JP5203077B2 - Speech coding apparatus and method, speech decoding apparatus and method, and speech bandwidth extension apparatus and method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To improve the subjective quality of a decryption signal by precisely adjusting the microstructure of a high-frequency spectrum dependent on frequency in speech coding and speech decoding in a band spreading technique. <P>SOLUTION: The speech decoder for decoding a speech signal including a frequency component corresponding to both low-frequency spectrum and high-frequency spectrum from a bit stream including encoded information of the low-frequency spectrum of speech signal and auxiliary information related to the nature of the high-frequency spectrum includes: a means for separating the bit stream into encoded information of the low-frequency spectrum of the speech signal and the auxiliary information related to the nature of the high-frequency spectrum; a means for decoding the low-frequency spectrum from the encoded information of the low-frequency spectrum of the speech signal; a means for obtaining a first high-frequency spectrum by copying part or the whole of the low-frequency spectrum obtained by decoding to the high-frequency region; a spectrum shape adjusting means for obtaining a second high-frequency spectrum by adjusting the shape of the first high-frequency spectrum utilizing the auxiliary information; and a means for decoding the speech signal utilizing the low-frequency spectrum and the second high-frequency spectrum. <P>COPYRIGHT: (C)2010,JPO&amp;INPIT

Description

  The present invention relates to a voice encoding apparatus and method for encoding a voice signal, a voice decoding apparatus and method for decoding a coded signal, and a voice band extending apparatus and method.

  A speech coding technique for compressing speech to a low bit rate is important for effectively utilizing a limited transmission capacity of a communication path in speech communication. In recent years, there has been an increasing demand for higher quality voice communication. In contrast, voice can be transmitted with high quality and various ambient sounds and music other than human voice can be transmitted. Research and development of speech coding methods is underway.

  Conventionally, for mobile phones and VoIP (Voice over Internet Protocol), Adaptive Multi Rate, an audio encoding method that can transmit narrowband signals of 300 Hz to 3.4 kHz with an information amount of about 4.75 kbit / s to 64 kbit / s. (AMR), G.711, G.729, and the like have been used. On the other hand, audio coding methods such as AMR-WB and G.722.1 that can transmit wideband signals up to 7KHz at a bit rate of about 6.6kbit / s to 32kbit / s have been standardized by 3GPP and ITU-T. Yes. In addition, G.722.1C, AAC-LD, etc. have been standardized by ITU-T and MPEG as voice / acoustic coding systems that can transmit ultra-wideband signals up to about 15 kHz at bit rates of 24 kbps to 128 kbps. Yes. By using these ultra-wideband coding methods, almost all frequency components included in the audible frequency band can be transmitted, and various ambient sounds and music other than human voice can be transmitted with quality close to the original sound. It is possible and high-quality voice communication is possible. On the other hand, these methods have a problem that the bit rate required for transmission becomes relatively high.

  On the other hand, as a method for encoding a signal having a wide frequency band at a low bit rate, there is a speech encoding method based on band expansion. An example of a speech encoding method based on band expansion is described in Patent Document 1. In the speech coding method based on bandwidth expansion, spectral coefficients were copied from low to high in transform regions such as QMF (Quadrature Mirror Filters), FFT (Fast Fourier Transform), and MDCT (Modified Discrete Cosine Transform). Later, high-quality decoded speech can be obtained by adjusting the spectral envelope of the copied coefficients. In a speech coding system using this method, it is not necessary to transmit information about the fine structure of the high-frequency spectrum of the signal, and only the outline of the spectrum needs to be transmitted at a low bit rate. The amount of stream information can be greatly reduced.

  As described above, the band extension technology is an effective means for reducing the bit rate of speech / acoustic coding. However, when the high frequency component is generated by copying the spectrum from the low frequency to the high frequency, the generated high frequency is generated. Inconsistency in the fine structure of the spectrum between the high frequency component of the original component and the high frequency component may lead to deterioration of the sound quality of the decoded audio. For the copied spectrum, the spectral envelope is usually adjusted using auxiliary information given from the encoder, but the influence of the fine structure of the spectrum still remains after the envelope adjustment. The influence of the fine structure of the spectrum is illustrated below.

  FIG. 1 shows an example of a spectrum of an audio signal obtained using STFT (Short-time Fourier Transform). In a low frequency range up to about 4 kHz, a steep spectral peak appears at a position that is an integral multiple of the fundamental frequency, so that it is understood that a tone component composed of harmonic series of the fundamental frequency is dominant in the signal. On the other hand, in the high region of 4 kHz or higher, it can be observed that the peak gradually weakens as the frequency increases, and the noise component having no tone property becomes dominant. Thus, in the voice, the noise component tends to be dominant as the frequency increases.

  FIG. 2 shows an example of the result of adjusting the spectral envelope after copying the spectrum from the low frequency to the high frequency for the signal shown in FIG. Although the spectrum of FIG. 2 has a global envelope similar to that of FIG. 1, the peak dip shape of the high-frequency spectrum is significantly different from that of FIG. 1, and the fine structure of the spectrum is greatly different from that of the original signal.

  The difference in the fine structure shown in the above examples has an adverse effect on subjective speech quality. In other words, when listening to band-expanded speech in which only the spectral envelope has been adjusted, an unnatural timbre different from the original speech is perceived due to the difference in the fine structure of the high-frequency spectrum. For example, in speech, the noise component becomes dominant at higher frequencies as described above, so unpleasant metallic tone due to the mismatch of the fine structure between the high frequency component generated by the band extension technique and the original signal is often found in the decoded speech. appear.

Therefore, in some band expansion technologies, fine structure adjustment is performed on the high-frequency spectrum generated by copying. As a method for adjusting the spectral fine structure, Patent Document 2 discloses a method of copying to a high band after uniformly limiting the amplitude of a low band spectrum.
JP-T-2001-521648 Republication 2005-111568

  However, according to the method disclosed in Patent Document 2, since the fine structure of the entire high-frequency spectrum generated by copying the low-frequency spectrum is uniformly adjusted, the fine structure of the spectrum of the original signal depends on the frequency. Even if it changes variously, the fine structure of the generated high frequency spectrum will be uniform. Therefore, the first problem to be solved by the present invention is that, in speech coding and speech decoding based on the band extension technology, the fine structure of the high-frequency spectrum depending on the frequency is accurately adjusted, and the decoded signal To improve subjective quality.

  In addition, when using a real-valued filter bank for conversion to the frequency domain due to the reduction in the amount of computation and the configuration of the codec, if the shape of the spectrum is deformed, harmful aliasing distortion occurs in the decoded signal, The subjective quality of the decoded signal may be adversely affected. As shown in FIG. 4, this distortion appears to spread to frequencies near the spectral coefficient whose amplitude is expanded and contracted. When the power spectral density of distortion is large and the power spectral density of noise components is small, the distortion may lead to significant deterioration in subjective sound quality. Therefore, the second problem to be solved by the present invention is to adjust the spectral fine structure while suppressing the deterioration of sound quality due to distortion caused by the deformation of the spectrum in the speech coding and speech decoding based on the band expansion technology. It is to improve the subjective quality of the decoded signal.

The speech decoding apparatus according to the present invention corresponds to both the low-frequency spectrum and the high-frequency spectrum from a bitstream including information obtained by encoding the low-frequency spectrum of the speech signal and auxiliary information regarding the nature of the high-frequency spectrum. An audio decoding device for decoding an audio signal including a frequency component to be transmitted, wherein the bit stream is separated into information obtained by encoding the low frequency spectrum of the audio signal and auxiliary information relating to the nature of the high frequency spectrum; Means for decoding the low-frequency spectrum from information obtained by encoding the low-frequency spectrum of the audio signal, and copying the part or the whole of the low-frequency spectrum obtained by decoding to the high frequency. By adjusting the shape of the first high-frequency spectrum using means for obtaining a spectrum and auxiliary information regarding the properties of the high-frequency spectrum. Spectral shape adjustment means for obtaining a second high-frequency spectrum Te, and means for decoding the audio signal using the above-described low band spectrum and the second high-frequency spectrum, and auxiliary information about the nature of the high frequency spectrum Spectrum shape adjustment method determining means for determining how to adjust the shape used by the spectrum shape adjusting means or the amount of noise to be superimposed based on the first high frequency spectrum, and the spectrum shape adjustment The method determining means uses the distortion generated in the output signal due to the adjustment of the shape of the spectrum and the masking amount by the high-frequency spectrum after the shape adjustment to determine how to adjust the shape or the amount of noise to be superimposed. It is characterized by determining . Here, it is preferable that the spectrum shape adjustment method determining unit compares the tone-to-noise ratio of the high frequency spectrum of the speech signal with the tone-to-noise ratio of the first high frequency spectrum.

