JP2018124304A - Speech coding apparatus, speech decoding apparatus, speech coding method, speech decoding method, program, and recording medium - Google Patents

Abstract

The present invention improves the quality of wideband speech. A band division filter (12) divides an input sound into a low frequency sound and a high frequency sound. The high frequency audio encoding unit 14 encodes the high frequency audio based on the decoded low frequency audio to generate a high frequency code. The low frequency speech encoding unit 13 encodes the low frequency speech and generates a low frequency code in which the high frequency code is embedded. However, the low-frequency code is assumed to be compatible with the ITU-T G.726 system. The low frequency audio decoding unit 16 decodes the low frequency code and generates decoded low frequency audio. The code sending unit 17 outputs the low frequency code as a voice code. [Selection] Figure 3

Description

  The present invention relates to voice / acoustic signal (hereinafter, also simply referred to as voice) communication using a digital communication network, and in particular, a voice encoding technique for encoding input voice and a voice decoding technique for generating voice from a received voice code. About.

  The frequency band of voice that can be transmitted by a conventional telephone system represented by an analog telephone is about 300 Hz to 3.4 kHz. This is determined by the International Telecommunication Union Telecommunication Standardization Sector (ITU-T) in consideration of the balance between the voice quality necessary to convey a message and the amount of information necessary for transmission. By being adopted in. In general, a voice whose upper limit of the frequency band is 4 kHz or less is called a narrowband signal (or also called narrowband voice or telephone voice), and a voice exceeding 4 kHz and about 7 kHz is called a wideband signal (or wideband voice). When voice is expressed by a pulse code modulation (PCM) method of a digital signal, it is desirable to sample a narrowband signal at 8 kHz and a wideband signal at 16 kHz according to the sampling theorem. For these reasons, a signal sampled at 8 kHz is sometimes called a narrowband signal, and a signal sampled at 16 kHz is sometimes called a wideband signal.

  With the recent development of acoustic technology and advancement of digital signal processing technology, the sound quality of equipment used in daily life has improved. In such a situation, there is a voice that calls for wider bandwidth in the voice of a telephone.

  In order to efficiently transmit a voice signal using a digital communication network, a voice coding method is used. There are international standard schemes such as ITU-T G.711 and ITU-T G.726 for speech coding for narrowband signals (also called narrowband speech coding). In addition, there are international standard systems such as ITU-T G.711.1 and ITU-T G.722 for speech coding for wideband signals (also called wideband speech coding). A terminal (hereinafter referred to as a terminal) that performs voice communication includes an encoding device and a decoding device that support at least one of the voice encoding schemes. When the terminal supports a plurality of audio encoding systems, the encoding system used for the communication is switched at the start of communication. Conventionally, a call control protocol called SIP or H.323 (also called signaling) is used for switching the coding method, and a coding method that is commonly supported by communicating terminals is determined in a predetermined priority order. Had chosen based on. For example, if both terminals support G.711.1 and G.711, communication is performed using G.711.1, which is wideband speech coding, one is G.711.1 and G.711, the other is G.722 and G. In the case of corresponding to .711, both terminals support wideband audio, but G.711 is used for audio encoding, and communication is performed using narrowband audio.

  The reason why the coding method is switched using the call control protocol at the start of communication is because the coding methods are not compatible, but the switching of the coding method based on the call control protocol establishes voice communication between terminals. Is complicated, causing connection problems. In addition, IP telephones that use the Internet as a communication network can be switched relatively freely using the call control protocol. However, the existing intra-company communication networks and interconnection networks between communication carriers can be used. In the voice communication that goes through, if there is equipment that allows only G.711 on the communication path, there is a problem that only G.711 can be used even if the terminal supports a plurality of encoding methods.

  With respect to this problem, Patent Document 1 describes that wideband speech coding having complete compatibility with G.711 can be realized. A wideband speech coding system that is fully compatible with G.711 greatly simplifies the procedure for switching between coding systems, and even if there is equipment that only allows G.711 on the communication path, Can be passed.

  Referring to FIG. 1, a speech encoding device described in Patent Document 1 is shown. Speech input to the speech coding apparatus is accumulated in the input buffer 81, divided into frames having a length of about 10 milliseconds to 20 milliseconds, and sent to the band division filter 82. The band division filter 82 divides the input sound into low frequency sound and high frequency sound. The low frequency audio is sent to the low frequency audio encoding unit 83, and the high frequency audio is sent to the high frequency audio encoding unit 84. The high frequency audio encoding unit 84 encodes the high frequency audio to generate a high frequency code, and sends the high frequency code to the low frequency audio encoding unit 83. The low frequency speech coding unit 83 receives the low frequency speech and the high frequency code, and embeds the high frequency code as a 1 or 0 bit string in the LSB (Least Significant Bit) or MSB (Most Significant Bit) of the G.711 code. A low frequency code is generated, and the low frequency code is sent to the packet construction unit 85. The packet configuration unit 85 receives the low frequency code from the low frequency audio encoding unit 83 and configures a packet using the low frequency code. The packet sending unit 86 receives the packet information created by the packet construction unit 85 and sends it as a voice packet to the packet communication network.

  With reference to FIG. 2, the speech decoding apparatus described in Patent Document 1 is shown. The voice packet output from the voice encoding device is received by the packet receiving unit 91 of the voice decoding device and accumulated in the reception buffer 92. The audio packet output from the reception buffer 92 is decoded by the low frequency audio decoding unit 94. The high frequency code extraction unit 95 extracts a high frequency code from the speech code. The high frequency audio decoding unit 96 decodes a high frequency audio component from the extracted high frequency code. The checksum detector 93 determines whether or not the high frequency code is embedded in the LSB or MSB of the low frequency code for the audio code output from the reception buffer 92. Is set to the high frequency audio decoding unit 96 side, and the high frequency audio component is sent to the band synthesis filter 98. As a result of the determination by the checksum detection unit 93, when it is determined that the high frequency code is not embedded in the LSB or MSB of the low frequency code, the switch 97 is set to the non-high frequency side. That is, no high frequency audio component is generated. The band synthesizing filter 98 synthesizes the output of the low frequency audio decoding unit 94 and the output of the high frequency audio decoding unit 96 into a wideband audio signal and outputs it.

Japanese Patent No. 4758687

  However, Patent Document 1 only describes a part of the configuration for realizing wideband speech coding that is completely compatible with G.711. Specifically, it is described that the high frequency speech encoding unit 84 simply encodes the high frequency speech, and the high frequency speech decoding unit 96 simply decodes the high frequency speech component from the high frequency code. Only that is described. In order to realize wideband speech coding fully compatible with G.711, the quality of the wideband speech reproduced from the decoding device is sufficiently good, and at least the narrowband decoded by the G.711 method It is necessary to be able to reproduce wide-band audio of higher quality than audio. In addition, it is expected that wideband speech coding having full compatibility with G.726 is realized by the same concept, but Patent Document 1 does not specifically describe the compatibility with G.726. .

