowband excitation signal to obtain a narrowband adaptive excitation code and a narrowband driving excitation signal, and a wideband adaptive excitation signal using the narrowband adaptive excitation code. Adaptive sound source estimating means for estimating, driving sound source estimating means for estimating a wide band driving sound source signal using a narrow band driving sound source signal, and generating a wide band sound source signal from the estimated wide band adaptive sound source signal and the wide band driving sound source signal. And an adding means.

Alternatively, the wideband sound source estimating means analyzes the input narrowband sound source signal to obtain a narrowband long-period prediction code and a narrowband long-period prediction residual signal. A long-period prediction residual estimator for estimating a wide-band long-period prediction residual signal using a wide-band long-period prediction code using a narrow-band long-period prediction code; And a long-period synthesizing means for synthesizing a wideband excitation signal from the wideband long-period prediction residual signal and the wideband long-period prediction code.

Another wideband audio restoration apparatus is an analyzing means for analyzing a narrowband audio signal to obtain a narrowband spectral parameter and narrowband amplitude information, and using the narrowband spectral parameter and the narrowband amplitude information at least a wideband spectral parameter or A spectrum estimating means for estimating broadband amplitude information, and a synthesizing means for generating a wideband speech signal from the estimated wideband spectral parameters and the wideband amplitude information or the wideband excitation signal are provided.

Or a broadband estimating unit for estimating a wideband audio signal using a narrowband audio signal, and a post-filter unit for performing post-filtering on the estimated wideband audio signal.

An analyzing means for analyzing a narrowband speech signal to obtain a narrowband spectral parameter; a spectrum estimating means for outputting a wideband spectral parameter using the narrowband spectral parameter as it is as a wideband spectral parameter; And a synthesizing means for generating a wideband audio signal from the audio signal.

Or analyzing means for analyzing a narrowband speech signal to obtain a narrowband spectral parameter, and converting the narrowband spectral parameter to another region as necessary, performing deformation, and inversely converting to a spectral parameter region to obtain a wideband spectrum. A spectrum estimating means for outputting parameters and a synthesizing means for generating a wideband speech signal from the outputted wideband spectral parameters are provided.

広 帯 域 The other wideband speech restoration apparatus includes spectrum decoding means for estimating a wideband spectrum parameter from a narrowband speech code, and synthesis means for generating a wideband speech signal from the estimated wideband spectrum parameter.

Alternatively, spectrum decoding means for estimating a wideband spectral parameter using a narrowband spectral code separated from a narrowband speech code, and a wideband for estimating a wideband excitation signal using a narrowband excitation code separated from the narrowband speech code Excitation decoding means and synthesis means for generating a wideband speech signal from the estimated wideband spectrum parameters and the wideband excitation signal are provided.

{Circle around (4)} Further, as the wideband excitation decoding means, zero filling means for inserting a predetermined zero value into each sample interval of the narrowband excitation signal restored from the narrowband excitation code is used.

Alternatively, the wideband excitation decoding means comprises a wideband adaptive excitation decoding means for estimating a wideband adaptive excitation signal using a narrowband adaptive excitation code separated from the input narrowband speech code, and a narrowband excitation separation means separated from the input narrowband speech code. Broadband driving excitation signal decoding means for estimating a broadband driving excitation signal using a driving excitation code, and adding means for generating a wideband excitation signal from the estimated wideband adaptive excitation signal and the wideband driving excitation signal.

Alternatively, the wideband sound source decoding unit includes a wideband long period prediction code decoding unit that estimates a wideband long period prediction code using a narrowband long period prediction code separated from the input narrowband speech code, and Wideband long-period prediction residual decoding means for estimating a wide-band long-period prediction residual signal using the separated narrow-band long-period prediction residual code, and these estimated wideband long-period prediction residual signal and wideband long-period prediction residual signal And an adding means for generating a broadband sound source signal from the above.

Alternatively, a narrow-band amplitude information decoding means for estimating narrow-band amplitude information using a narrow-band excitation code separated from a narrow-band speech code, and a narrow-band spectrum code and narrow-band amplitude information separated from the narrow-band speech code. A spectrum decoding means for estimating at least a wideband spectral parameter or a wideband amplitude information by use thereof, and a synthesizing means for generating a wideband speech signal from the estimated wideband spectral parameter and the wideband amplitude information or the wideband excitation signal as required. .