  Here, it is desirable that the spectrum shape adjusting means obtains the second high frequency spectrum by superimposing noise in addition to adjusting the shape of the first high frequency spectrum.

  Further, the spectrum shape adjusting means is configured to divide the first high frequency spectrum into a plurality of subbands and adjust the shape of the partial spectrum for each partial spectrum belonging to each of the divided subbands. Is desirable.

  At this time, the auxiliary information related to the nature of the high frequency spectrum includes information related to the envelope in the time domain of the high frequency component of the voice, and the spectrum shape adjusting means adjusts the envelope in the time domain of the superimposed noise and It is desirable to do.

A speech coding apparatus according to the present invention is a speech coding apparatus that generates a bit stream suitable for a speech decoding apparatus that decodes a high-frequency spectrum based on copying of a decoding result of a low-frequency spectrum to a high frequency. The voice signal is converted into the frequency domain, and after the low frequency spectrum is separated, a part or the whole of the remaining high frequency spectrum is separated as the first high frequency spectrum, and the low frequency spectrum is encoded. Means for obtaining a second high-frequency spectrum by copying a part or the whole of the low-frequency spectrum to a high frequency, and properties of the first high-frequency spectrum and the second high-frequency spectrum. Based on the spectrum shape adjustment method determining means for determining how to adjust the shape of the high frequency spectrum in the speech decoding apparatus, the encoded low frequency spectrum, and the determination Is provided with means for generating a bit stream by multiplexing the high frequency band spectrum of the manner of adjustment of the shape information, the spectral shape adjustment method determining means, the output signal due to the adjustment of the shape of the spectrum The method of adjusting the shape of the high-frequency spectrum or the amount of noise to be superimposed is determined using the distortion that occurs and the masking amount by the high-frequency spectrum after adjusting the spectral shape . Here, it is preferable that the spectrum shape adjustment method determination unit compares the tone-to-noise ratio of the first high-frequency spectrum with the tone-to-noise ratio of the second high-frequency spectrum.

  Here, the spectrum shape adjustment method determining means includes the speech decoding apparatus in addition to the method of adjusting the shape of the high frequency spectrum based on the properties of the first high frequency spectrum and the second high frequency spectrum. It is desirable that the amount of noise to be superimposed on the high frequency spectrum is determined in order to adjust the shape of the high frequency spectrum at.

  The spectrum shape adjustment method determining means divides the first high-frequency spectrum into a plurality of subbands, and determines how to adjust the shape of the partial spectrum for each of the divided subbands. It is desirable to adopt a configuration that does this.

An audio band extending apparatus according to the present invention is an audio band extending apparatus that restores an audio signal including a high frequency component not included in the low frequency spectrum from a low frequency spectrum including only a low frequency component of the audio. Means for obtaining a first high-frequency spectrum by copying a part or the whole of the low-frequency spectrum to a high frequency; and adjusting the shape of the first high-frequency spectrum to thereby obtain a second high-frequency spectrum. A spectral shape adjusting means for obtaining a sound signal including a high frequency component using the low frequency spectrum and the second high frequency spectrum, and a property of the first high frequency spectrum. A spectral shape adjustment method determining means for determining how to adjust the shape of the high frequency spectrum in the speech decoding device, the spectral shape adjustment method determining means, How to adjust the shape of the high-frequency spectrum or the amount of noise to be superimposed using the distortion generated in the output signal due to the spectral shape adjustment and the masking amount by the high-frequency spectrum after the spectral shape adjustment It is characterized by determining . Here, it is preferable that the spectrum shape adjustment method determination unit compares the tone-to-noise ratio of the first high-frequency spectrum with the tone-to-noise ratio of the high-frequency spectrum estimated from the low-frequency spectrum. The spectrum shape adjusting means is preferably configured to obtain the second high frequency spectrum by superimposing noise in addition to adjusting the shape of the first high frequency spectrum.

  Here, the spectrum shape adjusting means is configured to divide the first high-frequency spectrum into a plurality of subbands and adjust the shape of the partial spectrum for each partial spectrum belonging to each of the divided subbands. It is desirable.

  According to the present invention as described above, the subjective quality of a decoded signal can be improved in speech encoding and speech decoding based on a band extension technique.

By the way, the present invention relating to a speech decoding apparatus can also be regarded as an invention of a speech decoding method and can be described as follows. The speech decoding method according to the present invention corresponds to both the low-frequency spectrum and the high-frequency spectrum from a bitstream including information obtained by encoding the low-frequency spectrum of the speech signal and auxiliary information regarding the nature of the high-frequency spectrum. A speech decoding method executed by a speech decoding apparatus that decodes a speech signal including a frequency component to be encoded, wherein the bitstream is encoded with information obtained by encoding a low-frequency spectrum of the speech signal, and a property of a high-frequency spectrum And sub-information, and a step of decoding the low-frequency spectrum from the information obtained by encoding the low-frequency spectrum of the audio signal, and a part or the whole of the low-frequency spectrum obtained by the decoding is made high frequency. Obtaining the first high-frequency spectrum by copying, and using auxiliary information relating to the nature of the high-frequency spectrum A spectral shape adjustment step of obtaining a second high-frequency spectrum by adjusting the shape of one high-frequency spectrum, and a step of decoding an audio signal using the low-frequency spectrum and the second high-frequency spectrum; A spectral shape adjustment method for determining the shape adjustment method or the amount of noise to be superimposed used by the spectral shape adjustment step based on the auxiliary information on the nature of the high frequency spectrum and the first high frequency spectrum A step of determining the shape of the spectrum using the distortion generated in the output signal due to the adjustment of the shape of the spectrum and the masking amount by the high-frequency spectrum after the shape adjustment. It is characterized in that the adjustment method or the amount of noise to be superimposed is determined . At this time, in the spectrum shape adjustment step, it is desirable that the speech decoding apparatus obtains the second high frequency spectrum by superimposing noise in addition to the adjustment of the shape of the first high frequency spectrum.

Further, the present invention relating to a speech coding apparatus can also be understood as a speech coding method invention and can be described as follows. The speech coding method according to the present invention includes a speech decoding device that generates a bitstream suitable for a speech decoding device that decodes a high frequency spectrum based on a copy of a decoding result of a low frequency spectrum to a high frequency. A speech encoding method to be performed, comprising: converting a speech signal into a frequency domain, separating a low-frequency spectrum, and then separating a part or all of the remaining high-frequency spectrum as a first high-frequency spectrum; Encoding the low-frequency spectrum, obtaining a second high-frequency spectrum by copying a part or the whole of the low-frequency spectrum to the high frequency, the first high-frequency spectrum, and the first A spectral shape adjustment method determining step for determining how to adjust the shape of the high frequency spectrum in the speech decoding apparatus based on the properties of the high frequency spectrum of 2 And encoded the low frequency spectrum, and a step of generating a bit stream by multiplexing the information of how the adjustment of the shape of the high frequency band spectrum determined above, the spectral shape adjustment method determining step, Using the distortion generated in the output signal due to the adjustment of the spectrum shape and the masking amount by the high frequency spectrum after the spectrum shape adjustment, how to adjust the shape of the high frequency spectrum or the noise to be superimposed It is characterized by determining the quantity . At this time, in the spectrum shape adjustment method determination step, the speech encoding apparatus determines how to adjust the shape of the high frequency spectrum based on the properties of the first high frequency spectrum and the second high frequency spectrum. In addition, it is desirable to determine the amount of noise to be superimposed on the high frequency spectrum in order to adjust the shape of the high frequency spectrum in the speech decoding apparatus.

Further, the present invention relating to a voice band extending apparatus can also be regarded as an invention of a voice band extending method and can be described as follows. An audio band extending method according to the present invention is performed by an audio band extending apparatus that restores an audio signal including a high frequency component not included in the low frequency spectrum from a low frequency spectrum including only a low frequency component of the audio. A method for expanding a voice band, the step of obtaining a first high frequency spectrum by copying a part or the whole of the low frequency spectrum to a high frequency, and adjusting the shape of the first high frequency spectrum it is a step in which the spectral shape adjustment step of obtaining a second high-frequency spectrum, by using the above low frequency band spectrum and the second high-frequency spectrum, to restore the audio signals containing high frequency components, wherein A spectral shape adjustment method determining step for determining how to adjust the shape of the high frequency spectrum in the speech decoding apparatus based on the property of the first high frequency spectrum; And the spectral shape adjustment method determining step uses the distortion generated in the output signal due to the adjustment of the shape of the spectrum and the masking amount by the high frequency spectrum after the adjustment of the spectrum shape. It is characterized in that the method of adjusting the shape or the amount of noise to be superimposed is determined . In the spectrum shape adjustment step, it is desirable that the speech decoding apparatus obtains the second high frequency spectrum by superimposing noise in addition to adjusting the shape of the first high frequency spectrum.