  In view of the above points, an object of the present invention is to provide a speech coding technique capable of improving the quality of reproduced wideband speech in speech communication of wideband speech.

  In order to solve the above-described problem, the speech coding apparatus according to the first aspect of the present invention is based on a band dividing unit that divides an input speech into a low-frequency speech and a high-frequency speech, and a decoded low-frequency speech. A high-frequency speech encoding unit that encodes high-frequency speech to generate a high-frequency code, a low-frequency speech encoding unit that encodes low-frequency speech and generates a low-frequency code embedded with the high-frequency code, and a low frequency Includes a low-frequency audio decoding unit that decodes a code to generate decoded low-frequency audio, and a code sending unit that outputs the low-frequency code as an audio code, and the low-frequency code is compatible with the ITU-T G.726 system It has sex.

  The speech decoding apparatus according to the second aspect of the present invention includes a code receiving unit that receives the speech code output from the speech encoding apparatus according to the first aspect, and a low frequency that decodes the speech code and generates decoded low frequency speech. A speech decoding unit, a high frequency code extracting unit that extracts a high frequency code embedded in the speech code, and a high frequency speech decoding unit that generates a decoded high frequency speech by decoding the high frequency code based on the decoded low frequency speech And a band synthesizing unit that synthesizes the decoded low frequency sound and the decoded high frequency sound and outputs the decoded sound.

  According to the present invention, high-frequency speech can be encoded with a small number of bits while maintaining as much as possible the information necessary for reproducing wideband speech in speech encoding. In speech decoding, high-quality wideband speech can be reproduced by generating high-frequency speech that is audibly low in quality degradation. That is, the quality of the reproduced wideband voice can be improved in the wideband voice communication.

FIG. 1 is a diagram illustrating a functional configuration of a conventional speech coding apparatus. FIG. 2 is a diagram illustrating a functional configuration of a conventional speech decoding apparatus. FIG. 3 is a diagram illustrating a functional configuration of the speech encoding apparatus according to the embodiment. FIG. 4 is a diagram illustrating a functional configuration of the speech decoding apparatus according to the embodiment. FIG. 5 is a diagram illustrating a processing procedure of the speech encoding method according to the embodiment. FIG. 6 is a diagram illustrating a processing procedure of the speech decoding method according to the embodiment. FIG. 7 is a diagram illustrating a functional configuration of the high frequency speech encoding unit of the embodiment. FIG. 8 is a diagram illustrating a functional configuration of the coefficient encoding unit according to the embodiment. FIG. 9 is a diagram illustrating a functional configuration of the high frequency speech decoding unit according to the embodiment. FIG. 10 is a diagram illustrating a functional configuration of the coefficient decoding unit according to the embodiment. FIG. 11 is a diagram illustrating a functional configuration of the low frequency speech encoding unit of the embodiment. FIG. 12 is a diagram illustrating a functional configuration of a low-frequency speech encoding unit according to Modification A. As illustrated in FIG. FIG. 13 is a diagram illustrating a functional configuration of the low-frequency speech encoding unit according to Modification A. As illustrated in FIG. FIG. 14 is a diagram illustrating a functional configuration of the low-frequency speech decoding unit according to the embodiment. FIG. 15 is a diagram illustrating a functional configuration of the low frequency speech decoding unit according to Modification B. As illustrated in FIG. FIG. 16 is a diagram illustrating a functional configuration of the low frequency speech decoding unit according to Modification B. As illustrated in FIG. FIG. 17 is a diagram illustrating a functional configuration of the high frequency speech encoding unit according to Modification C. As illustrated in FIG. FIG. 18 is a diagram for explaining the effect of the modification C. FIG. 19 is a diagram illustrating a functional configuration of the designated band signal extraction unit of the modification C. As illustrated in FIG.

  Hereinafter, embodiments of the present invention will be described in detail. In addition, the same number is attached | subjected to the component which has the same function in drawing, and duplication description is abbreviate | omitted.

The symbol “ ⁻ ” used in the sentence should be described immediately above the character immediately before, but is described immediately after the character due to restrictions on text notation. In the mathematical expression, these symbols are described in their original positions, that is, directly above the characters.

  In an embodiment of the present invention, a speech encoding device that encodes input speech into speech codes and outputs the speech, and a speech decoding device that decodes speech codes output by the speech encoding device and outputs decoded speech are described. . When the speech encoding apparatus converts the input speech for one frame into a speech code and outputs it, it processes the input speech of the next frame and repeats this in the time period of the frame. When the speech decoding apparatus processes the speech code for one frame and outputs the decoded speech, it performs the speech code processing of the next frame and repeats this in the time period of the frame.

  As shown in FIG. 3, the speech encoding apparatus according to the embodiment includes an input buffer 11, a band division filter (also referred to as a band division unit) 12, a low frequency speech encoding unit 13, a high frequency speech encoding unit 14, and a delay unit. 15, a low frequency speech decoding unit 16 and a code sending unit 17. The speech encoding apparatus according to the embodiment is realized by the processing of each step described later by the speech encoding device.

  As shown in FIG. 4, the speech decoding apparatus according to the embodiment includes a code receiving unit 21, a low frequency speech decoding unit 22, a high frequency code extraction unit 23, a delay unit 24, a high frequency speech decoding unit 25, and a band synthesis filter ( 26). The speech decoding apparatus according to the embodiment is realized by the processing of each step described later by the speech decoding device.

  A speech encoding device and a speech decoding device, for example, have a special program read into a known or dedicated computer having a central processing unit (CPU), a main storage device (RAM), and the like. It is a special device constructed. Each device executes each process under the control of the central processing unit, for example. Data input to each device and data obtained by each process are stored in, for example, a main storage device, and the data stored in the main storage device is read out as necessary and used for other processing. . Further, at least a part of each processing unit of each device may be configured by hardware such as an integrated circuit.

  With reference to FIG. 5, the processing procedure of the speech coding method of the embodiment will be described.

In step S11, speech is input to the speech coding apparatus. The input sound x is stored in the input buffer 11, divided into frames having a length of about 10 milliseconds to 20 milliseconds, and sent to the band division filter 12. The input sound x is wideband sound, and the sampling frequency is 16 kHz. The input sound x is divided by the band division filter 12 into a low frequency sound x _L and a high frequency sound x _H having a sampling frequency of 8 kHz. The low frequency speech x _L is sent to the low frequency speech encoding unit 13, and the high frequency speech x _H is sent to the high frequency speech encoding unit 14. The band division filter 12 may be a quadrature mirror filter (QMF) used in G.711.1 or G.722. Alternatively, using an appropriate low-pass filter and high-pass filter, the low-pass sound x _L is obtained by applying a low-pass filter to the input sound x to reduce the number of samples to 1/2, and the input sound x is subjected to a high-pass filter. the thinned signal to 2 number of samples may be the high-band speech x _H.