Or a wideband speech decoding means for estimating a wideband speech signal using a narrowband speech code, and post-filter means for performing post-filtering on the decoded and estimated wideband speech signal.

{Circle around (4)} The wideband speech restoration apparatus described above synthesizes a wideband speech signal from the wideband spectrum parameters estimated using the narrowband spectrum parameters and the wideband excitation signal estimated using the narrowband excitation signal.

{Circle around (2)} A wideband sound source signal is generated by inserting a predetermined number of zeros between each sample of the narrowband sound source signal, and a wideband speech signal is synthesized using this and the estimated wideband spectrum.

In estimating the wideband excitation signal, the input narrowband excitation signal is analyzed to calculate a narrowband adaptive excitation code and a narrowband driving excitation signal, and the wideband adaptive excitation signal estimated using the narrowband adaptive excitation code is calculated. And the broadband driving sound source signal estimated using the narrow band driving sound source were added to obtain a wideband sound source signal. A wideband speech signal is synthesized using this and the estimated wideband spectrum.

As another method of estimating the wideband excitation signal, the input narrowband excitation signal is analyzed to calculate a narrowband long-period prediction code and a narrowband long-period residual signal, and the estimation is performed using the narrowband long-period prediction code. The wideband long-period prediction code and the wideband long-period residual signal estimated using the narrow-band long-period residual signal are used as a wideband excitation signal. A wideband speech signal is synthesized using this and the estimated wideband spectrum.

Further, other wideband speech restoration apparatuses analyze a narrowband speech signal to calculate a narrowband spectrum parameter, narrowband amplitude information, and a narrowband excitation signal, and use the narrowband spectrum parameter and the narrowband amplitude information to calculate a wideband spectrum. One or both of the parameters and the broadband amplitude information are estimated. Thereafter, a wideband audio signal is synthesized with these signals and a wideband excitation signal estimated from the narrowband excitation signal.

In another wideband audio restoration device, post-filtering is performed on a wideband audio signal estimated using a narrowband audio signal, and mainly high-frequency characteristics are processed.

Another wideband speech restoration apparatus extends a characteristic of a narrowband spectrum parameter to the whole band and synthesizes a wideband speech signal using the wideband spectrum parameter.

Further, other wideband speech restoration apparatuses use wideband spectrum parameters up to a specific order and inversely convert the narrowband spectrum parameters to corresponding spectrum parameters to obtain wideband spectrum parameters, which are used to synthesize a wideband speech signal. .

Another wide-band speech restoration apparatus generates a narrow-band synthesized sound using a narrow-band speech code and estimates a wide-band speech signal, and converts the narrow-band synthesized sound into an up-sampled signal or a narrow-band synthesized sound. A signal obtained by extracting mainly high-band components other than the narrow-band synthesized sound of the audio signal is added to synthesize a wide-band audio signal.

{Circle around (2)} In another wideband speech restoration apparatus, a wideband speech signal is synthesized using a wideband spectrum parameter estimated using a narrowband spectrum code and a wideband excitation signal estimated using a narrowband excitation code.

Furthermore, a wideband excitation signal is generated by inserting a predetermined number of zero values between each sample of the narrowband excitation decoded by using the narrowband excitation code by the wideband excitation decoding means, and the estimated wideband excitation signal is generated. A wideband audio signal is synthesized using the spectrum.

{Circle around (4)} The wideband excitation signal estimated by using the narrowband adaptive excitation code and the wideband excitation signal estimated from the narrowband excitation signal are added by another wideband excitation decoding means to generate a wideband excitation signal. A wideband speech signal is synthesized using this and the estimated wideband spectrum.

Further, a wideband excitation signal is obtained from another wideband excitation decoding means from a wideband long-period prediction code estimated using the narrowband excitation code and a wideband long-period residual signal estimated from the narrowband long-period prediction residual signal. Synthesized. A wideband speech signal is synthesized using this and the estimated wideband spectrum.

Another wideband speech restoration apparatus estimates either or both of the wideband spectrum parameter and the wideband amplitude information using the narrowband spectrum code and the narrowband amplitude information. Thereafter, a wideband speech signal is synthesized with these pieces of information and the wideband sound source signal estimated from the narrowband sound source signal.