  According to the present invention, in speech coding and speech decoding based on the band extension technique, even if the fine structure of the high-frequency spectrum in the original signal differs between subbands, as shown in FIG. The fine structure of the region spectrum can be adjusted with high accuracy. In addition, the spectral fine structure can be adjusted while minimizing the perceptual effect of distortion associated with the adjustment. As described above, the subjective quality of the decoded signal can be improved in speech encoding and speech decoding based on the band extension technique.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings.

(Embodiment 1)
FIG. 5 is a block diagram showing the main configuration of speech encoding apparatus 501 according to Embodiment 1 of the present invention. As illustrated in FIG. 5, the speech encoding apparatus 501 includes a frequency domain conversion unit 503 that converts an input signal into a frequency domain, a low frequency encoding unit 509 that encodes a low frequency spectrum, and a low frequency A low frequency decoding unit 515 that decodes the code given from the encoding unit 509 and a part or the whole of the input low frequency spectrum are copied, so that a high frequency spectrum having a component equal to or higher than a predetermined boundary frequency is obtained. A spectrum copying unit 517 to be created, subband division units 511 and 521 that divide a high-frequency spectrum into a plurality of subbands based on a predetermined division rule, and a spectral envelope of the high-frequency spectrum decomposed into subbands Spectral envelope calculation unit 527 and spectral shape for determining the shape adjustment method of the high frequency spectrum for each subband using the high frequency spectrum divided into subbands A bit stream obtained by multiplexing the code provided from the adjustment method determination unit 525, the low frequency encoding unit 509, the spectrum shape adjustment information provided from the spectrum shape adjustment method determination unit 525, and the spectrum envelope information provided from the spectrum envelope calculation unit 527 And a bit stream multiplexing unit 529 for generating.

  As shown in FIG. 22, the speech encoding apparatus 501 includes, for example, a CPU 2201 that executes an operating system and application programs, a main storage unit 2202 including ROM and RAM, a hard disk, and the like. Auxiliary storage unit 2203, communication control unit 2204 for performing data communication with the outside, display unit 2205 configured with a liquid crystal monitor, etc., and operation unit 2206 for performing character / number input and execution instructions, etc. The The functions of the blocks in FIG. 5 described above are realized by reading and executing predetermined software in the main storage unit 2202 under the control of the CPU 2201 shown in FIG.

  The audio encoding device 501 performs a series of processes shown in the flowchart of FIG. 6 on the input signal, and outputs the encoded bit stream to the outside. Hereinafter, a series of processes in FIG. 6 will be described.

  In step 601 of FIG. 6, the frequency domain conversion unit 503 converts the input signal into the frequency domain. Here, as a conversion method, discrete Fourier transform (DFT), discrete cosine transform (DCT), modified discrete cosine transform (MDCT), cosine modulation filter bank, or the like is used. The spectrum converted into the frequency domain is separated into a low frequency spectrum 505 and a high frequency spectrum 507 using a predetermined boundary frequency Fc. Further, the high frequency spectrum 507 may include only a part of the frequency band equal to or higher than the boundary frequency Fc. The frequency domain transforming unit 503 or the frequency domain transforming step 601 first separates the input signal into a low-frequency signal and a high-frequency signal using a band-pass filter or a filter bank, and then converts each signal into the frequency domain. It can also be realized by doing.

  In the next step 603, the low frequency encoding unit 509 encodes the low frequency spectrum 505 and gives the obtained code to the low frequency decoding unit 515. In the next step 605, the low frequency decoding unit 515 decodes the code given from the low frequency encoding unit 509 and gives the low frequency spectrum obtained by the decoding to the spectrum copying unit 517. As an alternative implementation means, the low band decoding unit 515 in FIG. 5 and step 605 in FIG. 6 can be omitted, and the low band spectrum 505 can be directly provided to the spectrum copying unit 517.

  In the next step 607, the spectrum copying unit 517 creates a high frequency spectrum 519 having a component equal to or higher than the boundary frequency Fc by copying a part or the whole of the input low frequency spectrum, and the subband dividing unit 521. To give. The high frequency spectrum 519 may include only a part of the frequency band not lower than the boundary frequency Fc. As the above copying method, copying is performed by shifting the high-frequency spectrum to the high frequency as shown in FIG. 7A, or so as to be line-symmetric with respect to the boundary frequency Fc as shown in FIG. 7B. It is possible to adopt a method of performing copying by inverting the spectrum. When the bandwidth of the high-frequency spectrum generated by copying is wider than the low-frequency spectrum used for copying, the copying can be repeated as shown in FIG.

  In the next step 609, the subband division units 511 and 521 divide the high-frequency spectra 507 and 519 into a plurality of subbands based on the same division rule, and give them to the spectrum shape adjustment method determination unit 525. Note that the subband splitting units 511 and 521 in FIG. 5 and step 609 in FIG. 6 are not essential, and the entire high-frequency spectra 507 and 519 may be given to the spectrum shape adjustment method determining unit 525.

  In next step 611, spectrum shape adjustment method determination section 525 determines the shape adjustment method of the high frequency spectrum on the speech decoding apparatus side for each subband using subband-divided high frequency spectra 513 and 523. Then, information indicating the determined method is provided to the bit stream multiplexing unit 529. When the subband division units 511 and 521 in FIG. 5 and step 609 in FIG. 6 are omitted, the spectrum shape adjustment method determination unit 525 handles the entire high frequency spectrum as one subband. The processing in step 611 will be described in detail later with reference to FIGS.

  In the next step 613, the spectrum envelope calculation unit 527 calculates the spectrum envelope of the high frequency spectrum decomposed into subbands by the subband division unit 511, and provides the spectrum envelope to the bitstream multiplexing unit 529. The spectral envelope can be expressed in the form of the average power of the spectrum within each subband. As an alternative configuration, the spectrum envelope calculation unit 527 can directly receive the input signal, perform LPC analysis, and provide the obtained LPC parameters to the bitstream multiplexing unit 529. The spectrum envelope calculation unit 527 in FIG. 5 and step 613 in FIG. 6 are not essential to obtain the effect of the present invention, and can be omitted.

  In the next step 615, the bit stream multiplexing unit 529 receives the code given from the low frequency coding unit 509, the spectrum shape adjustment information given from the spectrum shape adjustment method determination unit 525, and the spectrum envelope calculation unit 527. A bit stream is generated by multiplexing the given spectral envelope information.

  As described above, the speech encoding apparatus 501 generates a bit stream and outputs it to the outside.

  Hereinafter, the configuration of the spectral shape adjustment method determination unit 525 in FIG. 8 and the processing in step 611 in FIG. 6 will be described in detail.

  As described above, in step 611 of FIG. 6, the spectrum shape adjustment method determination unit 525 uses the highband spectrums 513 and 523 obtained by subband division to change the shape adjustment method of the highband spectrum in the decoder to the subband. Information indicating the determined method is given to the bit stream multiplexing unit 529.

スペクトル正規化部８０１、８０３は省略することもできる。 The spectrum shape adjustment method determination unit 525 that performs the above processing has the configuration of FIG. Spectral normalization sections 801 and 803 shown in FIG. 8 normalize the amplitudes of the subband-divided high frequency spectra 513 and 523 for each subband. The normalization method follows Formula 1. However, the spectrum coefficient before normalization is represented by S ′ (k), and the spectrum coefficient after normalization is represented by S (k). The index k indicates a frequency, and the i-th subband includes a frequency of F _i ≦ k <F _{i + 1} . However, F _i is the lower limit of the frequency included in the i-th subband. Further, max () represents a maximum value and || represents an absolute value.

The spectrum normalization units 801 and 803 can be omitted.

  Note that the spectrum shape adjustment method determination unit 525 in FIG. 8 may be provided with an LPC analysis unit 805 that performs LPC analysis of the high frequency spectrum of the input signal. For example, when performing the time envelope adjustment of the noise waveform by the LPC filter in the speech decoding apparatus, the high frequency spectrum of the input signal is LPC analyzed by the LPC analysis unit 805, and the parameters obtained by the analysis are used as the spectrum shape adjustment information. Can be included.

式２、式３、式４のいずれの定義に従っても、Ｔ_ｉはトーン成分の雑音成分に対する比率が高いほど大きな値をとる。 The tone-to-noise ratio calculation units 807 and 809 calculate an index T _i indicating the tone-to-noise ratio of each subband using the normalized spectrum given from the spectrum normalization units 801 and 803, and the parameter determination unit 811 give. T _i is can be defined in several ways, for example Formula 2, Formula 3, can be used to define, as shown in Equation 4. Here, mean () represents an arithmetic mean, gmean () represents a geometric mean, and min () represents a minimum value.