In step S12, the high frequency audio encoding unit 14 encodes the high frequency audio x _H using the decoded low frequency audio x ⁻ _L received from the low frequency audio decoding unit 16 described later, and generates the high frequency code c _H. This is sent to the delay unit 15. Details of the processing of the high frequency speech encoding unit 14 will be described later. The delay unit 15 has a memory for storing the high frequency code c _H for one frame, and sends the high frequency code one frame before to the low frequency speech encoding unit 13 and stores the received high frequency code. As will be described later, since the delay unit 15 can be omitted, the high-frequency code that is the output of the high-frequency speech encoding unit 14 and the high-frequency code that is one frame before that is the output of the delay unit 15 are Without distinction, it is simply referred to as a high-frequency code c _H.

In step S13, the low frequency speech encoding unit 13 can use the same configuration as the low frequency speech encoding unit 83 provided in the conventional speech encoding device. That is, the low frequency speech x _L and the high frequency code c _H are received, and the low frequency code c _{L in} which the high frequency code is embedded as a 1 or 0 bit string in the LSB or MSB of the G.711 code is output. The output of the low frequency speech encoding unit 13 is sent to the code sending unit 17 and also to the low frequency speech decoding unit 16.

In step S <b> 14, the low frequency speech decoding unit 16 decodes the low frequency code c _L received from the low frequency speech encoding unit 13 and sends the decoded low frequency speech x ⁻ _L to the high frequency speech encoding unit 14. The low frequency speech decoding unit 16 can use the same configuration as the low frequency speech decoding unit 94 provided in the conventional speech decoding device.

In step S15, the code transmission unit 17 transmits the low frequency code c _L received from the low frequency audio encoding unit 13 to the communication network as a audio code.

The speech code c _L sent from the speech encoding device has complete bit compatibility with G.711, and when the speech decoding device corresponding to the conventional G.711 scheme receives the speech code c _L , G.711 can play narrowband speech by decoding scheme, if the speech decoding apparatus receives a voice code c _L of the present invention, it is possible to reproduce a wideband speech by the speech decoding method described later. Moreover, the voice code c _L can pass through the communication network corresponding to only the existing G.711.

  With reference to FIG. 6, the processing procedure of the speech decoding method of the embodiment will be described.

In step S <b> 21, the code receiving unit 21 receives the speech code c _L from the communication network and sends it to the low frequency speech decoding unit 22 and the high frequency code extracting unit 23.

In step S <b> 22, the low frequency audio decoding unit 22 decodes the audio code c _L by the G.711 method, and sends the decoded low frequency audio x ⁻ _L to the delay unit 24. The delay unit 24 has a memory for storing the decoded low-frequency audio x ⁻ _L for one frame, sends the decoded low-frequency audio one frame before to the high-frequency audio decoding unit 25 and the band synthesis filter 26, and receives the received decoded low frequency Memorize the voice. As will be described later, since the delay unit 24 can be omitted, the decoded low-frequency audio that is the output of the low-frequency audio decoding unit 22 and the decoded low-frequency audio one frame before that that is the output of the delay unit 24 Are not distinguished and are simply referred to as decoded low-frequency audio x ⁻ _L .

In step S23, the high frequency encoding extracting unit 23 extracts the high-frequency encoding c _H from the voice code c _L. The configuration of the high frequency code extracting unit 23 can be the same as that of the conventional high frequency code extracting unit 95. That is, back a bit string of 1 or 0 embedded in G.711 code LSB or MSB to high frequency code c _H. The high frequency code c _H is sent to the high frequency audio decoding unit 25.

In step S <b> 24, the high frequency audio decoding unit 25 decodes the high frequency code c _H using the decoded low frequency audio x ⁻ _L , and sends the decoded high frequency audio x ⁻ _H to the band synthesis filter 26. Details of the processing of the high frequency speech decoding unit 25 will be described later.

In step S25, the band synthesis filter 26, the decoded low-band speech x ^- _L and the decoded high-band speech x ^- from _H wideband decoded speech x ^- a synthesized and output. As the band division filter 26, a quadrature mirror filter (QMF: Quadrature Mirror Filter) used in G.711.1 and G.722 can be used as in the band division filter 12.

The speech decoding apparatus is configured to include a checksum detection unit 93 and a switch 97 as described in Patent Document 1, and determines whether or not the high-frequency code c _H is embedded in the received speech code c _L. It is also possible to perform a switching process of determining and outputting a wideband sound when embedded and outputting a narrowband sound when not embedded.

The delay unit 15 included in the speech encoding device and the delay unit 24 included in the speech decoding device may be omitted. Since the speech encoding apparatus has a feedback structure that decodes the low-frequency code c _{L in} which the high-frequency code c _H is embedded and encodes the high-frequency speech x _H , if the delay unit 15 is omitted, the speech encoding device decoded low-band speech x ^- _L and the decoded low-band speech x in the audio decoding device ^- and _L can not be matched. However, the difference is an indistinguishable difference in the sense of hearing, and there are few practical problems. If each delay unit is omitted, the delay time in voice communication can be shortened by one frame.

Hereinafter, a detailed configuration of the high frequency speech encoding unit 14 included in the speech encoding device will be described. As shown in FIG. 7, the high frequency speech encoding unit 14 includes a band division filter (also called a high frequency band division unit) 31 _H , a band division filter (also called a low frequency band division unit) 31 _L , and a power calculation unit 32. _H and 32 _L , linear prediction units 33 _H and 33 _L , a relative gain calculation unit 34, a coefficient encoding unit 35, a gain encoding unit 36, and a multiplexer (also referred to as a multiplexing unit) 37.

The high frequency speech encoding unit 14 receives high frequency speech x _H and decoded low frequency speech x ⁻ _L . As an example, when the sampling frequency of the input sound is 16 kHz and the frame length is 10 milliseconds, the high frequency sound x _H and the decoded low frequency sound x ⁻ _L both have a sampling frequency of 8 kHz and a frame length of 10 milliseconds. Yes, the number of samples per frame is 80 samples.

Band dividing filter 31 _L is decoded low-band speech x ^- _L sampling frequency each 4kHz of LL band speech x ^- is divided into the _LH ^- _LL and LH band speech x. The band division filter 31 _L may be the same as the band division filter 12 of the speech encoding apparatus, or may be a band division filter having a different number of taps and characteristics from the band division filter 12. LL band speech x ^- _LL because it does not use the high-band speech encoding unit 14, the band division filter 31 _L is LH band speech x ^- may be configured to output only _LH. The LH band speech x ⁻ _LH is input to the linear prediction unit 33 _L and the power calculation unit 32 _L.

