JP4985743B2 - Speech code conversion method - Google Patents

Description

  The present invention relates to a speech code conversion method, and in particular, a speech code encoded by a speech encoding device used in a network such as the Internet or a speech encoding device used in an automobile / mobile phone system or the like is differently encoded. The present invention relates to a voice code conversion method for converting into a voice code.

  In recent years, the number of mobile phone subscribers has increased explosively and is expected to continue to increase in the future. Voice communication over the Internet (Voice over IP: VoIP) has become widespread in fields such as corporate networks and long distance telephone services. In such a voice communication system, a voice coding technique for compressing voice is used in order to effectively use a communication line, but a voice coding method used for each system is different. For example, in W-CDMA, which is expected as a next-generation mobile phone system, an AMR (Adaptive Multi-Rate) system is adopted as a world-wide voice encoding system. On the other hand, in VoIP, the ITU-T recommendation G.729A system is widely used as a voice encoding system.

  In the future, with the spread of the Internet and mobile phones, it is considered that the volume of voice communication between Internet users and mobile phone users will increase further. However, as described above, the cellular phone network and the Internet network cannot communicate with each other because they use different voice encoding methods. For this reason, conventionally, it is necessary to convert a voice code encoded in one network into a voice code of a voice code method used in the other network by a voice code converter.

Speech code conversion FIG. 15 shows a principle diagram of a conventional typical speech code conversion method. Hereinafter, this method is referred to as Prior Art 1. In the figure, only the case where the voice input by the user A to the terminal 1 is transmitted to the terminal 2 of the user B is considered. Here, the terminal 1 possessed by the user A has only the encoder 1a of the encoding scheme 1, and the terminal 2 possessed by the user B has only the decoder 2a of the encoding scheme 2.

  The voice uttered by the user A on the transmission side is input to the encoder 1a of the encoding method 1 incorporated in the terminal 1. The encoder 1a encodes the input audio signal into an encoding method 1 audio code and sends it to the transmission line 1b. When a speech code is input via the transmission path 1b, the decoder 3a of the speech code conversion unit 3 once decodes the reproduced speech from the speech code of the encoding scheme 1. Subsequently, the encoder 3b of the voice code converter 3 converts the reproduced voice signal into a voice code of the encoding method 2 and sends it to the transmission line 2b. The voice code of this encoding scheme 2 is input to the terminal 2 through the transmission path 2b. When the voice code is input, the decoder 2a decodes the reproduced voice from the voice code of the encoding method 2. Thereby, the user B on the receiving side can listen to the reproduced sound. The process of decoding the speech once encoded as described above and encoding the decoded speech again is called tandem connection.

As described above, in the configuration of the conventional technique 1, since the speech code encoded by the speech encoding method 1 is once decoded into the encoded speech and then encoded again by the speech encoding method 2, the speech quality is improved. There were problems such as significant deterioration of the network and increased delay. As a method for solving such a problem of tandem connection, the speech code is decomposed into parameter codes such as LSP code and pitch lag code without returning to speech signals, and each parameter code is individually converted into another speech coding method. A method of converting to a code has been proposed (see Patent Document 1). FIG. 16 shows the principle diagram. Hereinafter, this is referred to as Prior Art 2.

  The encoding method 1 encoder 1a incorporated in the terminal 1 encodes the audio signal emitted by the user A into the encoding method 1 audio code and sends it to the transmission line 1b. The voice code conversion unit 4 converts the voice code of the coding method 1 input from the transmission line 1b into the voice code of the coding method 2 and sends it to the transmission line 2b. The decoder 2a of the terminal 2 uses the transmission line 2b. The user B can listen to the reproduced voice by decoding the reproduced voice from the voice code of the encoding method 2 input via the user.

  The encoding method 1 includes (1) a first LSP code obtained by quantizing an LSP parameter obtained from a linear prediction coefficient (LPC count) obtained by linear prediction analysis for each frame, and (2) a periodic sound source. A first pitch lag code that specifies an output signal of an adaptive codebook for outputting a signal, and (3) an output signal of an algebraic codebook (or a noise codebook) for outputting a noisy excitation signal. 1 is obtained by quantizing an algebraic code (noise code) of 1 and (4) a pitch gain representing the amplitude of the output signal of the adaptive codebook and an algebraic gain representing the amplitude of the output signal of the algebraic codebook. This is a method of encoding an audio signal with a gain code. The encoding method 2 is obtained by quantization by a quantization method different from the first speech encoding method (1) a second LSP code, (2) a second pitch lag code, (3) a second In this method, a speech signal is encoded with an algebraic code (noise code) and (4) a second gain code.

  The speech code conversion unit 4 includes a code separation unit 4a, an LSP code conversion unit 4b, a pitch lag code conversion unit 4c, an algebraic code conversion unit 4d, a gain code conversion unit 4e, and a code multiplexing unit 4f. The code separation unit 4a is a code of a plurality of components necessary for reproducing a speech signal from the speech code of the encoding method 1 inputted from the encoder 1a of the terminal 1 through the transmission line 1b, that is, (1) An LSP code, (2) a pitch lag code, (3) an algebraic code, and (4) a gain code are separated and input to the code conversion units 4b to 4e, respectively. Each of the code conversion units 4b to 4e converts the input LSP code, pitch lag code, algebraic code, and gain code according to the speech coding method 1 into an LSP code, pitch lag code, algebraic code, and gain code according to the speech coding method 2, respectively. The code multiplexing unit 4f multiplexes the converted codes of the audio coding method 2 and sends them to the transmission line 2b.

  FIG. 17 is a block diagram of the voice code conversion unit in which the configurations of the code conversion units 4b to 4e are clearly shown. The same reference numerals are given to the same parts as those in FIG. The code separation unit 4a separates the LSP code 1, the pitch lag code 1, the algebraic code 1, and the gain code 1 from the speech code of the coding method 1 that is input from the transmission line via the input terminal # 1, and the code conversion unit 4b. Enter in ~ 4e.

The LSP dequantizer 4b ₁ of the LSP code converter 4b dequantizes the LSP code 1 of the encoding scheme 1 and outputs an LSP dequantized value, and the LSP quantizer 4b ₂ outputs the LSP dequantized value. Is quantized using the LSP quantization table of encoding method 2 and LSP code 2 is output. The pitch lag dequantizer 4c ₁ of the pitch lag code conversion unit 4c dequantizes the pitch lag code ₁ of the encoding scheme 1 and outputs a pitch lag dequantized value, and the pitch lag quantizer 4c ₂ outputs the pitch lag dequantized value. Is quantized using the pitch lag quantization table of encoding method 2 and pitch lag code 2 is output. The algebraic code dequantizer 4d ₁ of the algebraic code converter 4d dequantizes the algebraic code ₁ of the encoding scheme 1 and outputs an algebraic code dequantized value. The algebraic code quantizer 4d ₂ The inverse quantization value is quantized using the algebraic code quantization table of encoding method 2 and algebraic code 2 is output. The gain dequantizer 4e ₁ of the gain code converter 4e dequantizes the gain code 1 of the encoding scheme 1 and outputs a gain dequantized value, and the gain quantizer 4e ₂ outputs the gain dequantized value. Is quantized using a gain quantization table of encoding method 2 and gain code 2 is output.
The code multiplexer 4f multiplexes the LSP code 2, the pitch lag code 2, the algebraic code 2, and the gain code 2 output from the quantizers 4b _{2 to} 4e ₂ to create a speech code according to the encoding scheme 2 and output it. Send to terminal # 2 to transmission line.