Formula 2, Formula 3, also in accordance with any of the definitions of Formula 4, T _i takes a larger value the higher the ratio of the noise component of the tone component.

The parameter determination unit 811 compares the tone-to-noise ratio of the copied high-frequency spectrum 523 and the high-frequency spectrum 5 13 of the input signal, and determines the spectral shape adjustment method of the high-frequency spectrum in the decoder.

The spectrum shape is adjusted according to Equation 5 with Sc (k) as the copied high-frequency spectrum 523 and St (k) as the high-frequency spectrum after the spectrum shape adjustment. That is, the determination of the spectral shape adjustment method is equivalent to determining the parameters α _i and β _i in the expression of Equation 5. Here, f (x, α _i ) is a nonlinear function having x as an input, and α _i is a parameter for controlling the shape of the function. Rv is noise having a predetermined power, and β _i is a parameter indicating the magnitude of noise superimposed on the spectrum. sign () indicates a positive or negative sign.
St (k) = sign (Sc (k)) · f (| Sc (k) |, α _i ) + β _i · rv
where F _i ≦ k <F _{i + 1} (Formula 5)
The specific shape of f (x, α _i ) can be determined as shown in Equation 6, Equation 7, and Equation 8, for example. However, max (a, b) represents the maximum value of a and b, and min (a, b) represents the minimum value of a and b.
_{f (x, α i) =} x αi ... ( Equation 6)
f (x, α _i ) = min (x, α _i ) (Expression 7)
f (x, α _i ) = max (0, x + α _i ) (Equation 8)
The parameter determination unit 811 sets the parameter α _i so that the tone-to-noise ratio in each subband of St (k) obtained by the adjustment according to Equation 5 matches the tone-to-noise ratio in each subband of the high frequency spectrum 513 of the input signal. , Β _i are determined. The parameter determination process executed by the parameter determination unit 811 follows the flowchart of FIG. 9 described below.

In step 901 of FIG. 9, the parameter determination unit 811 selects one from the unprocessed subbands, and acquires the tone-to-noise ratio of the high-frequency spectra 513 and 523 in step 903. In the next step 905, β _i is set to 0, and in step 907, the tone-to-noise ratios of the high-frequency spectra 513 and 523 are compared. Here, if the tone-to-noise ratio of the high-frequency spectrum 513 of the input signal is higher than the tone-to-noise ratio of the copied high-frequency spectrum 523, step 909 described later is executed, and the tone pair of the high-frequency spectrum 513 is executed. When the noise ratio is equal to or lower than the tone-to-noise ratio of the high frequency spectrum 523, Step 911 described later is executed.

In step 909, α _i that matches the tone-to-noise ratio of the high-frequency spectrum 513 and St (k) is searched under the condition of “α _i > 1”. This search can be performed by calculating the tone-to-noise ratio of St (k) while increasing α _i using a constant step width.

On the other hand, in step 911, α _{i in} which the tone-to-noise ratio of high frequency spectrum 513 and St (k) coincide is similarly searched under the condition of “α _i <1”. The next step 913 is executed by the masking determination unit 813. The processing in step 913 will be described later with reference to FIG. 10. In this step 913, the masking determination unit 813 determines whether distortion is masked when α _i obtained in step 911 is used. . If it is determined in step 913 that the distortion is not masked, a constant value is added to α _i in step 915, and the processing in step 911 is executed again. On the other hand, in step 913, after the case and steps 909 strain is determined to Ru masked is executed at step 917, the determined alpha _i, beta _i is output. Then, in step 919, an unprocessed subband is searched, and in the case where processing has not been completed for all subbands, the processing from step 901 onward is executed again for the unprocessed subband. The parameters α _i and β _i are determined by the processing of FIG. 9 as described above.

  Here, an outline of the distortion masking determination process executed by the masking determination unit 813 in step 913 in FIG. 9 will be described. This distortion masking determination processing follows the flowchart shown in FIG.

In step 1001 of FIG. 10, the masking determination unit 813 obtains the change amount d (k) of the spectrum amplitude accompanying the deformation of the spectrum amplitude according to Equation 9.
d (k) = f (Sc (k), α _i ) −Sc (k) (Equation 9)
In the next step 1003, the upper _limit D _sup (k) of the aliasing distortion is obtained according to Equation 10. Here, spr (k) is a diffusion function determined by the type of conversion to the frequency domain and the shape of the window function, and * represents a convolution operation.
D _sup (k) = spr (k) * | d (k) | (Formula 10)
In the next step 1005, the masking amount T _mask (k) is obtained according to Equation 11. However, the masking pattern indicating the simultaneous masking amount is F _cb (k).
T _mask (k) = | S _t (k) | * F _cb (k) (Formula 11)
An example of the relationship between T _mask (k) and D _sup (k) is shown in FIG.

In the next step 1007, T _mask (k) and D _sup (k) are compared, and if T _mask (k) ≧ D _sup (k) for all k, it is determined that the distortion is masked, , If T _mask (k) <D _sup (k) for any k, it is determined that the distortion is not masked.

The distortion masking determination process is performed as described above. In the masking determination, a certain amount of distortion K exceeding the masking amount is allowed, and when T _mask (k) + K ≧ D _sup (k), it can be determined that the distortion is masked.

(Modification 1 of Embodiment 1)
By the way, in the parameter determination process by the parameter determination unit 811, Modification 1 in which the amount of calculation is reduced is shown below. The first modification is executed according to the flowchart of FIG. In the flowchart of FIG. 12, the same processes as those in the flowchart of FIG. 9 are given the same numbers, and steps 1213, 1215, and 1217 in FIG. 12 are different from those in the flowchart of FIG. Below, a different point is demonstrated.

In the flowchart of FIG. 9, masking is determined in step 913, whereas in the flowchart of FIG. 12, α _i is compared with a given maximum value A in step 1213. If α _i exceeds the maximum value A, α _i = A is set in step 1215, and then β _{i in} which the tone-to-noise ratio of spectrum 513 and St (k) match is searched for in step 1217.

On the other hand, when α _i does not exceed the maximum value A in step 1213 and after execution of step 1217, the process proceeds to step 917, where the determined α _i and β _i are output. The maximum value A of α _i may be a fixed value or may be determined according to the tone-to-noise ratio of the spectrum 523.

(Modification 2 of Embodiment 1)
By fixing β _i to 0 in Equation 5, it is possible to adopt a spectral shape adjustment method that is not accompanied by superimposition of noise but only by deformation of the spectrum amplitude. In this case, α _i can be obtained according to the flowchart shown in FIG.

In the flowchart of FIG. 13, the same processes as those in the flowchart of FIG. 9 are denoted by the same numbers, and steps 1307, 1309, and 1311 in FIG. 13 are different from those in the flowchart of FIG. That is, α _{i in} which the tone-to-noise ratio of the spectrum 513 and St (k) match is obtained in step 1307, and α _i and A are compared in the next step 1309. Here, if α _i exceeds A, α _i = A may be set in step 1311. Note that the processing of step 1309 may be executed by masking determination according to the flowchart of FIG.

(Modification 3 of Embodiment 1)
By fixing α _i to an appropriate value Z in Equation 5, it is possible to adopt a spectral shape adjustment method that only involves noise superposition without being accompanied by deformation of the spectral amplitude. In this case, β _i can be obtained according to the flowchart shown in FIG.

In the flowchart of FIG. 14, the same processes as those in the flowchart of FIG. 9 are given the same numbers, and steps 1405, 1409, and 1411 in FIG. 14 are different from the flowchart in FIG. 9. That is, α _i = Z is set in step 1405, and the tone-to-noise ratios of the high-frequency spectra 513 and 523 are compared in the next step 907. Here, only when the tone-to-noise ratio of the high-frequency spectrum 523 is higher than the tone-to-noise ratio of the high-frequency spectrum 513, β _{i at} which the tone-to-noise ratio of the spectrum 513 and St (k) match is set in step 1411. Can be explored. On the other hand, if the tone-to-noise ratio of the high-frequency spectrum 523 is less than or equal to the tone-to-noise ratio of the high-frequency spectrum 513, β _i = 0 is always set in step 1409.

(Modification 4 of Embodiment 1)
By the way, the noise rv in Equation 5 can be a random number having a uniform distribution or a Gaussian distribution. When rv is a random number, the superimposed noise has a flat envelope in the time domain. As another method, the random number obtained by applying LPC filtering along the frequency axis can be used as rv. By performing LPC filtering in the frequency domain, it is possible to adjust the envelope in the time domain of the noise waveform when the inverse transformation to the time domain is performed. The ability to adjust the envelope of the waveform in the time domain by LPC filtering in the frequency domain is described in, for example, the document “J. Herre et al.,“ Enhancing the Performance of Perceptual Audio Coders by Using Temporal Noise Shaping (TNS) ”, AES convention 101, paper number 4384, 1996 ".