The linear prediction unit 33 _L applies linear prediction analysis to the LH band speech x ⁻ _LH and outputs a p-order LH band linear prediction coefficient a _LH (i) (where i = 1, 2,..., P). To do. Here, a value of about 4 to 10 is generally used for p. Note that the p-th order linear prediction coefficient is a set of p values, but in the following, except for the case of indicating a linear prediction coefficient at a specific i, the index i is omitted and simply expressed as a _LH . . a _LH can also be regarded as a vector and is also called a linear prediction coefficient vector.

The power calculation unit 32 _L calculates the power P _LH for one frame of the LH band audio x ⁻ _LH . At this time, the average power including the preceding and succeeding frames, for example, the power for three frames including the signal before one frame and the signal after one frame, or 1/3 thereof may be set as the power for one frame. The same applies to the calculation of power for one frame.

The band division filter 31 _H divides the high frequency sound x _H into HL band sound x _HL and HH band sound x _HH each having a sampling frequency of 4 kHz. The band division filter 31 _H may be the same as the band division filter 12 of the speech encoding apparatus, or may be a band division filter having a different number of taps and different characteristics from the band division filter 12. Since the HH band speech x _HH does not use the high-band speech encoding unit 14, the band division filter 31 _H may be configured to output only the HL band speech x _HL. The HL band speech x _HL is input to the linear prediction unit 33 _H and the power calculation unit 32 _H.

The linear prediction unit 33 _H applies linear prediction analysis to the HL band speech x _HL and outputs a p-th order HL band linear prediction coefficient a _HL (i) (where i = 1, 2,..., P). . Hereinafter, like the LH band linear prediction coefficient a _LH , the index i is omitted, and is simply expressed as a _HL . a _HL can also be regarded as a vector like a _LH, and is also called a linear prediction coefficient vector.

The power calculation unit 32 _H calculates the power P _HL for one frame of the HL band audio x _HL .

The relative gain calculator 34 calculates a relative gain G _HL defined by the following equation. The relative gain G _HL is HL band speech x _HL of LH band speech x ^- is the relative gain for _LH, LH band speech x ^- power of the signal obtained by multiplying the relative gain G _HL to each sample of _LH is, HL band speech x _HL Power P _HL will be the same.

The coefficient encoding unit 35 encodes the HL band linear prediction coefficient a _HL with M ₁ bits using the LH band linear prediction coefficient a _LH and sends the coefficient code c ₁ to the gain encoding unit 36 and the multiplexer 37. Will be described later how the provisions of M _1.

The gain encoding unit 36 encodes the relative gain G _HL with M ₂ bits using the LH band linear prediction coefficient a _LH and the coefficient code c ₁ , and sends the gain code c ₂ to the multiplexer 37. Will be described later how the provisions of M _2.

How to define M ₁ and M ₂ will be described. According to Patent Document 1, even if a high frequency code of 16 bits per 160 samples of low frequency speech, that is, 8 bits per 80 samples is embedded in the low frequency code, the subjective quality of the decoded low frequency speech is not embedded in the high frequency code. It is said that it will not deteriorate. Therefore, when the frame length is 10 milliseconds (80 samples), it is preferable to determine M ₁ and M ₂ so that M ₁ + M ₂ ≦ 8. As an example, M ₁ = 4 and M ₂ = 4.

Coefficient coding section 35, the the LH band linear prediction coefficients a _LH and HL band linear prediction coefficients a _HL using a correlation, to encode the HL band linear prediction coefficients a _HL. For example, the value of the HL band linear prediction coefficient a _HL may be estimated from the value of the LH band linear prediction coefficient a _LH , and the error between the HL band linear prediction coefficient a _HL and the estimated value a ′ _HL may be encoded. The estimation uses a statistical method using a speech database.

The coefficient encoding unit 35 includes an LSP conversion unit 351, an LSP conversion unit 352, an LSP estimation unit 353, and an error encoding unit 354, as shown in FIG. The LSP converter 351 converts the HL band linear prediction coefficient a _HL into an HL band line spectrum pair (hereinafter, the line spectrum pair is referred to as LSP) f _HL . LSP is a kind of linear prediction parameter, and a p-order linear prediction coefficient and a p-order LSP can be mutually converted. Similarly to the linear prediction coefficient notation, the LSP notation omits the index i (i = 1, 2,..., P), and can be regarded as a vector when the index i is omitted. The LSP conversion unit 352 converts the LH band linear prediction coefficient a _LH into the LH band LSPf _LH . The LSP estimation unit 353 estimates the value of the HL band LSPf _HL using the LH band LSPf _LH . As the estimation rule, a statistical method using a speech database can be used. For example, a conversion function may be defined, and the correspondence relationship between the distribution of the LH band LSPf _{LH and} the distribution of the HL band LSPf _HL It may be defined by statistical examination. The error encoding unit 354 encodes an error between the HL band LSPf _HL and the estimated value f ′ _HL of the HL band LSP using, for example, a vector quantization method, and outputs a coefficient code c ₁ .

The gain encoding unit 36 encodes the relative gain G _HL by utilizing the correlation between the combination of the LH band linear prediction coefficient a _LH and the coefficient code c ₁ and the relative gain G _HL . For example, to estimate the value of the relative gain G _HL from a combination of LH band linear prediction coefficients a _LH and coefficient code c _1, the error between the relative gain G _HL and the estimated value G _'HL on a logarithmic scale (or decibels) It is good to encode. For estimation, a statistical method using a speech database may be used.

The multiplexer 37 receives the coefficient code c ₁ output from the coefficient encoding unit 35 and the gain code c ₂ output from the gain encoding unit 36 and outputs the input as a high frequency code c _H.

Details on speech analysis including linear prediction analysis are described in Reference 1 below.
[Reference 1] Sadahiro Furui, “Digital Audio Processing”, Tokai University Press, pp. 60-98

  Hereinafter, a detailed configuration of the high frequency speech decoding unit 25 included in the speech decoding apparatus will be described. As shown in FIG. 9, the high frequency speech decoding unit 25 includes a demultiplexer (also referred to as a code separation unit) 40, a band division filter (also referred to as a band division unit) 41, a power calculation unit 42, a linear prediction unit 43, and an inverse filter. 44, duplication unit 45, coefficient decoding unit 46, relative gain decoding unit 47, synthesis filter 48, power calculation unit 49, gain calculation unit 50, multiplication unit (also referred to as HL band multiplication unit) 51, relative gain prediction unit 52, coefficient A prediction unit 53, a random number unit 54, a synthesis filter 55, a power calculation unit 56, a gain calculation unit 57, a multiplication unit (also referred to as an HH band multiplication unit) 58, and a band synthesis filter (also referred to as a band synthesis unit) 59 are provided.

The high frequency speech decoding unit 25 receives the decoded low frequency speech x ⁻ _L and the high frequency code c _H. The high frequency code c _H is input to the demultiplexer 40. The decoded low frequency sound x ⁻ _L is input to the band division filter 41.