  In the tandem connection method (prior art 1) in FIG. 15, the reproduced speech obtained by temporarily decoding the speech code encoded by the encoding method 1 is input and the encoding and decoding are performed again. For this reason, since speech parameters are extracted from reproduced speech that has a much smaller amount of information than the original sound by re-encoding (i.e., speech information compression), the resulting speech code is not necessarily optimal. There wasn't. On the other hand, according to the speech coding apparatus of the related art 2 in FIG. 16, in order to convert the speech code of the coding method 1 into the speech code of the coding method 2 through the process of inverse quantization and quantization, Compared with the tandem connection of the prior art 1, speech code conversion with much less deterioration is possible. Further, since there is no need to decode the speech once for speech code conversion, there is an advantage that the delay which has been a problem in the conventional tandem connection can be reduced.

-Non-voice compression By the way, an actual voice communication system generally has a non-voice compression function that effectively uses non-voice sections included in a voice conversation to further improve information transmission efficiency. FIG. 18 shows a conceptual diagram of the non-voice compression function. In human conversation, there are non-speech intervals such as silence and background noise between speech. In such a section, it is not necessary to transmit voice information, and the communication line can be used more effectively. This is the basic idea of non-voice compression. However, if there is no change, there will be no sound between the audio played back on the receiving side and audio will be unnatural, so natural noise (comfort noise) that is not audibly unnatural on the receiving side is usually generated. generate. In order to generate comfort noise similar to the input signal, it is necessary to transmit comfort noise information (hereinafter referred to as CN information) from the transmission side. However, the amount of information of CN information is smaller than that of speech, and in the non-speech interval Since the properties change slowly, it is not always necessary to send CN information. As a result, the amount of information to be transmitted can be significantly reduced as compared with the voice section, so that the transmission efficiency of the entire communication line can be further improved. Such non-speech compression functions include a VAD unit (Voice Activity Detection) that detects speech and non-speech intervals, and a DTX unit (Discontinuous Transmission) that performs CN information generation and transmission control on the transmission side. Transmission control) and a CNG unit (Comfort Noise Generator) that generates comfort noise on the receiving side.

Hereinafter, the operation principle of the non-voice compression function will be described. FIG. 19 shows the principle diagram.
On the transmission side, an input signal divided into frames of a certain length (for example, 80 samples / 10 msec) is input to the VAD unit 5a to detect a voice section. The VAD unit 5a outputs a determination result vad_flag of 1 in the voice interval and 0 in the non-voice interval. In the case of a speech section (vad_flag = 1), all the switches SW1 to SW4 are switched to the speech side, and the speech encoder 5b on the transmission side and the speech decoder 6a on the reception side are in a normal speech coding scheme (for example, G.729A or The audio signal is encoded and decoded according to AMR). On the other hand, in the case of non-speech period (vad_flag = 0), all switches SW1 to SW4 are switched to the non-speech side, and the non-speech encoder 5c on the transmission side encodes a non-speech signal under the control of the DTX unit (not shown) Processing, that is, CN information generation / transmission control is performed, and the non-speech decoder 6b on the receiving side generates decoding processing, that is, comfort noise under the control of a CNG unit (not shown).

Next, operations of the non-speech encoder 5c and the non-speech decoder 6b will be described. FIG. 20 shows a block diagram of each, and FIGS. 21 (a) and 21 (b) show respective processing flows.
The CN information generation unit 7a analyzes the input signal for each frame and calculates a CN parameter for generating comfort noise in the CNG unit 8a on the receiving side (step S101). As the CN parameter, generally, outline information of frequency characteristics and amplitude information are used. The DTX control unit 7b controls the switch 7c to control whether or not the obtained CN information is transmitted to the receiving side for each frame (S102). As a control method, there are a method of adaptively controlling according to the nature of the signal and a method of periodically controlling at regular intervals. When transmission is necessary, the CN parameter is input to the CN quantization unit 7d, and the CN quantization unit 7d quantizes the CN parameter to generate a CN code (S103), and transmits it to the receiving side as line data (S103). S104). Hereinafter, a frame in which CN information is transmitted is referred to as a SID (Silence Insertion Descriptor) frame. Other frames are non-transmission frames and nothing is transmitted (S105).

The CNG unit 8a on the receiving side generates comfort noise based on the transmitted CN code. That is, the CN code sent from the transmission side is input to the CN dequantization unit 8b, and the CN dequantization unit 8b dequantizes the CN code into CN parameters (S111), and the CNG unit 8a Comfort noise is generated using the parameters (S112). In a non-transmission frame in which no CN parameter is transmitted, comfort noise is generated using the last received CN parameter (S113).
As described above, in an actual voice communication system, a non-speech section in a conversation is determined, and only information for generating aurally natural noise on the receiving side is intermittently transmitted in the non-speech section. As a result, the transmission efficiency can be further improved. Such a non-voice compression function is also employed in the next-generation mobile phone network and VoIP network described above, and different systems are used for each system.

G.729A、AMRともCN情報としてLPC係数(線形予測計数)とフレーム信号電力が用いられる。LPC係数は入力信号の周波数特性の概形を表わすパラメータであり、フレーム信号電力は入力信号の振幅特性を表わすパラメータである。これらパラメータはフレーム毎に入力信号を分析することによって得られる。以下にG.729AとAMRのCN情報の生成方法を述べる。 Next, the non-speech compression function used in G.729A (VoIP) and AMR (next generation mobile phone), which are typical coding systems, will be described. Table 1 shows the specifications of both types.

Both G.729A and AMR use LPC coefficients (linear prediction count) and frame signal power as CN information. The LPC coefficient is a parameter representing an outline of the frequency characteristic of the input signal, and the frame signal power is a parameter representing the amplitude characteristic of the input signal. These parameters are obtained by analyzing the input signal every frame. The following describes how to generate CN information for G.729A and AMR.

  In G.729A, LPC information is obtained as an average value of LPC coefficients of the past six frames including the current frame. Further, in consideration of signal fluctuations in the vicinity of the SID frame, the obtained average value or the LPC coefficient of the current frame is finally used as CN information. Which one to select is determined by measuring the strain between both LPC coefficients. When it is determined that the signal has a variation (large distortion), the LPC coefficient of the current frame is used. The frame power information is obtained as a value obtained by averaging the logarithmic power of the LPC prediction residual signal in the past 0 to 3 frames including the current frame. Here, the LPC residual signal is a signal obtained by passing an input signal through an LPC inverse filter for each frame.

In AMR, LPC information is obtained as an average value of LPC coefficients of the past 8 frames including the current frame. The average value is calculated in a region where LPC coefficients are converted into LSP parameters. Here, LSP is a parameter in the frequency domain that can be mutually converted with the LPC coefficient. The frame signal power information is obtained as a value obtained by averaging the logarithmic power of the input signal over the past 8 frames (including the current frame).
As described above, both G.729A and AMR use LPC information and frame signal power information as CN information, but their generation (calculation) methods are different.