  When the time envelope adjustment of the noise waveform by the LPC filter is performed in the speech decoding apparatus, the LPC analysis unit 805 in FIG. 8 performs LPC analysis on the high frequency spectrum of the input signal, and the parameters obtained by the analysis are used as spectral shape adjustment information. Can be included. The LPC analysis unit 805 is not an essential configuration for obtaining the effects of the present invention, and can be omitted.

In Equation 5, the deformation of the spectral amplitude is realized by a scalar function, and the deformation for a certain spectral coefficient is not affected by the value of the surrounding spectral coefficient. However, the processing method in which the deformation result depends on the surrounding spectral coefficient. You may take For example, instead of transforming the spectrum amplitude | Sc (k) | by a non-linear scalar function f (x, α _i ), f (Sc (k), α _i ) / | Sc (k) | is superimposed on each spectral coefficient. It is possible to multiply | Sc (k) | after smoothing the gain along the frequency axis.

(Configuration and operation of speech decoding apparatus in embodiment 1)
FIG. 15 is a block diagram showing the main configuration of speech decoding apparatus 1501 according to Embodiment 1 of the present invention. As shown in FIG. 15, the speech decoding apparatus 1501 divides the multiplexed bit stream into (1) information 1505 obtained by encoding the low frequency spectrum of the audio signal and (2) a spectral shape adjustment method for the high frequency spectrum. A bit stream separation unit 1503 that separates the auxiliary information 1507 to be instructed into (3) spectrum envelope information 1509, and a low-frequency decoding unit 1511 that decodes the low-frequency spectrum from information 1505 that encodes the low-frequency spectrum of the audio signal. 5 is copied by the same method as the spectrum copying unit 517 in FIG. 5, and the copied spectrum is subdivided by the same dividing method as the subband dividing unit 521 in FIG. A subband dividing unit 1515 that divides the signal into bands, and the spectrum shape of the high-frequency spectrum for each subband based on the auxiliary information 1507 Based on spectrum shape adjustment unit 1517 for adjustment, spectrum envelope adjustment unit 1519 for adjusting the envelope of the high frequency spectrum based on spectrum envelope information 1509, and combining the low frequency spectrum and the high frequency spectrum, a spectrum over the entire frequency band is obtained. A spectrum combining unit 1521 and a frequency domain inverse transform unit 1523 that converts a spectrum into a time signal and outputs the converted signal to the outside as a decoded signal are provided.

  The speech decoding apparatus 1501 has, for example, the configuration shown in FIG. 22 as the hardware configuration, similar to the speech encoding apparatus 501 described above. That is, as shown in FIG. 22, the speech decoding apparatus 1501 includes a CPU 2201 that executes an operating system, an application program, and the like, a main storage unit 2202 that includes a ROM and a RAM, an auxiliary storage unit 2203 that includes a hard disk, It includes a communication control unit 2204 that performs data communication with the outside, a display unit 2205 that includes a liquid crystal monitor, and an operation unit 2206 that performs character / number input and execution instructions. The functions of the blocks in FIG. 15 described above are realized by reading and executing predetermined software in the main storage unit 2202 under the control of the CPU 2201 shown in FIG.

  The audio decoding device 1501 performs a series of processes shown in the flowchart of FIG. 16 on the input bitstream, and outputs the decoded signal to the outside. Hereinafter, a series of processes of FIG. 16 will be described.

  In step 1601 of FIG. 16, the bit stream demultiplexing unit 1503 converts the multiplexed bit stream into (1) information 1505 obtained by encoding the low frequency spectrum of the audio signal, and (2) a spectral shape adjustment method for the high frequency spectrum. Is divided into auxiliary information 1507 for instructing and (3) spectrum envelope information 1509. It should be noted that “auxiliary information relating to the nature of the high frequency spectrum” according to the present invention corresponds to the auxiliary information 1507 and the spectrum envelope information 1509 described above.

  In the next step 1603, the low frequency decoding unit 1511 decodes the low frequency spectrum from the information 1505 obtained by encoding the low frequency spectrum of the audio signal, and supplies the decoded low frequency spectrum to the spectrum copying unit 1513 and the spectrum combining unit 1521.

  In the next step 1605, the spectrum copying unit 1513 copies the spectrum to the high band by the same method as the above-described spectrum copying unit 517 of FIG. 5 and gives it to the subband dividing unit 1515. Here, when a part of the low frequency spectrum is copied to the high frequency, the same partial spectrum as that of the spectrum copying unit 517 can be copied in accordance with a predetermined copy range. When the spectrum copying method is changed in accordance with the input signal, the copying method is determined in step 607 of FIG. 6 by the spectrum copying unit 517 of the speech coding apparatus 501 and the determined copying method is shown. It is also possible to transmit the information by superimposing it on the bit stream, and the spectrum copying unit 1513 of the audio decoding device 1501 can copy the spectrum to a high frequency according to the transmitted auxiliary information.

  In the next step 1607, the subband splitting unit 1515 splits the copied spectrum into subbands by the same splitting method as the subband splitting unit 521 in FIG.

  In the next step 1609, the spectrum shape adjustment unit 1517 adjusts the spectrum shape of the high-frequency spectrum for each subband based on the auxiliary information 1507 according to Equation 5, and adjusts the spectral shape of the adjusted high-frequency spectrum to the spectrum envelope. Part 1519.

  For the noise rv in Equation 5, the envelope in the time domain can be adjusted by LPC filtering as described above. As the LPC parameter used for filtering, the LPC parameter included in the auxiliary information 1507 (that is, the LPC parameter obtained from the analysis of the high-frequency component of the original signal before encoding) can be used. Further, by using LPC parameters obtained by LPC analysis of the decoded low-frequency spectrum, the envelope of the noise waveform in the time domain can be brought close to the low-frequency component of the decoded signal.

  In the next step 1611, the spectrum envelope adjustment unit 1519 adjusts the envelope of the high frequency spectrum given from the spectrum shape adjustment unit 1517 based on the spectrum envelope information 1509, and gives it to the spectrum combination unit 1521.

  In the next step 1613, the spectrum combiner 1521 combines the low-frequency spectrum given from the low-frequency decoder 1511 and the high-frequency spectrum given from the spectrum envelope adjuster 1519 to obtain a spectrum over the entire frequency band. The obtained spectrum over the entire frequency band is given to the frequency domain inverse transform unit 1523.

  In the next step 1615, the frequency domain inverse transform unit 1523 converts the spectrum into a time signal and outputs the converted signal to the outside as a decoded signal.

  By using the speech coding apparatus 501 and speech decoding apparatus 1501 according to the first embodiment described above, the spectral fine structure can be adjusted with high quality accompanying band expansion, and the subjective quality of the decoded speech is improved. be able to.

(Embodiment 2)
In the following second embodiment, a speech decoding apparatus having a configuration different from that of the speech decoding apparatus 1501 described in the first embodiment will be described. The speech decoding apparatus according to the second embodiment applies a bit stream including information obtained by encoding the low frequency spectrum of the audio signal, tone-to-noise ratio information of the high frequency spectrum, and envelope information of the high frequency spectrum to the flowchart of FIG. A series of processing shown in the figure is performed, and the decoded audio signal is output to the outside.

  The speech decoding apparatus according to the second embodiment has a configuration similar to that of the speech decoding apparatus according to the first embodiment. However, in the first embodiment, auxiliary information that instructs a method for adjusting a spectrum shape is included in the transmitted bitstream. On the other hand, the second embodiment is different in that the auxiliary information describing the tone-to-noise ratio of the high frequency spectrum is included in the transmitted bit stream.

  FIG. 17 is a block diagram showing the main configuration of speech decoding apparatus 1701 according to Embodiment 2 of the present invention. As shown in FIG. 17, the speech decoding apparatus 1701 converts information 1705 obtained by encoding a low frequency spectrum of a speech signal, tone-to-noise ratio information 1707 of a high frequency spectrum, and high frequency spectrum from a multiplexed bit stream. A bit stream separation unit 1703 that separates the envelope information 1709, a low-frequency decoder 1713 that decodes the low-frequency spectrum based on the information 1705 obtained by encoding the low-frequency spectrum of the audio signal, and FIG. The spectrum copying unit 1715 for copying the spectrum to the high band in the same manner as the spectrum copying unit 517, the subband dividing unit 1717 for dividing the copied high band spectrum into subbands, and the block of FIG. Spectral shape adjustment method determination unit 1711 for determining a spectral shape adjustment method according to the figure, and subband A spectrum shape adjustment unit 1719 that adjusts the shape of the divided high-frequency spectrum according to Equation 5, a spectrum envelope adjustment unit 1721 that adjusts the envelope of the high-frequency spectrum based on the envelope information 1709 of the high-frequency spectrum, A spectrum combining unit 1723 that combines the high-frequency spectrum to obtain a spectrum over the entire frequency band, and a frequency domain inverse conversion unit 1725 that converts the spectrum into a time signal and outputs the converted signal to the outside as a decoded signal are provided. .