The band division filter 41 has the same configuration as that of the band division filter 31 _L of the high frequency audio encoding unit 14, and the decoded low frequency audio x ⁻ _L is obtained from the LL band audio x ⁻ _LL and the LH band audio x ⁻ with a sampling frequency of 4 kHz, respectively. Divide into _LH . Since the LL band voice x ^- _LL is not used in the high band voice decoding unit 25, the band division filter 41 may be configured to output only the LH band voice x ^- _LH . The LH band speech x ⁻ _LH is input to the linear prediction unit 43 and the power calculation unit 42.

Linear prediction unit 43, LH band speech x ^- by applying linear prediction analysis to _LH, and outputs a p-th order LH band linear prediction coefficients a _LH. The LH band linear prediction coefficient a _LH is input to the inverse filter 44, the coefficient decoding unit 46, the relative gain decoding unit 47, and the coefficient prediction unit 53.

Power calculation unit 42, like the power calculating portion 32 _L of the high-band speech encoding unit 14, LH band speech x ^- calculating the power P _LH of one frame of _LH. The power P _LH is input to the gain calculation unit 50 and the gain calculation unit 57.

The inverse filter 44 is an FIR filter using the LH band linear prediction coefficient a _LH as a filter coefficient, obtains the LH band linear prediction residual e _LH from the LH band speech x ⁻ _LH , and sends it to the duplicating unit 45. Here, x ⁻ _LH (j) is the jth sample of the LH band speech x ⁻ _LH , e _LH (j) is the jth sample of the LH band linear prediction residual, and j = 1 is the first sample of the current frame. , J = N represents the last sample of the current frame, e _LH (j) is expressed by the following equation.

  When one frame is composed of 80 samples, N = 80. When j-i is negative, the sample position in the past frame is represented as a relative sample position with respect to the first sample of the current frame. When representing a set of sample values for one frame, the index j is omitted.

The duplicating unit 45 duplicates the LH band linear prediction residual e _LH and outputs an HL band driving sound source e _HL as in the following equation. The HL band drive sound source e _HL is input to the synthesis filter 48.

Demultiplexer 40 divides the high-frequency encoding c _H in the coefficient code c ₁ and gain code c _2. The coefficient code c ₁ is input to the coefficient decoding unit 46, the relative gain decoding unit 47, the relative gain prediction unit 52, and the coefficient prediction unit 53. The gain code c ₂ is input to the relative gain decoding unit 47 and the relative gain prediction unit 52.

The coefficient decoding unit 46 decodes the coefficient code c ₁ using the LH band linear prediction coefficient a _LH and outputs the HL band decoded linear prediction coefficient a ⁻ _HL . The coefficient decoding unit 46 includes an LSP conversion unit 461, an LSP estimation unit 462, a reconstruction unit 463, and a coefficient conversion unit 464, as shown in FIG. The LSP conversion unit 461 and the LSP estimation unit 462 are the same as the LSP conversion unit 352 and the LSP estimation unit 353 of the coefficient encoding unit 35. The reconstructing unit 463 reconstructs the HL band decoded LSPf ^- _HL using the coefficient code c ₁ and the estimated value f ′ _HL of the HL band LSP by a decoding method corresponding to error coding. The coefficient conversion unit 464 converts the HL band decoded LSPf ^- _HL into the HL band decoded linear prediction coefficient a ^- _HL and outputs it. The HL band decoded linear prediction coefficient a ^- _HL is input to the synthesis filter 48.

The relative gain decoding unit 47 decodes the gain code c ₂ using a combination of the LH band linear prediction coefficient a _LH and the coefficient code c ₁ to obtain a decoded relative gain G ⁻ _HL . The decoded relative gain G ^- _HL is input to the gain calculation unit 50. As a decoding method, a method corresponding to the encoding method of the gain encoding unit 36 of the high frequency speech encoding unit 14 is used. For example, the relative gain G _HL is determined from the combination of the LH band linear prediction coefficient a _LH and the coefficient code c ₁ . The decoded relative gain G ⁻ _HL can be obtained by estimating the value and adding the error represented by the gain code c ₂ to the relative gain estimate G ′ _HL on a logarithmic scale or multiplying by a linear scale.

Synthesis filter 48, HL band decoded linear prediction coefficients received from the coefficient decoding unit 46 a ^- _HL of a IIR filter having a filter coefficient (also referred to as AR filter), HL band synthesis from HL band excitation e _HL speech y _HL Is output. The HL band synthesized speech y _HL is input to the power calculation unit 49 and the multiplication unit 51.

The power calculation unit 49 calculates the power P _HL for one frame of the HL band synthesized speech y _HL . The power P _HL is input to the gain calculation unit 50.

The gain calculation unit 50 calculates a gain g _HL represented by the following equation using the decoded relative gain G ⁻ _HL , the power P _LH , and the power P _HL . The gain g _HL is input to the multiplication unit 51.

Multiplication unit 51 multiplies the gain g _HL to HL band synthesized speech y _HL, decoding HL band speech x ^- calculating the _HL. The decoded HL band speech x ^- _HL is input to the band synthesis filter 59.

The relative gain prediction unit 52 predicts and obtains the predicted relative gain G ⁻ _HH using the coefficient code c ₁ and the gain code c ₂ . The predicted relative gain G ⁻ _HH is input to the gain calculation unit 57.

Coefficient prediction unit 53 uses the LH band linear prediction coefficients a _LH and coefficient code c _1, HH band linear prediction coefficients a ^- obtained by predicting _HH. The HH band linear prediction coefficient a ⁻ _HH is input to the synthesis filter 55.

The random number unit 54 generates a Gaussian random number and outputs a random number signal sequence e _HH having a length of 1 frame. The random number signal sequence e _HH is input to the synthesis filter 55.

The synthesis filter 55 is an IIR filter using the HH band linear prediction coefficient a ⁻ _HH as a filter coefficient, and outputs the HH band synthesized speech y _HH from the random number signal sequence e _HH . The HH band synthesized speech y _HH is input to the power calculator 56 and multiplier 58.

Power calculation unit 56 calculates the power P _HH of one frame of the HH band synthesized speech y _HH. The power _PHH is input to the gain calculator 57.

The gain calculation unit 57 calculates a gain g _HH represented by the following expression using the predicted relative gain G ⁻ _HH , the power P _LH , and the power P _HH . The gain g _HH is input to the multiplier 57.

Multiplier 58 multiplies the gain g _HH to HH band synthesized speech y _HH, decoding HH band speech x ^- calculating the _HH. The decoded HH band speech x ⁻ _HH is input to the band synthesis filter 59.