In another wideband speech restoration apparatus, post-filtering is performed on a wideband speech signal estimated using a narrowband speech code, and mainly high-frequency characteristics are processed.

As described above, since the wideband sound source signal is estimated using the narrowband sound source signal and the wideband sound signal is synthesized using the narrowband sound source signal, the characteristics of the narrowband sound source signal are favorably given to the wideband sound source signal. Therefore, there is an effect that a wideband voice with stable and natural sound quality can be restored with little dependence on a speaker.

In addition, since the wide-band sound source estimating means uses zero padding means for inserting a predetermined number of zeros between each sample of the narrow-band sound source signal, voiced / unvoiced determination and pitch extraction are not required, and voiced / unvoiced determination error and pitch extraction are not required. There is an effect that it is possible to estimate a good broadband sound source without the influence of an error, and to restore a wideband sound with stable and natural sound quality.

Also, as the wideband excitation estimation means, the narrowband adaptive excitation code and the narrowband excitation signal are used to estimate the wideband adaptive excitation signal and the broadband excitation signal, and the wideband excitation signal is generated from this. The characteristics related to the strength and fluctuation of the pitch periodicity of the band excitation signal are well reflected in the band excitation signal, and there is an effect that a broadband sound having good sound quality can be restored without pulse-like sound.

Furthermore, since the fundamental frequency and the frequency of its harmonic component are correctly aligned at integer multiples, the connection between the narrowband component and the restored broadband component in the wideband audio signal is good, and the practical properties of pitch periodicity can be restored. This has the effect of restoring high quality wideband speech.

Also, as the wideband excitation source estimating means, the wideband long-period prediction code and the wideband long-period residual signal are estimated using the narrow-band long-period prediction code and the narrow-band long-period residual signal. , The characteristics of the pitch periodicity of the narrow-band sound source signal, such as its strength and fluctuations, are well reflected in the wide-band sound source signal. There is an effect that can be.

Furthermore, since the fundamental frequency and the frequency of its harmonic component are correctly aligned at integer multiple positions, the connection between the narrowband component and the restored wideband component in the finally restored wideband audio signal is good, and the actual pitch periodicity is improved. Characteristics can also be taken in, and there is an effect that high-quality wideband speech can be restored.

Also, because one or both of the wideband spectral parameter and the wideband amplitude information is estimated using the narrowband spectral parameter and the narrowband amplitude information, the narrowband amplitude information is reflected in the estimation of the wideband spectral parameter, There is an effect that a good spectrum can be more stably estimated, and a wideband voice having a more correct amplitude can be restored.

Furthermore, since the post-filtering is performed on the wideband audio signal estimated using the narrowband audio signal, when the sound quality of the restored wideband audio signal is insufficient, the pitch periodicity is emphasized and the pole of the spectral envelope is reduced. This has the effect of improving sound quality such as emphasis.

{Circle around (4)} Since the narrowband spectrum parameter is expanded and used as a wideband spectrum parameter to synthesize a wideband speech signal, there is an effect that a rough broadband spectrum can be restored very easily. In addition, there is no need for a memory for storing the codebook, which has the effect of reducing the amount of calculation.

{Circle around (4)} Since the wideband spectrum parameter is obtained by inverting the narrowband spectrum parameter up to a predetermined order and converting it into a spectrum parameter, the broadband spectrum can be restored very easily. In addition, there is no need for a memory for storing the codebook, which has the effect of reducing the amount of calculation.

Further, according to the present invention, a narrow-band synthesized speech is generated using a narrow-band speech code and a wide-band speech signal is estimated, and the narrow-band synthesized speech is narrowed down to a signal obtained by up-sampling the narrow-band synthesized speech or a narrow-band synthesized speech. Since the components of the bands other than the band-synthesized sound are extracted and added, it is possible to restore the wide-band sound even from the encoded narrow-band sound, and the decoded narrow-band sound is not re-analyzed. There is an effect that can be done.

Alternatively, since the wideband speech signal is synthesized using the wideband spectrum parameters estimated using the narrowband spectrum code and the wideband excitation signal estimated using the narrowband excitation code, the decoded narrowband speech is reproduced. There is no need for analysis, and there is an effect that restoration can be performed with a small amount of processing. In addition, since distortion due to interpolation at the time of synthesis or windowing at the time of analysis is not superimposed, there is an effect that better quality wideband speech can be restored.