  Also, the speech decoding apparatus 1701 has the configuration shown in FIG. 22 as the hardware configuration, similar to the speech decoding apparatus 1501 described above. The functions of the blocks in FIG. 17 described above are realized by, for example, reading and executing predetermined software in the main storage unit 2202 under the control of the CPU 2201 shown in FIG.

  The speech decoding apparatus 1701 performs a series of operations shown in the flowchart of FIG. 18 on a bitstream including information obtained by encoding the low frequency spectrum of the audio signal, tone-to-noise ratio information of the high frequency spectrum, and envelope information of the high frequency spectrum. Processing is performed, and the decoded audio signal is output to the outside. Hereinafter, a series of processes of FIG. 18 will be described.

  In step 1801 of FIG. 18, the bit stream separation unit 1703 converts the multiplexed bit stream into information 1705 obtained by encoding the low frequency spectrum of the audio signal, tone-to-noise ratio information 1707 of the high frequency spectrum, and the high frequency spectrum. Separated into envelope information 1709.

  In the next step 1803, the low frequency decoding unit 1713 decodes the low frequency spectrum based on the information 1705 obtained by encoding the low frequency spectrum of the audio signal, and supplies the decoded low frequency spectrum to the spectrum copying unit 1715 and the spectrum combining unit 1723.

  In the next step 1805, the spectrum copying unit 1715 performs the copying of the spectrum to the high band by the same method as the spectrum copying unit 517 of FIG. 5 in the first embodiment (the same method as step 607 of FIG. 6). This is given to the shape adjustment method determination unit 1711 and the subband division unit 1717.

  In the next step 1807, the subband dividing unit 1717 divides the copied high frequency spectrum into subbands and supplies the subband to the spectrum shape adjusting unit 1719.

  In the next step 1809, the spectrum shape adjustment method determination unit 1711 determines the spectrum shape adjustment method according to the block diagram shown in FIG. The processing content of FIG. 19 is almost the same as the processing content of step 611 of FIG. 6 by the spectrum shape adjustment method determining unit 525 of FIG. 5 in the first embodiment, but unlike the step 611 of FIG. Noise-to-noise ratio information 1707 is given to the spectrum shape adjustment method determination unit 1711. Therefore, in the processing of FIG. 19, spectrum normalization and tone-to-noise ratio calculation are performed only for the high frequency spectrum after copying. The spectrum shape adjustment method determination unit 1711 can be modified in the same manner as in Modification 1 to Modification 4 of Embodiment 1. By applying the same modification as that of the second modification, it is possible to adopt a spectrum shape adjustment method only by the deformation of the spectrum amplitude without superimposing noise. Further, by applying the same modification as that of the modification 3, it is possible to adopt a spectrum shape adjustment method only by superimposing noise without accompanying a modification of the spectrum amplitude.

  In the next step 1811, the spectrum shape adjustment unit 1719 adjusts the shape of the subband-divided high frequency spectrum according to Equation 5, and gives the adjustment result to the spectrum envelope adjustment unit 1721.

  In the next step 1813, the spectrum envelope adjustment unit 1721 adjusts the envelope of the high frequency spectrum based on the envelope information 1709 of the high frequency spectrum, and gives the adjustment result to the spectrum combining unit 1723.

  In the next step 1815, the spectrum combiner 1723 combines the low-frequency spectrum given from the low-frequency decoder 1713 and the high-frequency spectrum given from the spectrum envelope adjuster 1721 to obtain a spectrum over the entire frequency band. The obtained spectrum over the entire frequency band is given to the frequency domain inverse transform unit 1725.

  In the next step 1817, the frequency domain inverse transform unit 1725 converts the spectrum into a time signal and outputs the converted signal to the outside as a decoded signal.

  Also in the second embodiment described above, as in the first embodiment, the spectral fine structure accompanying the band expansion can be adjusted with high quality, and the subjective quality of the decoded speech can be improved.

(Embodiment 3)
FIG. 20 is a block diagram showing the main configuration of the voice band extending apparatus 2001 according to Embodiment 3 of the present invention. The voice band expansion device 2001 performs a frequency domain conversion unit 2003 that obtains a low-frequency spectrum by performing a conversion process on an input signal to a frequency domain, and a frequency domain conversion unit in the same manner as the spectrum copying unit 517 in FIG. A spectrum copying unit 2005 for copying a low-frequency spectrum given from 2003 to a high frequency; a sub-band dividing unit 2011 for dividing the high-frequency spectrum given from the spectrum copying unit 2005 into a plurality of sub-bands; Tone-to-noise ratio estimation unit 2007 for estimating the tone-to-noise ratio in each subband of the spectrum from the low-frequency spectrum, spectral envelope estimation unit 2009 for estimating the spectral envelope of the high-frequency spectrum from the low-frequency spectrum, and subband dividing unit Subband-divided high-frequency spectrum given by 2011 and tone-to-noise A spectrum shape adjustment method determination unit 2015 that determines a spectrum shape adjustment method from the tone-to-noise ratio given from the estimation unit 2007 in the same manner as the spectrum shape adjustment method determination unit 525 of FIG. A spectrum shape adjustment unit 2013 that adjusts the shape of the high frequency spectrum in the same manner as the spectrum shape adjustment unit 1517 of FIG. The spectrum envelope adjustment unit 2017 that adjusts the envelope of the high-frequency spectrum using the spectrum envelope information given from the above, the high-frequency spectrum given from the spectrum envelope adjustment unit 2017, and the low-frequency spectrum given from the frequency domain conversion unit 2003 Spectral coupling unit 2019 coupling Converts the spectrum after binding to a time signal, the high-frequency range and a frequency domain inverse transform unit 2021 outputs the converted signal as a decompressed wideband speech signal to the outside.

  The speech band expansion device 2001 has a hardware configuration as shown in FIG. 22, for example, as with the speech encoding device 501 and speech decoding device 1501 described above. The functions of the blocks in FIG. 20 described above are realized by reading and executing predetermined software in the main storage unit 2202 under the control of the CPU 2201 shown in FIG.

  The voice band extending apparatus 2001 performs a series of processes shown in the flowchart of FIG. 21 on the band-limited input signal including only the low frequency component of the voice, and outputs a wideband voice signal whose high frequency component is restored to the outside. . Hereinafter, a series of processes of FIG. 21 will be described.

  In step 2101 of FIG. 21, the frequency domain conversion unit 2003 performs conversion to the frequency domain on the input signal to obtain a low frequency spectrum. As a conversion method, discrete Fourier transform (DFT), discrete cosine transform (DCT), modified discrete cosine transform (MDCT), cosine modulation filter bank, or the like can be used. If the input speech is given as a signal in the frequency domain instead of the time domain, the step 2101 and the frequency domain transform unit 2003 can be omitted.

  In the next step 2103, the spectrum copying unit 2005 converts the low band spectrum given from the frequency domain conversion unit 2003 to the high band in the same manner as the step 607 in FIG. 6 by the above-described spectrum copying unit 517 in FIG. Copying is performed and given to the subband dividing unit 2011.

  In the next step 2105, the subband dividing unit 2011 divides the high-frequency spectrum given from the spectrum copying unit 2005 into a plurality of subbands, and gives it to the spectrum shape adjusting method determining unit 2015 and the spectrum shape adjusting unit 2013. Note that the subband splitting unit 2011 in FIG. 20 or step 2105 in FIG. 21 is not essential, and the entire high-frequency spectrum that is not split may be given to the spectrum shape adjusting unit 2013.

  In the next step 2107, the tone-to-noise ratio estimation unit 2007 estimates the tone-to-noise ratio in each subband of the high-frequency spectrum from the low-frequency spectrum, and provides it to the spectral shape adjustment method determination unit 2015. For this estimation, a vector quantizer, a mixed Gaussian model (GMM), a hidden Markov model (HMM), or the like previously learned using a spectrum of wideband speech can be used.

  In the next step 2109, the spectrum envelope estimation unit 2009 estimates the spectrum envelope of the high frequency spectrum from the low frequency spectrum, and provides it to the spectrum envelope adjustment unit 2017. As in the estimation of the tone-to-noise ratio, a vector quantizer, a mixed Gaussian model (GMM), a hidden Markov model (HMM), or the like previously learned using a wideband speech spectrum can be used for the estimation here. .