  The CN information is quantized into a CN code and transmitted to the decoder. Table 1 shows the bit allocation of G.729A and AMR CN codes. G.729A quantizes LPC information with 10 bits and frame power information with 5 bits. On the other hand, in AMR, LPC information is quantized with 29 bits and frame power information with 6 bits. Here, the LPC information is converted into LSP parameters and quantized. In this way, bit allocation for quantization is different between G.729A and AMR. 22 (a) and 22 (b) are non-speech code (CN codes) configuration diagrams in G.729A and AMR, respectively.

In G.729A, as shown in FIG. 22 (a), the size of the non-voice code is 15 bits, and is composed of an LSP code I_LSPg (10 bits) and a power code I_POWg (5 bits). Each code is composed of a codebook index (element number) held by a G.729A quantizer, and details are as follows. That is, (1) LSP code I_LSPg is composed of codes L _G1 (1 bit), L _G2 (5 bits), L _G3 (4 bits), and L _G1 is the prediction information switching information of the LSP quantizer, L _G2 , L _G3 is an index of codebooks CB _G1 and CB _G2 of the LSP quantizer, and (2) power code is an index of codebook CB _G3 of the power quantizer.
In AMR, as shown in FIG. 22 (b), the size of the non-voice code is 35 bits, and is composed of an LSP code I_LSPa (29 bits) and a power code I_POWA (6 bits). Each code is composed of a codebook index of the AMR quantizer, and details are as follows. That is, (1) LSP code I_LSPa is composed of codes L _A1 (3 bits), L _A2 (8 bits), L _A3 (9 bits), L _A4 (9 bits), and each code is a code book GB of the LSP quantizer. Each index of _A1 , GB _A2 , GB _A3 , GB _A4 , (2) The power code is an index of the code book GB _A5 of the power quantizer.

-DTX control Next, the DTX control method is described. FIG. 23 shows the time flow of G.729A, and FIGS. 24 and 25 show the AMR DTX control. First, G.729A DTX control will be described with reference to FIG.
In G.729A, when VAD detects a change from a voice interval (VAD_flag = 1) to a non-voice interval (VAD_flag = 0), the first frame of the non-voice interval is set as an SID frame. The SID frame is created by generating CN information and quantizing CN information by the above-described method, and transmitted to the receiving side. In the non-voice interval, signal fluctuation is observed for each frame, only the frame in which the fluctuation is detected is set as an SID frame, and CN information is transmitted again. A frame determined not to change is set as a non-transmission frame, and information is not transmitted. In addition, the SID frame is limited to include at least two non-transmission frames. The change is detected by measuring the amount of change in CN information between the current frame and the last transmitted SID frame. As described above, in G.729A, the setting of the SID frame is adaptively performed with respect to the fluctuation of the non-voice signal.

  Next, AMR DTX control will be described with reference to FIGS. In AMR, as shown in FIG. 24, the SID frame setting method is basically periodically set every 8 frames unlike the adaptive control of G.729A. However, hangover control is performed as shown in FIG. 25 at the point of change to a non-voice section after a long voice section. Specifically, seven frames after the change point are set as speech sections regardless of the non-speech section (VAD_flag = 0), and normal speech coding processing is performed. This section is called hangover. This hangover is set when the number of frames (P-FRM) elapsed since the last SID frame was set is 23 frames or more. This prevents CN information at the change point (start point of the non-speech interval) from being obtained from the feature parameters of the speech interval (past 8 frames), and improves the sound quality at the change point from speech to non-speech. I can do it.

  Thereafter, the eighth frame is set as the first SID frame (SID_FIRST frame), but CN information is not transmitted in the SID_FIRST frame. This is because CN information can be generated from the decoded signal by the receiving decoder in the hangover period. After the SID_FIRST frame, the third frame is set as the SID_UPDATE frame, and CN information is transmitted for the first time here. In the subsequent non-voice section, a SID_UPDATA frame is set every 8 frames. The SID_UPDATA frame is created by the method described above and transmitted to the receiving side. Other frames are set as non-transmission frames and CN information is not transmitted.

  Also, as shown in FIG. 24, when the number of frames that have elapsed since the last SID frame was set is 23 frames or less, hangover control is not performed. In this case, the frame at the change point (the first frame of the non-voice interval) is set as SID_UPDATE, but the CN information transmitted last is transmitted again without calculating the CN information. As described above, the AMR DTX control transmits CN information by fixed control without performing adaptive control like G.729A. Therefore, appropriate hangover control is performed in consideration of the change point from voice to non-voice. As described above, the non-voice compression functions of G.729A and AMR have the same basic principle, but are different in CN information generation, quantization, and DTX control methods.

In prior art 1, FIG. 26 shows a configuration diagram when each communication system has a non-voice compression function. In the case of tandem connection, as described above, the speech code of encoding method 1 is once decoded into a reproduction signal, and is encoded again by encoding method 2. When each system has a non-voice compression function, as shown in FIG. 26, the VAD unit 3c of the code conversion unit 3 uses a reproduction signal encoded / decoded (information compression) by the encoding method 1 as a target for a voice / non-voice section. Judgment will be made. For this reason, the determination accuracy of the voice / non-speech section of the VAD unit 3c is lowered, and there is a problem that the head is cut off due to erroneous determination, and the sound quality may be deteriorated. For this reason, the encoding method 2 can be considered to treat all as speech sections, but this cannot perform optimum non-sound compression, and the effect of improving the transmission efficiency by the original non-sound compression is impaired. Furthermore, since CN information of encoding scheme 2 is obtained from comfort noise generated by decoder 1a of encoding scheme 1 in the non-speech section, CN information for generating noise similar to the input signal is used. Not necessarily optimal.
The prior art 2 is an excellent speech code conversion method with less sound quality degradation and transmission delay than the prior art 1 (tandem connection), but there is a problem that the non-speech compression function is not considered. In other words, in the prior art 2, since the input voice code is always assumed to be information encoded as a voice section, normal conversion operation can be performed when a non-voice compression function generates a SID frame or non-transmission frame. Absent.

Japanese Patent Application 2001-75427 (Japanese Patent Laid-Open No. 2002-202799)

An object of the present invention is to communicate non-speech communication systems between two different speech communication systems without receiving a CN code encoded by a non-speech encoding method on the transmission side into a CN signal. Conversion to a CN code according to the speech encoding method.
Another object of the present invention is to convert the CN code on the transmission side into the CN code on the reception side in consideration of the difference in frame length between the transmission side and the reception side and the difference in DTX control.
Another object of the present invention is to realize high-quality non-speech code conversion and speech code conversion in communication between two speech communication systems having different non-speech coding methods and speech coding methods.

  The present invention relates to a first speech code obtained by encoding a speech signal in a speech interval in a frame unit by a first speech encoding method, and a non-speech signal in a non-speech interval. The first non-speech code obtained by encoding the first non-speech code is mixed and transmitted from the transmission side, and the first speech code and the first non-speech code are respectively transmitted to the second non-speech code. The second speech code by the speech coding method and the second non-speech code by the second non-speech coding method are respectively converted, and the second speech code and the second non-speech code obtained by the conversion are respectively converted. This is a voice code conversion method in a voice communication system that mixes and transmits to the receiving side.