In addition, as the wideband excitation decoding means, a zero padding means for inserting a predetermined number of zeros between each sample of the narrowband excitation decoded using the narrowband excitation code is used, so that a sound source having an intermediate characteristic between voiced and unvoiced is used. Can be satisfactorily restored, and there is an effect that a broadband sound with stable and natural sound quality can be restored.

Also, as the wideband excitation decoding means, the wideband adaptive excitation signal and the wideband driving excitation signal estimated using the narrowband excitation code are estimated, and the sum is added to obtain the wideband excitation signal. Therefore, a wideband sound source signal is directly generated without performing the above processing, and there is an effect that restoration can be performed with a small amount of processing.

(4) Since the characteristics related to the strength and fluctuation of the pitch periodicity included in the narrowband excitation code are well reflected in the wideband excitation signal, there is an effect that a wideband speech having good sound quality can be restored.

Also, as the wideband excitation decoding means, the wideband long-period prediction code and the wideband long-period residual signal estimated using the narrowband excitation code are estimated, and the wideband excitation signal is synthesized using these. Thus, the wideband excitation signal is directly generated without decoding the narrowband excitation signal, and the effect is that the restoration can be performed with a small amount of processing.

(4) Since the characteristics related to the strength and fluctuation of the pitch periodicity included in the narrowband excitation code are well reflected in the wideband excitation signal, there is an effect that a wideband speech having good sound quality can be restored.

In addition, since the wideband spectral parameter and / or the wideband amplitude information are estimated using the narrowband spectral code and the narrowband amplitude information, the narrowband amplitude information is reflected in the estimation of the wideband spectral parameter. Since a good spectrum can be more stably estimated and the difference in the narrowband spectral code can be reflected in the estimation of the broadband amplitude information, there is an effect that a wideband speech having a more correct amplitude can be restored.

Furthermore, since the post-filtering is performed on the wideband speech signal estimated using the narrowband speech code, when the post-filter processing is applied to the narrowband synthesized sound, the narrowband portion and the restored band are used. This has the effect of improving continuity. Further, when the sound quality of the restored wideband audio signal is insufficient, there is an effect that sound quality can be improved such as enhancement of pitch periodicity and enhancement of poles of a spectrum envelope.

BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a configuration diagram of a wideband audio restoration device according to a first embodiment of the present invention. FIG. 3 is an explanatory diagram illustrating a process of a zero filling unit according to the first embodiment of the present invention. FIG. 9 is a configuration diagram of a wideband sound source estimating unit in the wideband sound restoration apparatus according to the second embodiment of the present invention. FIG. 9 is an explanatory diagram illustrating an example of an adaptive excitation signal according to the second embodiment of the present invention. FIG. 11 is a configuration diagram of a wideband driving sound source estimating unit in the wideband audio restoration apparatus according to the third embodiment of the present invention. FIG. 13 is a configuration diagram of a wideband driving sound source estimating unit in a wideband audio restoration apparatus according to a fourth embodiment of the present invention. FIG. 14 is a configuration diagram of a wideband sound source estimating unit in a wideband speech restoration apparatus according to a fifth embodiment of the present invention. FIG. 14 is a configuration diagram of a wideband driving sound source estimating unit in the wideband audio restoration apparatus according to the sixth embodiment of the present invention. FIG. 14 is a configuration diagram of a wideband driving sound source estimating means in the wideband audio restoration apparatus according to the seventh embodiment of the present invention. FIG. 18 is a configuration diagram of a wideband sound source estimating unit in the wideband speech restoration apparatus according to the eighth embodiment of the present invention. FIG. 19 is a configuration diagram of a wideband audio restoration apparatus according to a ninth embodiment of the present invention. FIG. 21 is a configuration diagram of a wideband audio restoration apparatus according to Embodiment 10 of the present invention. FIG. 21 is a configuration diagram of a wideband audio restoration apparatus according to Embodiment 11 of the present invention. FIG. 21 is a configuration diagram of a wideband audio restoration apparatus according to Embodiment 12 of the present invention. FIG. 21 is an explanatory diagram illustrating a general relationship between a narrowband spectrum and a wideband spectrum in Embodiment 13 of the present invention. FIG. 19 is an explanatory diagram for explaining a general relationship between a narrowband spectrum and a wideband spectrum in Embodiment 14 of the present invention. FIG. 21 is a configuration diagram of a wideband spectrum estimating means in a wideband speech restoration apparatus according to Embodiment 15 of the present invention. FIG. 21 is a configuration diagram of a wideband audio restoration apparatus according to Embodiment 17 of the present invention. FIG. 28 is a configuration diagram of a wideband sound source decoding means in the wideband audio restoration apparatus according to the eighteenth embodiment of the present invention. FIG. 21 is a configuration diagram of a wideband sound source decoding means in a wideband sound restoration apparatus according to a nineteenth embodiment of the present invention. FIG. 21 is a configuration diagram of a wideband sound source decoding means in a wideband speech restoration apparatus according to a twentieth embodiment of the present invention. FIG. 27 is a configuration diagram of a wideband audio restoration apparatus according to Embodiment 22 of the present invention.