  In the next step 2111, the spectrum shape adjustment method determination unit 2015 uses the subband-divided high-frequency spectrum given from the subband division unit 2011 and the tone-to-noise ratio given from the tone-to-noise ratio estimation unit 2007. A spectrum shape adjustment method is determined by the same method as the spectrum shape adjustment method determination unit 525 of FIG. 5 described above, and is given to the spectrum shape adjustment unit 2013. The spectrum shape adjustment method determination unit 2015 can be modified in the same manner as in Modification 1 to Modification 4 of Embodiment 1. By applying the same modification as that of the second modification, it is possible to adopt a spectrum shape adjustment method only by the deformation of the spectrum amplitude without superimposing noise. Further, by applying the same modification as that of the modification 3, it is possible to adopt a spectrum shape adjustment method only by superimposing noise without accompanying a modification of the spectrum amplitude.

  In the next step 2113, the spectrum shape adjustment unit 2013 uses the spectrum shape adjustment method given from the spectrum shape adjustment method determination unit 2015 in the same manner as the spectrum shape adjustment unit 1517 of FIG. Adjust the shape.

  In the next step 2115, the spectrum envelope adjustment unit 2017 adjusts the envelope of the high frequency spectrum using the spectrum envelope information given from the spectrum envelope estimation unit 2009, and gives it to the spectrum combining unit 2019.

  In the next step 2117, the spectrum combining unit 2019 combines the high-frequency spectrum given from the spectrum envelope adjustment unit 2017 and the low-frequency spectrum given from the frequency domain conversion unit 2003, and the combined spectrum is converted into the frequency domain. This is given to the inverse transform unit 2021.

  In the next step 2119, the frequency domain inverse transform unit 2021 converts the combined spectrum into a time signal, and outputs the converted signal to the outside as a wideband audio signal whose high frequency is restored.

(Modification of Embodiment 3)
In the voice band extending apparatus 2001 shown above, the case where the input voice is given as a signal in the time domain is exemplified, but the input signal may be a frequency domain signal as described above. Further, the signal in the frequency domain may be a signal obtained in the middle of the decoding process of speech encoded by a speech encoding method based on transform encoding. In a speech decoding apparatus that performs speech decoding processing of speech encoded by a speech encoding method based on transform encoding, the signal is inversely transformed from the frequency domain to the time domain at the final stage of the decoding processing. By supplying the signal in the previous frequency domain to the voice band extending apparatus 2001, it is possible to obtain wideband decoded speech whose high band is restored as the output of the voice band extending apparatus 2001.

  Also in the third embodiment described above, as in the first embodiment, the spectral fine structure can be adjusted with a high quality as a result of band expansion, and the subjective quality of the restored speech can be improved.

It is a figure which shows the example of the spectrum of an audio | voice signal. It is a figure which shows the example of the result of having copied the spectrum from the low region to the high region, and having adjusted the spectrum envelope. It is a figure which shows the example of the spectrum which divided | segmented into the subband and adjusted the fine structure. It is a schematic diagram which shows distortion generation accompanying compression of a spectrum amplitude. 1 is a block diagram illustrating a main configuration of a speech encoding apparatus according to Embodiment 1. FIG. 4 is a flowchart showing main steps of the speech encoding method according to Embodiment 1. It is a schematic diagram which shows the various examples of the copying method from the low region of a spectrum to a high region. It is a block diagram which shows the main structures of the spectrum shape adjustment method determination part 525. 10 is a flowchart showing main steps of processing by a parameter determination unit 811. It is a flowchart which shows the detail of the judgment process of step 913 of FIG. It is a schematic diagram which shows the masking of the distortion accompanying a deformation | transformation of a spectrum amplitude. 10 is a flowchart showing another mode of processing by the parameter determination unit 811. It is a flowchart which shows another aspect of the process by the parameter determination part 811 when (alpha) _i is fixed. It is a flowchart which shows another aspect of the process by the parameter determination part 811 when (beta) _i is made fixed. It is a block diagram which shows the main structures of the speech decoding apparatus in Embodiment 1. 6 is a flowchart showing main steps of the speech decoding method according to Embodiment 1. It is a block diagram which shows the main structures of the speech decoding apparatus in Embodiment 2. 10 is a flowchart showing main steps of the speech decoding method according to Embodiment 2. It is a block diagram which shows the main structures of the spectrum shape adjustment method determination part 1711. FIG. It is a block diagram which shows the main structures of the audio | voice band expansion apparatus in Embodiment 3. FIG. It is a block diagram which shows the main steps of the audio | voice band expansion method in Embodiment 3. FIG. It is a figure which shows the hardware structural example of an audio | voice encoding apparatus, an audio | voice decoding apparatus, and an audio | voice band expansion apparatus.

Explanation of symbols

501: Speech coding apparatus, 503: Frequency domain transform unit, 505 ... Low frequency spectrum, 507 ... High frequency spectrum, 509 ... Low frequency encoding unit, 511 ... Subband division unit, 513 ... High frequency spectrum, 515 ... Low Area decoding section, 517 ... Spectrum copying section, 519 ... High band spectrum, 521 ... Subband splitting section, 523 ... High band spectrum, 525 ... Spectral shape adjustment method determining section, 527 ... Spectrum envelope calculation section, 529 ... Bit stream Multiplexer, 801, 803 ... spectrum normalization unit, 805 ... LPC analysis unit, 807,809 ... tone-to-noise ratio calculation unit, 811 ... parameter determination unit, 813 ... masking determination unit, 1501 ... voice decoding device, 1503 ... bitstream separation unit, 1505 ... encoded information, 1507 ... auxiliary information, 1509 ... spect Envelope information, 1511 ... Low band decoding unit, 1513 ... Spectrum copying unit, 1515 ... Subband dividing unit, 1517 ... Spectrum shape adjusting unit, 1519 ... Spectrum envelope adjusting unit, 1521 ... Spectrum combining unit, 1523 ... Frequency domain inverse transform , 1701... Speech decoding device, 1703... Bitstream separation unit, 1705... Encoded information, 1707... Tone-to-noise ratio information, 1709... Envelope information, 1711. 1715 ... Spectrum copying unit, 1717 ... Subband dividing unit, 1719 ... Spectral shape adjusting unit, 1721 ... Spectral envelope adjusting unit, 1723 ... Spectrum combining unit, 1725 ... Frequency domain inverse converting unit, 2001 ... Audio band expanding device , 2003 ... frequency domain transform unit, 2005 ... spectrum Copy unit 2007 ... tone-to-noise ratio estimation unit 2009 ... spectrum envelope estimation unit 2011 ... subband division unit 2013 ... spectral shape adjustment unit 2015 ... spectral shape adjustment method determination unit 2017 ... spectrum envelope adjustment unit 2019 ... spectrum combining unit, 2021 ... frequency domain inverse transformation unit, 2201 ... CPU, 2202 ... main storage unit, 2203 ... auxiliary storage unit, 2204 ... communication control unit, 2205 ... display unit, 2206 ... operation unit.

Claims (19)