  In the first speech code conversion method of the present invention, a non-speech code is transmitted only in a predetermined frame in a non-speech interval, and a non-speech code is not transmitted in other frames. Adds frame type information indicating whether the frame, non-voice frame, or non-transmission frame that does not transmit a code, identifies which frame code is based on the frame type information, and in the case of a non-voice frame or non-transmission frame In consideration of the difference in frame length in the first and second non-speech coding methods and the difference in transmission control of the non-speech code, the first non-speech code is converted into the second non-speech code, In the case of a non-speech frame, if the first non-speech code to be converted cannot be obtained, the second non-speech code is obtained using the speech code of the past first speech frame, To convert non-voice code of the.

  In the second speech code conversion method of the present invention, a non-speech code is transmitted only in a predetermined frame in a non-speech interval, and a non-speech code is not transmitted in other frames. Adds frame type information indicating whether the frame, non-voice frame, or non-transmission frame that does not transmit a code, identifies which frame code is based on the frame type information, and in the case of a non-voice frame or non-transmission frame Takes into account the difference in frame length between the first and second non-speech coding methods and the difference in transmission control of the non-speech code, and converts the first non-speech code to the second non-speech code, When the first non-speech code is periodically converted to the second non-speech code, an average value obtained by averaging the received first non-speech code regardless of the presence or absence of the non-speech code First Used as a non-speech code, the second non-speech code periodically generated by.

  In the second speech code conversion method, the second non-speech coding method is configured such that at the change point from the speech interval to the non-speech interval, the continuous n frames including the frame at the change point are regarded as speech frames. When the code is transmitted, the dequantized values of a plurality of element codes obtained by dequantizing the non-speech code of the first non-speech frame and the dequantization of another predetermined or random element code Are generated in a continuous n-frame speech code of the second speech coding scheme, and a speech code of the second speech coding scheme for the n frames is output.

According to the present invention, in communication between two speech communication systems having different non-speech encoding methods, a non-speech code (CN code) encoded by a non-speech encoding method on the transmission side is not decoded into a CN signal. Can also be converted into a non-speech code (CN code) according to the non-speech encoding method on the receiving side, and high quality non-speech code conversion can be realized.
According to the present invention, a non-speech code (CN code) on the transmission side is converted to a non-speech code on the reception side without decoding into a non-speech signal in consideration of a difference in frame length between the transmission side and the reception side and a difference in DTX control. (CN code) can be converted, and high-quality non-voice code conversion can be realized.

It is a principle explanatory view of the present invention. 1 is a configuration diagram of a first embodiment of non-speech code conversion according to the present invention. FIG. It is a processing frame of G.729A and AMR. It is a frame type conversion control procedure from AMR to G.729A. It is a processing flow of an electric power correction part. It is a block diagram of 2nd Example of this invention. It is a block diagram of 3rd Example of this invention. It is conversion control explanatory drawing in an audio | voice area. It is conversion control explanatory drawing in a non-voice area. It is conversion control explanatory drawing (conversion control for every AMR8 frame) in a non-voice area. It is a block diagram of 4th Example of this invention. It is a block diagram of the audio | voice code | symbol conversion part in 4th Example. It is conversion control explanatory drawing in the audio | voice non-voice change point. It is conversion control explanatory drawing in the non-voice-> voice change point. FIG. 6 is an explanatory diagram of prior art 1 (tandem connection). It is explanatory drawing of the prior art 2. FIG. FIG. 10 is a more detailed explanatory diagram of the prior art 2. It is a conceptual diagram of a non-voice compression function. It is a principle diagram of a non-voice compression function. It is a processing block diagram of a non-voice compression function. It is a processing flow of a non-voice compression function. It is a non-speech code configuration diagram. It is a DTX control explanatory diagram of G.729A. It is AMR DTX control (at the time of non-hangover control) explanatory drawing. It is AMR DTX control (at the time of hangover control) explanatory drawing. It is a block diagram in the case of having a non-voice compression function in the prior art.

(A) Principle of the present invention FIG. 1 is an explanatory diagram of the principle of the present invention. As coding system 1 and coding system 2, coding based on CELP (Code Excited Linear Prediction) systems such as AMR and G.729A is used. It is assumed that each encoding method has the above-described non-voice compression function. In FIG. 1, when an input signal xin is input to the encoder 51a of the encoding method 1, the encoder 51a encodes the input signal and outputs code data bst1. At this time, the encoder 51a of the encoding scheme 1 performs the encoding process of the speech / non-speech section according to the determination result (VAD_flag) of the VAD unit 51b by the non-speech compression function. Therefore, the code data bst1 is composed of a voice code or a CN code. The code data bst1 includes frame type information Ftype1 indicating whether the frame is a voice frame or a SID frame (or a non-transmission frame).

The frame type detection unit 52 detects the frame type Ftype1 from the input code data bst1, and outputs the frame type information Ftype1 to the conversion control unit 53. The conversion control unit 53 identifies a speech section and a non-speech section based on the frame type information Ftype1, selects an appropriate conversion process according to the identification result, and switches the control switches S1 and S2.
If the frame type information Ftype1 is an SID frame, the non-voice code converting unit 60 is selected. In the non-voice code converting unit 60, first, the code data bst1 is input to the code separating unit 61. The code separation unit 61 separates the code data bst1 into one element CN code of the coding method. Each element CN codes are inputted to the CN code converting unit 62 ₁ ~62n, the CN code converting unit 62 ₁ ~62n Each element CN codes of encoding scheme 2 without decoding each element CN codes CN information Direct conversion. The code multiplexing unit 63 multiplexes the converted element CN codes, and inputs them to the decoder 54 of the encoding scheme 2 as the non-voice code bst2 of the encoding scheme 2.

If the frame type information Ftype1 is a non-transmission frame, no conversion process is performed. In this case, the non-voice code bst2 includes only the frame type information of the non-transmission frame.
When the frame type information Ftype1 is a speech frame, the speech code conversion unit 70 configured according to the prior art 1 or the prior art 2 is selected. The speech code conversion unit 70 performs speech code conversion processing according to the prior art 1 or the prior art 2, and outputs code data bst2 composed of the speech code of the encoding scheme 2.
As described above, since the frame type information Ftype1 is included in the speech code, the frame type can be identified by referring to the information. For this reason, the VAD part can be made unnecessary in the encoding method conversion part, and the erroneous determination of the speech section and the non-speech section can be eliminated.

In addition, since the CN code of encoding method 1 is directly converted into the CN code of encoding method 2 without returning it to the decoded signal (CN signal), it is possible to obtain optimum CN information for the input signal on the receiving side. it can. Thereby, natural background noise can be reproduced without impairing the effect of improving the transmission efficiency by the non-voice compression function.
Also, normal code conversion processing can be performed for SID frames and non-transmission frames in addition to voice frames. As a result, code conversion between different audio encoding systems having a non-audio compression function is possible.
Also, code conversion between two audio coding systems with different non-speech / sound compression functions is possible while maintaining the transmission efficiency improvement effect of the non-speech compression function and suppressing quality degradation and transmission delay. Therefore, the effect is great.

(B) First Embodiment FIG. 2 is a block diagram of a first embodiment of non-voice code conversion according to the present invention. An example in which AMR is used as the encoding method 1 and G.729A is used as the encoding method 2 is shown. Show. In FIG. 2, the NMR frame data, that is, the voice code bst1 (n) is input to the terminal 1 from an AMR encoder (not shown). The frame type detection unit 52 extracts the frame type information Ftype1 (n) included in the line data bst1 (n) and outputs it to the conversion control unit 53. There are four types of AMR frame type information Ftype (n): a voice frame (SPEECH), a SID frame (SID_FIRST), a SID frame (SID_UPDATE), and a non-transmission frame (NO_DATE) (see FIGS. 24 to 25). The non-voice code conversion unit 60 performs CN code conversion control according to the frame type information Ftype1 (n).