The band synthesizing filter 59 is a band synthesizing filter corresponding to the band dividing filter 31 _H of the high frequency audio encoding unit 14 (that is, as inverse transform), and includes the decoded HL band audio x ^- _HL and the decoded HH band audio x ^- _HH. Is used to generate and output decoded high frequency audio x ⁻ _H . Note that the sampling frequency of the decoded HL band audio x ^- _HL and the decoded HH band audio x ^- _HH are both 4 kHz, and the sampling frequency of the decoded high band audio x ^- _H is 8 kHz.

  The points of the speech coding apparatus and speech decoding apparatus in the present invention will be described.

  In the speech coding apparatus, wideband speech is band-divided into low-frequency speech and high-frequency speech, low-frequency speech is further divided into LL-band signals and LH-band signals, high-frequency speech is further transformed into HL-band signals and HH. Band division into band signals. That is, the wideband voice is divided into four bands of LL band, LH band, HL band, and HH band.

  In order to embed high-frequency speech information in a low-frequency code without degrading the quality of the decoded low-frequency speech, it is necessary to encode the high-frequency speech with as few bits as possible. Therefore, the spectral envelope information and power information of the HL band are encoded with a small number of bits that do not deteriorate the quality of the decoded low-frequency speech and are embedded in the low-frequency code. In order to encode these pieces of information with a small number of bits, encoding is performed by utilizing the correlation between parameters to the maximum. At this time, information on the HH band is not sent.

  In the speech decoding apparatus, HL band spectrum envelope information and power information are extracted from the low band code, and an HL band signal and an HH band signal are generated. In general, a speech coding method using linear prediction requires spectral envelope information, excitation information for driving a synthesis filter, and information representing power. However, a speech coding apparatus uses excitation information for driving a synthesis filter. Therefore, it is necessary to pseudo-generate sound source information for driving the synthesis filter from other information obtained by the speech decoding apparatus. Therefore, it is assumed that the linear prediction residual signal in the LH band is the same as the sound source information that drives the synthesis filter in the HL band, and driving the synthesis filter in the HL band by driving the HL band synthesis filter with the linear prediction residual signal in the LH band. Generate a signal. In addition, since no information is sent from the speech coding apparatus for the HH band, a pseudo HH band signal is generated from the LH band and HL band information obtained by the speech decoding apparatus. Specifically, spectral envelope information of the HH band and information representing power are predicted from the information of the LH band and the HL band by a statistical method, and the synthesis filter is driven with a Gaussian random number.

  By the above method, the high frequency sound is expressed by 8 bits per 10 milliseconds, and the wideband sound having sufficiently good quality can be reproduced from the sound decoding device. Note that the reproduced wideband sound has good perceptual quality, but the SN ratio with the input sound, particularly the high-frequency SN ratio is not high. The reason why the human auditory characteristic is good even though the S / N ratio is not high is that if the human auditory characteristics are reproduced in a state where the spectral envelope and power are close to those of the input speech in the high frequency range, the driving sound source of linear prediction That is, it is insensitive to the fine structure and phase of the spectrum. In addition, high-frequency spectral envelope and high-reproducibility coding with less power, especially that the HH band can be reproduced without sending information, the high-frequency spectral envelope and power It is realized by utilizing the high correlation with the low-frequency spectrum envelope and power.

[Modification A]
The above embodiment is an example of a wideband speech coding apparatus and wideband speech decoding apparatus that are fully compatible with G.711. As shown below, this is a wideband that is fully compatible with G.726. It can be modified to a speech encoding device and a wideband decoding device.

FIG. 11 is a configuration example using the speech encoding unit described in Patent Literature 1 as the low frequency speech encoding unit 13 of the embodiment. As shown in FIG. 11, the low-frequency speech encoding unit 13 includes a speech signal buffer 131, a bit analysis unit 132, a switch 133, a G.711 all code search unit 134, a G.711 even code search unit 135, and a G.711. An odd code search unit 136 and a speech code buffer 137 are provided. Audio signal buffer 131 outputs separated at predetermined intervals called frames low-band speech x _L. The bit analysis unit 132 decomposes the high-frequency code c _H into a bit string for each frame, and controls the switch 133 according to the value of each bit. Among the sample points of the frame, at the sample point corresponding to each bit of the predetermined bit string, if the value of the bit is 0, the G.711 even code search unit 135 corresponds to the even code of G.711. From the 128 codes to be searched, a code having a quantization value close to the input voice sample value is searched, and the LSB outputs a G.711 voice code representing the value of the bit according to the search result. If the value of the bit is 1 at the sample point corresponding to each bit, the G.711 odd code search unit 136 performs the quantization on the input audio sample value from the 128 code corresponding to the odd code of G.711. A code with a similar value is searched, and an LSB outputs a G.711 speech code representing the value of the bit according to the search result. Except for the sample points corresponding to each of the bits, the G.711 all code search unit 134 searches for a code whose quantization value is close to the input voice sample value from all 256 codes of G.711, and according to the search result. Outputs G.711 audio code. The voice code buffer 137 outputs the G.711 voice code for one frame as a voice code (also a low frequency code) c _L.

  FIG. 12 shows a modified example of the low-frequency speech encoding unit for realizing wideband speech encoding that is completely compatible with G.726. Similar to the low frequency speech encoding unit 13, the low frequency speech encoding unit 13A of the modified example includes an audio signal buffer 131, a bit analysis unit 132, and a speech code buffer 137, and further includes a G.726 compatible encoder 140. . The G.726 compatible encoder 140 includes a switch 143, a G.726 all code search unit 144, a G.726 even code search unit 145, a G.726 odd code search unit 146, an adaptive prediction unit 148, and a differential signal calculation unit 149. Prepare. G.726 is called an adaptive differential pulse code modulation (ADPCM) system. In G.726, the adaptive prediction unit 148 predicts the current audio sample value using the quantized value of the past audio sample, and the difference signal calculation unit 149 sends the current audio sample sent from the audio signal buffer 131. The difference is quantized with 2 to 5 bits per sample by calculating a difference signal between the value and the predicted value of the speech sample sent from the adaptive prediction unit 148. In general, a mode that uses 4 bits per sample (also called 32kbit / s mode) is generally used, such as digital cordless telephones. Therefore, the following explanation uses 4 bits per sample as an example. To do.

  Among the sample points of the frame, at the sample point corresponding to each bit of the predetermined bit string, if the value of the bit is 0, the G.726 even code search unit 145 corresponds to the even code of G.726. A code having a quantization value close to that of the difference signal is searched from the 8 codes, and the LSB outputs a G.726 speech code representing the value of the bit according to the search result. If the value of the bit is 1 at the sample point corresponding to each bit, the G.726 odd code search unit 146 uses the 8 code corresponding to the odd code of G.726 to calculate the quantized value to the differential signal. Search for a close code, and according to the search result, the LSB outputs a G.726 speech code representing the value of the bit. Except for the sample points corresponding to the respective bits, the G.726 all-code search unit 144 searches for a code having a quantization value close to the difference signal from all 16 codes of G.726, and according to the search result, G.726 Is output. The G.726 speech code is sent to the speech code buffer 137 and also to the adaptive prediction unit 148.