Explanation of reference numerals

1} narrowband speech signal, 2} analysis means, 3} spectrum analysis means, 4} narrowband spectrum parameter, 5} inverse filter, 6} narrowband excitation signal, 7} wideband spectrum estimation means, 8} vector quantization means, 9] narrowband spectrum codebook, 10 spectrum code, 11 inverse quantization means, 12 wideband spectrum codebook, 13 wideband spectrum parameters, 14 wideband excitation estimation means, 15 zero-filling means, 16 wideband excitation signal, 17 synthesis filter, 18 bandpass filter, 19 upsampling means, 20 wideband speech signal, 21 sound source analysis means, 22 narrowband adaptive codebook, 23 distortion minimization means, 24 narrowband drive excitation signal, 25 narrowband adaptive lag length, 26 narrowband adaptive gain, 27 wideband drive excitation estimation means, 28 zero filling means, 29 wide band Driving excitation signal, 30 、 wideband adaptive excitation estimating means, 31 wideband adaptive excitation codebook, 32 wideband adaptive excitation signal, 33 wideband adaptive lag length, 34 wideband adaptive gain, 35 power calculation means, 36 noise generation means, 37 narrowband long period Prediction analysis means, 38 narrow band long cycle delay, 39 narrow band long cycle prediction coefficient, 40 long cycle inverse filter, 41 narrow band long cycle prediction residual signal, 42 wide band long cycle prediction residual estimation means, 43 zero filling means, 44 wideband long-period prediction parameter estimation means, 45 wideband long-period delay coefficient, 46 wideband long-period prediction coefficient, 47 long-period synthesis filter, 48 wideband long-period prediction residual signal, 49 upsampling means, 50 zeroing means, 51 narrowband Power calculation means, 52 narrow band sound source power, 53 narrow band power included spectrum code , 54 excitation normalizing means, 55 narrowband normalized excitation signal, 56 wideband normalized excitation signal, 57 wideband power codebook, 58 wideband excitation power estimation, 59 wideband excitation power estimation means, 60 wideband power spectrum codebook, 61 post Filter means, 62 spectrum parameter conversion means, 63 degree reduction means, 64 spectrum parameter inverse conversion means, 101 narrowband speech code, 102 separation means, 103 narrowband spectrum code, 104 narrowband excitation code, 105 wideband spectrum decoding means, 106 Broadband excitation decoding means, 107 narrowband spectrum decoding means, 108 narrowband excitation decoding means, 109 narrowband speech decoding means, 110 narrowband pitch code, 111 narrowband power code, 112 wideband pitch decoding means, 113 wideband Pitch period, 114 {Wide band power decoding means, 115} Wide band power decoding means, 116} Excitation generation means, 117 # Narrow band driving excitation code, 118 # Narrow band adaptive excitation decoding means, 119 # Wide band driving excitation decoding means, 120 # Wide band driving excitation decoding means, 121 # Narrow Bandwidth adaptive excitation decoding means, 122 {narrowband driving excitation decoding means, 123} narrowband long-period prediction parameter decoding means, 124 wideband long-period prediction parameter decoding means, 125 {narrow-band long-period prediction parameter decoding means, 126} narrow-band long-period prediction residual code 127, wideband long-period prediction residual decoding means, 128, narrowband long-period prediction residual decoding means, 129, narrowband power decoding means, 130, wideband normalized excitation decoding means.