A speech signal including a frequency component corresponding to both the low-frequency spectrum and the high-frequency spectrum is decoded from a bitstream including information obtained by encoding the low-frequency spectrum of the audio signal and auxiliary information regarding the nature of the high-frequency spectrum. A speech decoding device comprising:
Means for separating the bitstream into information encoded from a low frequency spectrum of an audio signal and auxiliary information relating to the nature of the high frequency spectrum;
Means for decoding a low-frequency spectrum from information obtained by encoding the low-frequency spectrum of the audio signal;
Means for obtaining a first high-frequency spectrum by copying a part or the whole of the low-frequency spectrum obtained by decoding to a high frequency;
Spectrum shape adjusting means for obtaining a second high-frequency spectrum by adjusting the shape of the first high-frequency spectrum using auxiliary information on the nature of the high-frequency spectrum;
Means for decoding an audio signal using the low frequency spectrum and the second high frequency spectrum;
Spectral shape adjustment method determination for determining how to adjust the shape used by the spectral shape adjusting means or the amount of noise to be superimposed based on auxiliary information regarding the nature of the high frequency spectrum and the first high frequency spectrum Means,
With
The spectrum shape adjustment method determining means uses the distortion generated in the output signal due to the adjustment of the shape of the spectrum and the masking amount due to the high-frequency spectrum after the shape adjustment to superimpose or superimpose the shape. A speech decoding apparatus characterized by determining an amount of noise . The spectrum shape adjustment method determining means further compares the tone-to-noise ratio of the high frequency spectrum of the speech signal with the tone-to-noise ratio of the first high frequency spectrum.
  The speech decoding apparatus according to claim 1. The spectrum shape adjusting means obtains the second high frequency spectrum by superimposing noise in addition to adjusting the shape of the first high frequency spectrum.
The speech decoding apparatus according to claim 1 or 2 , characterized in that The spectrum shape adjusting means divides the first high-frequency spectrum into a plurality of subbands, and adjusts the shape of the partial spectrum for each of the partial spectra belonging to each of the divided subbands. The speech decoding apparatus according to any one of claims 1 to 3 . The auxiliary information related to the nature of the high frequency spectrum includes information related to the envelope in the time domain of the high frequency components of the speech,
The speech decoding apparatus according to claim 1, wherein the spectrum shape adjusting unit adjusts an envelope in a time domain of superimposed noise. A speech encoding device that generates a bitstream compatible with a speech decoding device that decodes a high-frequency spectrum based on a copy of a decoding result of a low-frequency spectrum to a high frequency,
Means for separating the low frequency spectrum after converting the audio signal into the frequency domain, and then separating a part or all of the remaining high frequency spectrum as a first high frequency spectrum;
Means for encoding the low frequency spectrum;
Means for obtaining a second high frequency spectrum by copying a part or the whole of the low frequency spectrum to a high frequency;
A spectral shape adjustment method determining means for determining how to adjust the shape of the high frequency spectrum in the speech decoding apparatus based on the properties of the first high frequency spectrum and the second high frequency spectrum;
Means for generating a bit stream in which the encoded low frequency spectrum and information on how to adjust the shape of the determined high frequency spectrum are multiplexed ;
Equipped with a,
The spectrum shape adjustment method determining means uses the distortion generated in the output signal due to the adjustment of the shape of the spectrum and the masking amount by the high frequency spectrum after the spectrum shape adjustment, to form the shape of the high frequency spectrum. A speech coding apparatus that determines how to adjust the amount of noise or the amount of noise to be superimposed . The spectrum shape adjustment method determining means compares the tone-to-noise ratio of the first high-frequency spectrum with the tone-to-noise ratio of the second high-frequency spectrum.
  The speech encoding apparatus according to claim 6. The spectrum shape adjustment method determining means, in addition to the method of adjusting the shape of the high frequency spectrum based on the properties of the first high frequency spectrum and the second high frequency spectrum, the high frequency in the speech decoding apparatus. The speech coding apparatus according to claim 6 or 7 , wherein the amount of noise to be superimposed on the high frequency spectrum is determined for adjusting the shape of the high frequency spectrum. The spectrum shape adjustment method determining means divides the first high frequency spectrum into a plurality of subbands, and determines how to adjust the shape of the partial spectrum for each partial spectrum belonging to each of the divided subbands. The speech coding apparatus according to any one of claims 6 to 8, wherein An audio band extending apparatus for restoring an audio signal including a high frequency component not included in the low frequency spectrum from a low frequency spectrum including only a low frequency component of the audio,
Means for obtaining a first high frequency spectrum by copying a part or the whole of the low frequency spectrum to a high frequency;
A spectrum shape adjusting means for obtaining a second high frequency spectrum by adjusting the shape of the first high frequency spectrum;
Means for reconstructing an audio signal including a high frequency component using the low frequency spectrum and the second high frequency spectrum;
Spectrum shape adjustment method determining means for determining how to adjust the shape of the high frequency spectrum in the speech decoding apparatus based on the property of the first high frequency spectrum;
Equipped with a,
The spectrum shape adjustment method determining means uses the distortion generated in the output signal due to the adjustment of the shape of the spectrum and the masking amount by the high frequency spectrum after the spectrum shape adjustment, to form the shape of the high frequency spectrum. A voice band extending apparatus for determining an adjustment method or an amount of superimposed noise . The spectrum shape adjustment method determining means compares the tone-to-noise ratio of the first high-frequency spectrum with the tone-to-noise ratio of the high-frequency spectrum estimated from the low-frequency spectrum.
  The voice band extending apparatus according to claim 10. The spectrum shape adjusting means obtains the second high frequency spectrum by superimposing noise in addition to adjusting the shape of the first high frequency spectrum.
The voice band extending apparatus according to claim 10 or 11 , wherein The spectrum shape adjusting means divides the first high-frequency spectrum into a plurality of subbands, and adjusts the shape of the partial spectrum for each of the partial spectra belonging to each of the divided subbands. The voice band extending apparatus according to any one of claims 10 to 12 . A speech signal including a frequency component corresponding to both the low-frequency spectrum and the high-frequency spectrum is decoded from a bitstream including information obtained by encoding the low-frequency spectrum of the audio signal and auxiliary information regarding the nature of the high-frequency spectrum. A speech decoding method executed by a speech decoding device,
Separating the bitstream into information encoded from the low frequency spectrum of the audio signal and auxiliary information relating to the nature of the high frequency spectrum;
Decoding the low-frequency spectrum from information obtained by encoding the low-frequency spectrum of the audio signal;
Obtaining a first high-frequency spectrum by copying a part or the whole of the low-frequency spectrum obtained by decoding to a high frequency;
A spectral shape adjustment step of obtaining a second high-frequency spectrum by adjusting the shape of the first high-frequency spectrum using auxiliary information on the nature of the high-frequency spectrum;
Decoding an audio signal using the low frequency spectrum and the second high frequency spectrum;
Spectral shape adjustment method determination for determining the shape adjustment method or the amount of noise to be superimposed used by the spectral shape adjustment step based on the auxiliary information on the nature of the high frequency spectrum and the first high frequency spectrum Steps,
With
The spectrum shape adjustment method determining step uses the distortion generated in the output signal due to the adjustment of the shape of the spectrum and the masking amount by the high-frequency spectrum after the shape adjustment, or the method of shape adjustment or superimposition. A speech decoding method characterized by determining an amount of noise . In the spectral shape adjustment step, the speech decoding apparatus obtains the second high frequency spectrum by superimposing noise in addition to the adjustment of the shape of the first high frequency spectrum.
The speech decoding method according to claim 14 , wherein: A speech coding method that is executed by a speech coding apparatus that generates a bitstream suitable for a speech decoding apparatus that decodes a high-frequency spectrum based on copying of a decoding result of a low-frequency spectrum to a high frequency, ,
Separating the audio signal into the frequency domain, separating the low-frequency spectrum, and then separating part or all of the remaining high-frequency spectrum as a first high-frequency spectrum;
Encoding the low-pass spectrum;
Obtaining a second high frequency spectrum by copying a part or the whole of the low frequency spectrum to a high frequency;
A spectral shape adjustment method determining step for determining how to adjust the shape of the high frequency spectrum in the speech decoding apparatus based on the properties of the first high frequency spectrum and the second high frequency spectrum;
Generating a bitstream in which the encoded low-frequency spectrum and information on how to adjust the shape of the determined high-frequency spectrum are multiplexed ;
Equipped with a,
The spectrum shape adjustment method determination step uses the distortion generated in the output signal due to the adjustment of the shape of the spectrum and the masking amount by the high frequency spectrum after the spectrum shape adjustment, and the shape of the high frequency spectrum A speech encoding method characterized by determining an adjustment method or an amount of noise to be superimposed . In the spectral shape adjustment method determining step, the speech encoding apparatus, in addition to the method of adjusting the shape of the high frequency spectrum based on the properties of the first high frequency spectrum and the second high frequency spectrum, The speech coding method according to claim 16 , wherein the amount of noise to be superimposed on the high frequency spectrum is determined for adjusting the shape of the high frequency spectrum in the speech decoding apparatus. A voice band extending method executed by a voice band extending apparatus for restoring a voice signal including a high frequency component not included in the low frequency spectrum from a low frequency spectrum including only a low frequency component of the voice,
Obtaining a first high frequency spectrum by copying a part or the whole of the low frequency spectrum to a high frequency;
A spectral shape adjustment step for obtaining a second high frequency spectrum by adjusting the shape of the first high frequency spectrum;
Reconstructing an audio signal including a high frequency component using the low frequency spectrum and the second high frequency spectrum;
A spectral shape adjustment method determining step for determining how to adjust the shape of the high frequency spectrum in the speech decoding apparatus based on the property of the first high frequency spectrum;
Equipped with a,
The spectrum shape adjustment method determination step uses the distortion generated in the output signal due to the adjustment of the shape of the spectrum and the masking amount by the high frequency spectrum after the spectrum shape adjustment, and the shape of the high frequency spectrum A method of expanding a voice band , characterized by determining a method of adjusting the amount of noise or an amount of noise to be superimposed . In the spectral shape adjustment step, the speech decoding apparatus obtains the second high frequency spectrum by superimposing noise in addition to the adjustment of the shape of the first high frequency spectrum.
The method of extending a voice band according to claim 18 .

Images