  In this CN code conversion control, it is necessary to consider the difference in frame length between AMR and G.729A. As shown in FIG. 3, the frame length of AMR is 20 ms, while the frame length of G.729A is 10 ms. Therefore, the conversion process converts one AMR frame (nth frame) as two G.729A frames (m, m + 1 frames). FIG. 4 shows the conversion control procedure for the frame type from AMR to G.729A. Each case will be described in turn below.

(a) When Ftype1 (n) = SPEECH As shown in FIG. 4 (a), when Ftype1 (n) = SPEECH, control switches S1 and S2 in FIG. The code conversion process is performed in the unit 70.
(b) When Ftype1 (n) = SID_UPDATE Next, a case where Ftype1 (n) = SID_UPDATE will be described. If one AMR frame is a SID_UPDATE frame as shown in FIG. 4 (b-1), the m-th frame of G.729A is set as an SID frame and CN code conversion processing is performed. That is, the switch in FIG. 2 is switched to the terminal 3, and the non-voice code converting unit 60 converts the AMR CN code bst1 (n) into the G.729A m-th frame CN code bst2 (m). Also, as described with reference to FIG. 23, in G.729A, SID frames are not set continuously, so the m + 1th frame of the next frame is set as a non-transmission frame. The operation of each CN element code converter (LSP converter 62 ₁ , frame power converter 62 ₂ ) will be described below.

  First, if the CN code bst1 (n) is input to the code separation unit 61, the code separation unit 61 separates the CN code bst1 (n) into the LSP code I_LSP1 (n) and the frame power code I_POW1 (n), and I_LSP1 ( n) is input to the LSP inverse quantizer 81 having the same quantization table as the AMR, and I_POW1 (n) is input to the frame power inverse quantizer 91 having the same quantization table as the AMR.

  The LSP inverse quantizer 81 inverse quantizes the input LSP code I_LSP1 (n) and outputs an AMR LSP parameter LSP1 (n). That is, the LSP inverse quantizer 81 inputs the LSP parameter LSP1 (n), which is the inverse quantization result, to the LSP quantizer 82 as the LSP parameter LSP2 (m) of the m.th frame of G.729A. The LSP quantizer 82 quantizes LSP2 (m) and outputs a G.729A LSP code I_LSP2 (m). Here, the quantization method of the LSP quantizer 82 is arbitrary, but the quantization table to be used is the same as that used in G.729A.

The frame power inverse quantizer 91 inversely quantizes the input frame power code I_POW1 (n) and outputs an AMR frame power parameter POW1 (n). Here, the frame power parameters of AMR and G.729A are different in signal areas when calculating frame power, as shown in Table 1, such that AMR is an input signal area and G.729A is an LPC residual signal area. Therefore, the frame power correction unit 92 corrects the AMR POW1 (n) in the LSP residual signal area according to the procedure described later so that it can be used in G.729A. As described above, the frame power correction unit 92 receives POW1 (n) as an input and outputs the G.729A frame power parameter POW2 (m). The frame power quantizer 93 quantizes POW2 (m) and outputs a G.729A frame power code I_POW2 (m). Here, the quantization method of the frame power quantizer 93 is arbitrary, but the quantization table to be used is the same as that used in G.729A.
The code multiplexing unit 63 multiplexes I_LSP2 (m) and I_POW2 (n), and outputs the result as a CN code bst2 (m) of G.729A. Since the (m + 1) th frame is set as a non-transmission frame, no conversion process is performed. Therefore, bst2 (m + 1) includes only frame type information representing a non-transmission frame.

(c) When Ftype1 (n) = NO_DATA Next, when the frame type information Ftype1 (n) = NO_DATA, both the m-th and m + 1-th frames are set as non-transmission frames as shown in FIG. 4C. In this case, conversion processing is not performed, and bst2 (m) and bst2 (m + 1) include only frame type information representing a non-transmission frame.

である。err(n) (n＝0，．．．，N₁-1，N₁：G.729Aのフレーム長（80サンプル）)はLPC残差信号であり、入力信号s(n)（n=0，．．．，N₁-1）とs(n)から求めたLPC係数α_i(i＝1，．．．，10)を用いて次式 (d) Frame power correction method
The logarithmic electricity POW1 of G.729A is calculated based on the following equation.
POW1 ＝ 20log ₁₀ E1 （1）
here,

It is. err (n) (n = 0,..., N ₁ -1, N ₁ : G.729A frame length (80 samples)) is an LPC residual signal, and the input signal s (n) (n = 0 , ..., N ₁ -1) and LPC coefficients α _i (i = 1, ..., 10) obtained from s (n)

により求められる。

Is required.

である。また、N2は、AMRのフレーム長（160サンプル）である。
式（2）、式（5）から明らかなように、G.729AとAMRでは電力E1、E2を算出するのに各々残差err(n)、入力信号s(n)と異なる領域の信号を用いている。したがって、その間を変換する電力修正部が必要となる。修正方法は任意であるが、例えば以下の方法が考えられる。 On the other hand, the logarithmic power POW2 of AMR is calculated based on the following equation.
POW2 ＝ log ₂ E2 （4）
here,

It is. N2 is the AMR frame length (160 samples).
As is clear from Equation (2) and Equation (5), G.729A and AMR calculate the power E1 and E2, respectively, using the residual err (n) and the signal in a different region from the input signal s (n). Used. Therefore, a power correction unit that converts between them is required. Although the correction method is arbitrary, for example, the following method can be considered.

-Modification from G.729A to AMR Figure 5 (a) shows the processing flow. First, the power E1 is obtained from the logarithmic power POW1 of G.729A.
E1 = 10 ^{(POW1 / 20)} (6)
Next, a pseudo LPC residual signal d_err (n) (n = 0,..., N ₁ −1) is generated by the following equation so that the power becomes E1.
d_err (n) = E1 · q (n) (7)
Here, q (n) (n = 0,..., N ₁ −1) is a random noise signal whose power is normalized to 1. d_err (n) is passed through the LPC synthesis filter to generate a pseudo signal (input signal region) d_s (n) (n = 0,..., N ₁ −1).

ここで、α_i(i＝1，．．．，10)はLSP逆量化値から求められたG.729AのLPC係数である。またd_s（-i）（i＝1，．．．，10）の初期値は0とする。ｄ_s(n)の電力を算出し、AMRの電力E2として用いる。したがって、AMRの対数電力POW2は、次式で求められる。

Here, α _i (i = 1,..., 10) is an LPC coefficient of G.729A obtained from the LSP dequantization value. The initial value of d_s (-i) (i = 1, ..., 10) is 0. The power of d_s (n) is calculated and used as the AMR power E2. Therefore, the logarithmic power POW2 of AMR is obtained by the following equation.

-Modification from AMR to G.729A Fig. 5 (b) shows the processing flow. First, electric power E2 is obtained from logarithmic electric power POW2 of AMR.
E2 = 2 ^POW2 (10)
A pseudo input signal d_s (n) (n = 0,..., N ₂ −1) with power E2 is generated from the following equation.
d_s (n) ＝ E2 ・ q (n) (11)
Here, q (n) is a random noise signal whose power is normalized to 1. d_s (n) is passed through an LPC inverse synthesis filter to generate a pseudo signal (LPC residual signal region) d_err (n) (n = 0,..., N ₂ −1).