Claims (12)

Spectrum decoding means for decoding a narrow-band spectral parameter from the narrow-band spectral code, extending the spectrum envelope of the decoded narrow-band spectral parameter in the frequency axis direction, and outputting as a wide-band spectral parameter;
A wideband speech restoration apparatus comprising a synthesizing means for generating a wideband speech signal using the outputted wideband spectrum parameters. Comprising a wideband sound source estimating means for estimating a wideband sound source signal,
2. The wideband audio restoration apparatus according to claim 1, wherein the synthesis unit is configured to generate a wideband audio signal from the wideband sound source signal and a wideband spectrum parameter. A spectrum decoding step of decoding a narrow-band spectrum parameter from the narrow-band spectrum code, extending a spectrum envelope of the decoded narrow-band spectrum parameter in the frequency axis direction, and outputting as a wide-band spectrum parameter;
A wide-band speech restoration method including a synthesis step of generating a wide-band speech signal using the outputted wide-band spectrum parameters. (4) A wideband audio restoration apparatus that expands a spectrum envelope of a narrowband spectral parameter obtained by analyzing a narrowband audio signal in a frequency axis direction and generates a wideband audio signal using the wideband spectrum parameter. Comprising a wideband sound source estimating means for estimating a wideband sound source signal,
The wideband audio restoration apparatus according to claim 4, wherein a wideband audio signal is generated from the generated wideband sound source signal and the wideband spectrum parameter. (4) A wideband audio restoration method for generating a wideband audio signal by extending a spectrum envelope of a narrowband spectral parameter obtained by analyzing a narrowband audio signal in a frequency axis direction and using the spectrum envelope as a wideband spectral parameter. In a speech transmission system including a speech encoding device that encodes a narrowband speech signal and outputs a narrowband speech code, and a wideband speech restoration device that receives the narrowband speech code and generates a wideband speech signal,
The wideband audio restoration device,
Spectral decoding for decoding a narrowband spectral parameter from the narrowband spectral code included in the received narrowband speech code, extending the spectrum envelope of the decoded narrowband spectral parameter in the frequency axis direction, and outputting as a wideband spectral parameter. Means,
An audio transmission system including a synthesizing unit that generates a wideband audio signal using the output wideband spectral parameters. A voice encoding step of encoding a narrowband voice signal to output a narrowband voice code, and a wideband voice restoration step of receiving the narrowband voice code and generating a wideband voice signal,
The wideband audio restoration step includes:
Spectral decoding for decoding a narrowband spectral parameter from the narrowband spectral code included in the received narrowband speech code, extending the spectrum envelope of the decoded narrowband spectral parameter in the frequency axis direction, and outputting as a wideband spectral parameter. Steps and
A voice transmission method including a synthesis step of generating a wideband voice signal using the output wideband spectral parameters. The said spectrum decoding means performs the deformation | transformation which suppresses the pole of the spectrum envelope of the decoded narrow-band spectrum parameter, and outputs the narrow-band spectrum parameter which performed the deformation | transformation which suppresses the said pole as a wide-band spectrum parameter. Item 2. The wideband audio restoration device according to Item 1. The spectrum decoding means performs a deformation to smooth the pole structure of the spectral envelope of the decoded narrowband spectrum parameter, and outputs the narrowband spectrum parameter subjected to the deformation to smooth the pole structure as a wideband spectrum parameter. 2. The wideband audio restoration apparatus according to claim 1, wherein: The spectrum decoding step performs a modification for suppressing a pole of a spectrum envelope of the decoded narrowband spectral parameter, and outputs the narrowband spectral parameter subjected to the modification for suppressing the pole as a wideband spectral parameter. Item 3. The wideband audio restoration method according to Item 3. The spectrum decoding step performs a deformation to smooth the pole structure of the spectral envelope of the decoded narrowband spectrum parameter, and outputs the narrowband spectrum parameter that has been deformed to smooth the pole structure as a wideband spectrum parameter. The method of claim 3, wherein the wideband sound is restored.

Images