Details of G.726 other than the above are described in ITU-T G.726. In addition, since G.726 has a small number of quantization bits per sample, when a low frequency code in which a high frequency code is embedded is generated, the sense of noise of the decoded low frequency audio obtained by decoding the low frequency code is G.711. There are more problems. As a method of reducing this noise sensation, as shown in FIG. 13, “a method of effectively reducing the harshness of quantization noise” described in Reference 2 can be combined.
[Reference Document 2] Japanese Patent No. 5014493

  The method of embedding the high-frequency code as a bit string of 1 or 0 in LSB or MSB is as follows. The present invention is not limited to G.726 and can be applied to a wide range of existing speech coding schemes. By replacing G.726 in the above configuration example with other existing speech coding schemes, it is fully compatible with the existing speech coding schemes. A wideband speech encoding apparatus and a wideband speech decoding apparatus having the above can be realized.

  In this modification, the even code is searched when the embedded high-frequency code bit is 0, and the odd code is searched when the embedded high-frequency code bit is 0. However, when the number is 1, a correspondence relationship for searching for an even code may be used.

[Modification B]
FIG. 14 is a configuration example of the low frequency speech decoding unit 16 in the speech encoding device of the embodiment and the low frequency speech decoding unit 22 in the speech decoding device of the embodiment. An existing G.711 decoder can be simply used for the low-band speech decoding unit of the wideband speech coding apparatus and wideband speech decoding apparatus that are fully compatible with G.711. 15 is a modification of the low-band speech encoding unit of the wide-band speech coding apparatus and the wide-band decoding apparatus for realizing wide-band speech coding fully compatible with G.726. The .711 decoder 162 is replaced with the existing G.726 decoder 163.

  Since G.726 has a small number of quantized bits per sample, there is a problem that the noise sensation of decoded low-frequency speech is higher than that of G.711. As a method for reducing this noise sensation, as shown in FIG. 16, a post filter 164 widely used as a narrowband speech coding technique can be used after the G.726 decoder 163. The post filter is widely used as a method for improving the quality of decoded speech of so-called “CELP system” such as ITU-T G.729 and PSI-CELP used in Japanese mobile phones. Therefore, detailed description is omitted here.

[Modification C]
FIG. 17 is a modification of the high frequency audio decoding unit 25 shown in FIG. The high frequency speech decoding unit 25A of the modification is different from the high frequency speech decoding unit 25 in that it includes a designated band signal extraction unit 61, a linear prediction unit 62, and an inverse filter 63. The high frequency speech decoding unit 25 generates an HL band signal by driving an HL band synthesis filter with an LH band linear prediction residual signal. Here, the linear prediction residual signal in the LH band is a linear prediction residual signal in the 2 to 4 kHz band, but in the high frequency speech decoding unit 25A, instead of the 2 to 4 kHz band,
F to (F + 2) kHz band However, 0 ≦ F ≦ 2
The synthesis filter in the HL band is driven by the linear prediction residual signal. Here, F is a predetermined constant. In the configuration of FIG. 17, the designated band signal extraction unit 61 extracts the F to (F + 2) kHz band components of the decoded low frequency sound x ⁻ _L to obtain the designated band signal x ⁻ _LF of 4 kHz sampling. The linear prediction unit 62 applies linear prediction analysis to the designated band signal x ⁻ _LF , and outputs a p-th order designated band linear prediction coefficient a _LF . The designated band linear prediction coefficient a _LF is input to the inverse filter 63. The inverse filter 63 is an FIR filter using the designated band linear prediction coefficient a _LF as a filter coefficient, obtains the designated band linear prediction residual e _LF from the designated band signal x ⁻ _LF , and sends it to the duplicating unit 45. Replication unit 45 outputs the HL band excitation e _HL duplicates the specified band linear prediction residual e _LF.

The high frequency speech decoding unit 25 shown in FIG. 9 corresponds to F = 2, but a case where F <2 is more preferable will be described. FIG. 18 (A) is an example of a frequency spectrum of the low-band speech x _L, and FIG. 18 (B) is decoded low-band speech x corresponding to the example of FIG. 18 (A) ^- the frequency spectrum of the _L. Comparing FIG. 18A and FIG. 18B, in FIG. 18B, noise is superimposed over the entire band due to coding distortion, and the SN ratio in the 2-4 kHz band is, for example, in the 1-3 kHz band. It can be said that it is worse than the SN ratio. In such a case, rather than driving a synthesis filter in the HL band with a linear prediction residual signal in the 2 to 4 kHz band,
F to (F + 2) kHz band However, 0 ≦ F <2
Driving the HL band synthesis filter with the linear prediction residual signal can reduce the noise feeling of the generated HL band signal.

[Modification D]
FIG. 19 is a configuration example of the designated band signal extraction unit 61. The designated band signal extraction unit 61 includes a Fourier transform unit 161, a frequency component circulation unit 162, a Fourier inverse transform unit 163, and a band division filter (also referred to as a designated band division unit) 164. The Fourier transform unit 161 performs Fourier transform on the decoded low frequency sound x ^- _L to convert it into a frequency domain decoded low frequency sound X ^- _L . The frequency component circulating unit 162 has the fkHz component of the frequency domain decoded low-frequency speech X ⁻ _L ,
f '= f + 2-F when f≤2 + F
When f> 2 + F f '= f-2-F
The frequency domain cyclic decoded low-frequency speech X ⁻ _LS having the f ′ kHz component is calculated. Inverse Fourier transform unit 163, frequency domain cyclic decoded low-band speech X ^- the _LS and inverse Fourier transform, cyclic decoded low-band speech x ^- calculating the _LS. Similarly to the band division filter 41 of the high frequency audio decoding unit 25 shown in FIG. 9, the band division filter 164 converts the input cyclic decoded low frequency audio x ^- _LS into a high frequency audio and a low frequency having a sampling frequency of 4 kHz, respectively. Divided into voices, high frequency voice is output as designated band signal x ^- _LF .

  As described above, the embodiments of the present invention have been described, but the specific configuration is not limited to these embodiments, and even if there is a design change or the like as appropriate without departing from the spirit of the present invention, Needless to say, it is included in this invention. The various processes described in the embodiments are not only executed in time series according to the description order, but may also be executed in parallel or individually as required by the processing capability of the apparatus that executes the processes.

[Program, recording medium]
When various processing functions in each device described in the above embodiment are realized by a computer, the processing contents of the functions that each device should have are described by a program. Then, by executing this program on a computer, various processing functions in each of the above devices are realized on the computer.