ここで、α_i(i＝1，．．．，10)はLSP逆量子化値から求められたAMRのLPC係数である。また、ｄ_s(-i) （i＝1，．．．，10）の初期値は0とする。d_err（n）の電力を算出し、G.729Aの電力E1として用いる。したがって、G.729Aの対数電力POW1は、次式

で求められる。

Here, α _i (i = 1,..., 10) is an AMR LPC coefficient obtained from the LSP inverse quantization value. The initial value of d_s (-i) (i = 1,..., 10) is 0. The power of d_err (n) is calculated and used as the power E1 of G.729A. Therefore, the logarithmic power POW1 of G.729A is given by

Is required.

(e) Effects of First Embodiment As described above, according to the first embodiment, an LSP code and a frame power code, which are AMR CN codes, can be directly converted into G.729A CN codes. Also, by switching between the voice code conversion unit 70 and the non-voice code conversion unit 60, the non-voice compression is performed without decoding the code data (voice code, non-voice code) from the AMR having the non-voice compression function into the playback voice once. It can be normally converted into G.729A code data with a function.

(C) Second Embodiment FIG. 6 is a block diagram of a second embodiment of the present invention. The same reference numerals are given to the same parts as those of the first embodiment of FIG. In the second embodiment, when AMR is used as the encoding method 1 and G.729A is used as the encoding method 2 as in the first embodiment, the frame type of the AMR detected by the frame type detecting unit 52 is Ftype1 (n ) = Conversion processing when SID_FIRST is realized.
As shown in FIG. 4 (b-2), when one AMR frame is a SID_FIRST frame, as in the case of the SID_UPDATE frame in the first embodiment (FIG. 4 (b-1)), the G.729A Conversion processing can be performed by setting m frames as SID frames and m + 1 frame as non-transmission frames. However, as described in FIG. 25, it is necessary to consider that the CN code has not been transmitted by the hangover control in the SID_FIRST frame of AMR. That is, in the configuration of the first embodiment shown in FIG. 2, bst1 (n) is not sent, so LSP2 (m) and POW2 (m) that are CN parameters of G.729A cannot be obtained as they are.

Therefore, in the second embodiment, these are calculated using information of the past seven audio frames transmitted immediately before the SID_FIRST frame. The conversion process will be described below.
As described above, LSP2 (m) in the SID_FIRST frame is the LSP parameter OLD_LSP (1) for the past seven frames output from the LSP inverse quantization unit 4b ₁ (see FIG. 17) of the LSP code conversion unit 4b in the speech code conversion unit 70. ), (l = n-1, n-7) as an average value. Therefore, the LSP buffer unit 83 always holds the LSP parameters of the past seven frames with respect to the current frame, and the LSP average value calculation unit 84 stores the LSP parameters OLD_LSP (l), (l = n-1, n− for the past seven frames. Calculate and hold the average value of 7).

Similarly, POW2 (m) is calculated as an average value of frame power OLD_POW (l), (l = n−1, n−7) of the past seven frames. OLD_POW (l) is obtained as the frame power of the excitation signal EX (l) generated by the gain code conversion unit 4e (see FIG. 17) in the speech code conversion unit 70. Accordingly, the power calculation unit 94 calculates the frame power of the sound source signal EX (l), and the frame power buffer unit 95 always holds the frame power OLD_POW (l) of the past seven frames with respect to the current frame, and the power average value The calculation unit 96 calculates and holds the average value of the frame power OLD_POW (l) for the past seven frames.
If the frame type is not SID_FIRST in the non-voice interval, the LSP quantizer 82 and the frame power quantizer 93 are notified by the conversion control unit 53. Therefore, the LSP inverse quantizer 81 and the frame power inverse quantizer The LSP code I_LSP2 (m) and the frame power code I_POW2 (m) of G.729A are obtained and output using the LSP parameter and the frame power parameter output from the generator 91.

However, if the frame type is SID_FIRST in the non-voice interval, that is, if Ftype1 (n) = SID_FIRST, the conversion control unit 53 notifies that fact. Thereby, the LSP quantizer 82 and the frame power quantizer 93 use the average LSP parameter and the average frame power parameter for the past seven frames held in the LSP average value calculation unit 84 and the power average value calculation unit 96. Thus, the LSP code I_LSP2 (m) and the frame power code I_POW2 (m) of G.729A are obtained and output.
The code multiplexing unit 63 multiplexes the LSP code I_LSP2 (m) and the frame power code I_POW2 (m) and outputs them as bst2 (m).
Also, conversion processing is not performed in the (m + 1) th frame, and bst2 (m + 1) is transmitted including only frame type information representing a non-transmission frame.

  As described above, according to the second embodiment, even when the CN code to be converted is not obtained by the AMR hangover control, the CN parameter is obtained using the speech parameter of the past speech frame, and the G.729A CN code can be generated.

(D) Third Embodiment FIG. 7 shows a block diagram of a third embodiment of the present invention, and the same reference numerals are given to the same parts as those of the first embodiment. The third embodiment shows an example in which G.729A is used as the encoding method 1 and AMR is used as the encoding method 2.
In FIG. 7, m-th frame data, that is, a voice code bst1 (m) is input to a terminal 1 from a G.729A encoder (not shown). The frame type detection unit 52 extracts the frame type Ftype (m) included in bst1 (m) and outputs it to the conversion control unit 53. There are three types of F.type (m) of G.729A: a voice frame (SPEECH), a SID frame (SID), and a non-transmission frame (NO_DATA) (see FIG. 23). The conversion control unit 53 identifies the voice section and the non-voice section based on the frame type and switches the control switches S1 and S2.

  The non-voice code conversion unit 60 controls the CN code conversion process in accordance with the frame type information Ftype (m) in the non-voice section. Here, it is necessary to consider the difference in the frame length between AMR and G.729A as in the first embodiment. That is, two G.729A frames (m-th and m + 1-th frames) are converted as one AMR frame (n-th frame). Also, in the conversion from G.729A to AMR, it is necessary to control the conversion process in consideration of the difference in DTX control.

As shown in FIG. 8, when both Ftype1 (m) and Ftype1 (m + 1) are speech frames (SPEECH), the AMR nth frame is also set as a speech frame. That is, the control switches S1 and S2 in FIG. 7 are switched to terminals 2 and 4, and the speech code conversion unit 70 performs speech code conversion processing according to the related art 2.
Also, as shown in FIG. 9, when both Ftype1 (m) and Ftype1 (m + 1) are non-transmission frames (NO_DATA), the AMR nth frame is also set as a non-transmission frame and no conversion process is performed. . That is, the control switches S1 and S2 in FIG. 7 are switched to the terminals 3 and 5, and the code multiplexing unit 63 transmits only the frame type information of the non-transmission frame. Therefore, bst2 (n) includes only frame type information representing a non-transmission frame.

  Next, a CN code conversion method in a non-voice section as shown in FIG. 10 will be described. FIG. 10 shows a temporal flow of the CN code conversion method in the non-voice section. In the non-voice section, the switches S1 and S2 in FIG. 7 are switched to terminals 3 and 5, and the non-voice code conversion unit 60 performs a CN code conversion process. In this conversion process, it is necessary to consider the difference between DTX control between G.729A and AMR. Transmission control of SID frames in G.729A is adaptive, and SID frames are set irregularly according to changes in CN information (non-voice signal). On the other hand, in AMR, the SID frame (SID_UPDATA) is set periodically every 8 frames. Therefore, in the non-speech period, as shown in Fig. 10, regardless of the G.729A frame type (SID or NO_DATA) of the conversion source, every 8 frames (corresponding to 16 frames in G.729A) according to the AMR of the conversion destination To SID frame (SID_UPDATA). The other seven frames are converted so as to be in the non-transmission section (NO_DATA).

  Specifically, in the conversion of the AMR in the nth frame of FIG. 10 into the SID_UPDATA frame, the past 16 frames (m-14,..., M +) including the current frame (m, m + 1) are included. 1) The average value is obtained from the CN parameter of the SID frame received during (corresponding to 8 frames in AMR), and converted to the CN parameter of the SID_UPDATA frame of AMR. The conversion process will be described with reference to FIG.

  When the G.729A SID frame is received in the kth frame, the code separation unit 61 separates the CN code bst1 (k) into the LSP code I_LSP1 (k) and the frame power code I_POW1 (k), and I_LSP1 (k) Is input to the LSP inverse quantizer 81 having the same quantization table as G.729A, and I_POW1 (k) is input to the frame power inverse quantizer 91 having the same quantization table as G.729A. The LSP inverse quantizer 81 inversely quantizes the LSP code I_LSP1 (k) and outputs an LSP parameter LSP1 (k) of G.729A. The frame power dequantizer 91 dequantizes the frame power code I_POW1 (k) and outputs the G.729A frame power parameter POW1 (k).

As shown in Table 1, the G.729A and AMR frame power parameters have different signal areas when calculating frame power, such as G.729A is an LPC residual signal area and AMR is an input signal area. Therefore, the frame power correction unit 92 corrects the parameter POW1 (k) of the G.729A LSP residual signal region into the input signal region so that it can be used in AMR. As a result, the frame power correction unit 92 receives POW1 (k) and outputs the AMR frame power parameter POW2 (k).
The obtained LSP (k) and POW2 (k) are input to the buffer units 85 and 97, respectively. Here, k = m−14,..., M + 1, and each CN parameter of the SID frame received in the past 16 frames is held in the buffer units 85 and 97. Here, if there is no SID frame received in the past 16 frames, the CN parameter of the last received SID frame is used.

  The average value calculation units 86 and 98 calculate the average value of the buffer holding data and output it as the AMR CN parameters LSP2 (n) and POW2 (n). The LSP quantizer 82 quantizes LSP2 (n) and outputs an AMR LSP code I_LSP2 (n). Here, the quantization method of the LSP quantizer 82 is arbitrary, but the quantization table to be used is the same as that used in AMR. The frame power quantizer 93 quantizes POW2 (n) and outputs an AMR frame power code I_POW2 (n). Here, the quantization method of the frame power quantizer 93 is arbitrary, but the quantization table to be used is the same as that used in AMR. The code multiplexing unit 63 multiplexes I_LSP2 (n) and I_POW2 (n), adds frame type information (= U), and outputs the result as bst2 (n).

As described above, according to the third embodiment, when the CN code conversion processing is periodically performed in accordance with the DTX control of the conversion destination AMR regardless of the conversion source G.729A frame type in the non-voice section. The AMR CN code can be generated by using the average value of the CN parameters of G.729A received until the conversion process is performed as the CN parameter of AMR.
Also, by switching between the voice code converter and the CN code converter, G.729A code data (speech code, non-speech code) provided with a non-speech compression function is temporarily decoded without reproducing it into reproduced speech. Can be normally converted to AMR code data.

(E) Fourth Embodiment FIG. 11 is a block diagram of a fourth embodiment of the present invention. Components identical with those of the third embodiment of FIG. FIG. 12 is a block diagram of the speech code converter 70 in the fourth embodiment. In the fourth embodiment, as in the third embodiment, when G.729A2 is used as the coding method 1 and AMR is used as the coding method 2, the CN code conversion process at the change point from the voice section to the non-voice section is performed. Is realized.
FIG. 13 shows a temporal flow of the conversion control method. When the G.729A m-th frame is a voice frame and the m + 1-th frame is a SID frame, this is a transition point from a voice section to a non-voice section. In AMR, hangover control is performed at such change points. Note that if the number of elapsed frames in the AMR from the last conversion to the SID_UPDATA frame to the section change frame is 23 frames or less, the hangover control is not performed. Hereinafter, a case where the elapsed frame is larger than 23 frames and hangover control is performed will be described.

  When hangover control is performed, it is necessary to set 7 frames (nth,..., N + 7th frames) from the conversion point frame as audio frames regardless of non-audio frames. Therefore, as shown in FIG. 13 (a), the m + 1th frame to the m + 13th frame of G.729A are the non-voice frames (SID frame or non-transmission frame), but the AMR DTX of the conversion destination In accordance with the control, the conversion process is performed considering the voice frame. Hereinafter, the conversion process will be described with reference to FIGS. 11 and 12.

  In order to convert from a G.729A to an AMR speech frame at a conversion point from a speech segment to a non-speech segment, there is only a conversion process using the speech code conversion unit 70. However, since the G.729A side is a non-speech frame after the conversion point, the G.729A speech parameters (LSP, pitch lag, algebraic code, pitch gain, algebraic code gain) that are input to the speech code conversion unit 70 will remain as they are. Can't get. Therefore, as shown in FIG. 12, the LSP and the algebraic code gain are substituted with the CN parameters LSP1 (k), POW1 (k) (k <n) received last by the non-voice code converting unit 60, and other parameters ( The pitch lag lag (m), pitch gain Ga (m), and algebraic code code (m)) are arbitrarily set to such an extent that the pitch lag generating unit 101, algebraic code generating unit 102, and pitch gain generating unit 103 have no audible adverse effects. Generate. The generation method may be generated randomly or by a fixed value. However, it is desirable to set the minimum value (0.2) for the pitch gain.

At the time of switching from the voice section and the voice to the non-voice section, the voice code conversion unit 70 operates as follows.
In the speech section, the code separation unit 71 determines the LSP code ILSP1 (m), the pitch lag code I, from the input G.729A speech code. LAG1 (m), algebraic code I CODE1 (m), gain code I GAIN1 (m) is separated and input to the LSP inverse quantizer 72a, pitch lag inverse quantizer 73a, algebraic code inverse quantizer 74a, and gain inverse quantizer 75a, respectively. In addition, in the voice section, the switching units 77a to 77e are operated by the LSP inverse quantizer 72a, the pitch lag inverse quantizer 73a, the algebraic code inverse quantizer 74a, and the gain inverse quantizer 75a according to an instruction from the conversion control unit 53. Select an output.