  The program describing the processing contents can be recorded on a computer-readable recording medium. As the computer-readable recording medium, for example, any recording medium such as a magnetic recording device, an optical disk, a magneto-optical recording medium, and a semiconductor memory may be used.

  This program is distributed by selling, transferring, or lending a portable recording medium such as a DVD or CD-ROM in which the program is recorded. Furthermore, the program may be distributed by storing the program in a storage device of the server computer and transferring the program from the server computer to another computer via a network.

  A computer that executes such a program first stores, for example, a program recorded on a portable recording medium or a program transferred from a server computer in its own storage device. When executing the process, the computer reads a program stored in its own recording medium and executes a process according to the read program. As another execution form of the program, the computer may directly read the program from a portable recording medium and execute processing according to the program, and the program is transferred from the server computer to the computer. Each time, the processing according to the received program may be executed sequentially. A configuration in which the above-described processing is executed by a so-called ASP (Application Service Provider) type service that realizes a processing function only by an execution instruction and result acquisition without transferring a program from the server computer to the computer. It is good. Note that the program in this embodiment includes information that is used for processing by an electronic computer and that conforms to the program (data that is not a direct command to the computer but has a property that defines the processing of the computer).

  In this embodiment, the present apparatus is configured by executing a predetermined program on a computer. However, at least a part of these processing contents may be realized by hardware.

DESCRIPTION OF SYMBOLS 11 Input buffer 12 Band division | segmentation filter 13 Low frequency audio | voice encoding part 14 High frequency audio | voice encoding part 15 Delay part 16 Low frequency audio decoding part 17 Code transmission part 21 Code receiving part 22 Low frequency audio decoding part 23 High frequency code extraction part 24 delay unit 25 high frequency speech decoding unit 26 band synthesis filter

Claims (8)

A band dividing unit that divides the input sound into low-frequency sound and high-frequency sound;
A high frequency audio encoding unit that encodes the high frequency audio based on the decoded low frequency audio to generate a high frequency code;
A low frequency speech encoding unit that encodes the low frequency speech and generates a low frequency code in which the high frequency code is embedded;
A low frequency audio decoding unit that decodes the low frequency code to generate the decoded low frequency audio;
A code sending unit that outputs the low frequency code as a voice code;
Including
The above low-pass code is compatible with the ITU-T G.726 system.
Speech encoding device. A code receiving unit that receives a voice code output by the voice encoding device according to claim 1;
A low frequency audio decoding unit that decodes the audio code to generate a decoded low frequency audio;
A high-frequency code extraction unit that extracts a high-frequency code embedded in the speech code;
A high-frequency speech decoding unit that generates the decoded high-frequency speech by decoding the high-frequency code based on the decoded low-frequency speech;
A band synthesizing unit that synthesizes the decoded low-frequency sound and the decoded high-frequency sound and outputs the decoded sound;
A speech decoding device. The speech decoding apparatus according to claim 2, wherein
The high frequency speech decoding unit is
A band dividing unit for dividing the decoded low-frequency sound into LH band sound and LL band sound;
A code separation unit for separating the speech code into a gain code and a coefficient code;
A coefficient decoding unit that decodes the coefficient code using the linear prediction coefficient of the LH band speech to obtain an HL band decoded linear prediction coefficient;
A relative gain decoding unit that decodes the gain code using the linear prediction coefficient of the LH band speech and the coefficient code to obtain a decoded relative gain;
A coefficient prediction unit that predicts an HH band linear prediction coefficient using the LH band linear prediction coefficient and the coefficient code;
A relative gain prediction unit that predicts a predicted relative gain using the gain code and the coefficient code;
A designated band signal extraction unit that obtains a designated band signal by extracting F to (F + 2) kilohertz band components of the decoded low-frequency speech, with F being a predetermined constant of 0 ≦ F ≦ 2,
A duplicating unit for duplicating a designated band linear prediction residual obtained from the designated band signal using a linear prediction coefficient of the designated band signal as a filter coefficient and obtaining an HL band driving sound source;
The HH band voice is multiplied by the gain calculated from the predicted relative gain based on the ratio of the power of the HH band voice obtained from Gaussian random numbers and the power of the LH band voice using the HH band linear prediction coefficient as a filter coefficient. An HH band multiplier for generating decoded HH band speech;
The gain calculated from the decoded relative gain based on the ratio of the power of the HL band synthesized speech obtained from the HL band driving sound source and the power of the LH band speech using the HL band decoded linear prediction coefficient as a filter coefficient is the HL band. An HL band multiplier that multiplies the synthesized voice to generate decoded HL band voice;
A band synthesizing unit that synthesizes the decoded HH band voice and the decoded HL band voice and outputs the decoded high band voice;
Including
Speech decoding device. The speech decoding apparatus according to claim 3, wherein
The designated band signal extraction unit
A Fourier transform unit that converts the decoded low-frequency sound into a frequency domain to generate a frequency-domain decoded low-frequency sound;
F ′ = f + 2-F when the f-kilohertz component of the above frequency domain decoded low-frequency speech is f ≦ 2 + F, and f ′ = f-2-F when f> 2 + F. 'Frequency component cyclic unit for calculating frequency domain cyclic decoded low-frequency speech composed of kilohertz components;
A Fourier inverse transform unit that converts the frequency domain cyclic decoded low-frequency speech into the time domain to generate cyclic decoded low-frequency speech;
A designated band dividing unit that divides the cyclic decoded low frequency sound into high frequency sound and low frequency sound, and outputs the high frequency sound as the designated band signal;
Including
Speech decoding device. The band division unit divides the input sound into low-frequency sound and high-frequency sound,
A high frequency audio encoding unit generates the high frequency code by encoding the high frequency audio based on the decoded low frequency audio,
The low frequency speech encoding unit generates a low frequency code in which the low frequency speech is encoded and the high frequency code is embedded,
A low-frequency speech decoding unit decodes the low-frequency code to generate the decoded low-frequency speech,
The code sending unit outputs the low frequency code as a voice code,
The above low-pass code is compatible with the ITU-T G.726 system.
Speech encoding method. The code receiving unit receives the voice code output by the voice coding method according to claim 5,
A low-frequency audio decoding unit generates the decoded low-frequency audio by decoding the audio code,
The high frequency code extraction unit extracts the high frequency code embedded in the speech code,
A high frequency audio decoding unit generates the decoded high frequency audio by decoding the high frequency code based on the decoded low frequency audio,
A band synthesizing unit that synthesizes the decoded low-frequency sound and the decoded high-frequency sound and outputs the decoded sound;
Speech decoding method.   A program for causing a computer to function as the speech encoding device according to claim 1 or the speech decoding device according to any one of claims 2 to 4.   A computer-readable recording medium on which a program for causing a computer to function as the speech encoding device according to claim 1 or the speech decoding device according to any one of claims 2 to 4 is recorded.

Images