The LSP dequantizer 72a dequantizes the G.729A LSP code and outputs an LSP dequantized value, and the LSP quantizer 72b uses the AMR LSP quantization table to output the LSP dequantized value. Quantize and LSP code I Outputs LSP2 (n). The pitch lag dequantizer 73a dequantizes the pitch lag code of G.729A and outputs a pitch lag dequantized value, and the pitch lag quantizer 73b uses the AMR pitch lag quantization table to output the pitch lag dequantized value. Quantize and pitch lag code I Output LAG2 (n). The algebraic code dequantizer 74a dequantizes the G.729A algebraic code and outputs an algebraic code dequantized value, and the algebraic code quantizer 74b converts the algebraic code dequantized value into the AMR algebraic code quantizer. Algebraic code I quantized using the quantization table Output CODE2 (n). The gain dequantizer 75a dequantizes the G.729A gain code and outputs a pitch gain dequantized value Ga and an algebraic gain dequantized value Gc. The pitch gain quantizer 75b The quantized value Ga is quantized using the pitch gain quantization table of AMR, and the pitch gain code I Output GAIN2a (n). Also, the algebraic gain quantizer 75c quantizes the algebraic gain inverse quantization value Gc using the AMR gain quantization table, and algebraic gain code I Output GAIN2c (n).

The code multiplexing unit 76 multiplexes the LSP code, pitch lag code, algebraic code, pitch gain code, and algebraic gain code output from each quantizer 72b to 75b, 75c, and adds frame type information (= S). Create and send AMR voice code.
In the speech section, the above operation is repeated, and the G.729A speech code is converted into an AMR speech code and output.
On the other hand, if hangover control is to be performed at the time of switching from speech to non-speech section, switching unit 77a obtains from the last received LSP code by non-speech code conversion unit 60 in accordance with an instruction from conversion control unit 53. The selected LSP parameter LSP1 (k) is selected and input to the LSP quantizer 72b. Further, the switching unit 77b selects the pitch lag parameter lag (m) generated from the pitch lag generation unit 101 and inputs it to the pitch lag quantizer 73b. The switching unit 77c selects the algebraic code parameter code (m) generated from the algebraic code generating unit 102 and inputs it to the algebraic code quantizer 74b. The switching unit 77d selects the pitch gain parameter Ga (m) generated from the pitch gain generation unit 103 and inputs it to the pitch gain quantizer 75b. Further, the switching unit 77e selects the frame power parameter POW1 (k) obtained from the frame power code IPOW1 (k) last received by the non-voice code converting unit 60 and inputs it to the algebraic gain quantizer 75c.

The LSP quantizer 72b quantizes the LSP parameter LSP1 (k) input from the non-voice code conversion unit 60 via the switching unit 77a using the AMR LSP quantization table, and performs the LSP code I Outputs LSP2 (n). The pitch lag quantizer 73b quantizes the pitch lag parameters input from the pitch lag generation unit 101 via the switching unit 77b using the AMR pitch lag quantization table to generate the pitch lag code I. Output LAG2 (n). The algebraic code quantizer 74b quantizes the algebraic code parameter input from the algebraic code generating unit 102 via the switching unit 77c using the AMR algebraic code quantization table, and algebraic code I Output CODE2 (n). The pitch gain quantizer 75b quantizes the pitch gain parameter input from the pitch gain generating unit 103 via the switching unit 77d using the pitch gain quantization table of AMR, and performs pitch gain code I. Output GAIN2a (n). The algebraic gain quantizer 75c quantizes the frame power parameter POW1 (k) input from the non-speech code converting unit 60 via the switching unit 77e using the AMR algebraic gain quantization table, and algebraic gain code I Output GAIN2c (n).

The code multiplexing unit 76 multiplexes the LSP code, pitch lag code, algebraic code, pitch gain code, and algebraic gain code output from each quantizer 72b to 75b, 75c, and adds frame type information (= S). Create and send AMR voice code.
At the transition point from the speech section to the non-speech section, the speech code conversion unit 70 repeats the above operation until sending the AMR speech code for 7 frames, and if the transmission of the 7 frames of speech code is completed, The voice code output is stopped until a voice segment is detected.

When the transmission of the speech code for 7 frames is completed, the switches S1 and S2 in FIG. 11 are switched to the terminals 3 and 5 side by the control of the conversion control unit 53. Thereafter, the CN code conversion processing by the non-speech code conversion unit 60 is performed. Done.
As shown in Figure 13 (a), the m + 14th and m + 15th frames (the n + 7th frame on the AMR side) after the hangover must be set as SID_FIRST frames in accordance with AMR DTX control. . However, transmission of the CN parameter is not necessary, and therefore the code multiplexing unit 63 outputs only the information indicating the frame type of SID_FIRST included in bst2 (m + 7). Thereafter, CN code conversion is performed as in the third embodiment of FIG.

The above is CN code conversion when hangover control is performed, but when the number of elapsed frames in AMR from the last conversion to SID_UPDATA frame to the change point frame is 23 frames or less, hangover There is no control. FIG. 13B shows a control method when such hangover control is not performed.
The m-th and m + 1-th frames, which are the boundary frames between the speech section and the non-speech section, are converted into AMR speech frames by the speech code conversion unit 70 in the same manner as at the time of hangover, and are output.

The next m + 2 and m + 3 frames are converted into SID_UPDATA frames.
For the frames after the (m + 4) th frame, the same method as the conversion method in the non-voice section described in the third embodiment is used.
Next, a CN code conversion method at a change point from a non-voice section to a voice section will be described. FIG. 14 shows a temporal flow of the conversion control method. When the m.th frame of G.729A is a non-voice frame (SID frame or non-transmission frame) and the m + 1th frame is a voice frame, this is a transition point from a non-voice section to a voice section. In this case, the AMR nth frame is converted as a voice frame in order to prevent the voice head from being cut off (the rising edge of the voice disappears). Therefore, the mth frame of G.729A converts a non-voice frame as a voice frame. As in the case of the hangover, the conversion method converts the voice code conversion unit 70 into an AMR voice frame and outputs it.

  As described above, according to the present embodiment, when it is necessary to convert a non-voice frame of G.729A into a voice frame of AMR at the transition point from the voice section to the non-voice section, the CN parameter of G.729A is set to AMR. It is possible to generate an AMR speech code by substituting as a speech parameter.

Supplementary Note (Supplementary Note 1) In a speech code conversion method for converting a first speech code obtained by encoding an input signal using a first speech coding method to a second speech code of a second speech coding method ,
Second speech coding without first decoding the first non-speech code obtained by coding the non-speech signal included in the input signal by the non-speech compression function of the first speech coding method Convert to the second non-voice code of the scheme,
A speech code conversion method characterized by:

(Supplementary note 2) In a speech code conversion method for converting a first speech code obtained by encoding an input signal by a first speech encoding method into a second speech code of a second speech encoding method,
Separating the first non-speech code obtained by encoding the non-speech signal included in the input signal by the non-speech compression function of the first speech encoding method into a plurality of first element codes;
Converting a plurality of first element codes into a plurality of second element codes constituting the second non-voice code;
Multiplex the second plurality of element codes obtained by the conversion to output a second non-voice code,
A speech code conversion method characterized by the above.

(Supplementary Note 3) The first element code is obtained by dividing a non-speech signal into frames each having a predetermined number of samples and analyzing a feature parameter representing a feature of the non-speech signal obtained by analysis for each frame. It is a code obtained by quantizing using a method-specific quantization table,
The second element code is a code obtained by quantizing the feature parameter using a quantization table unique to the second speech encoding method.
The speech code conversion method according to supplementary note 2, characterized by:
(Supplementary Note 4) The characteristic parameter is an LPC coefficient (linear prediction coefficient) representing an outline of the frequency characteristic of a non-speech signal and a frame signal power representing an amplitude characteristic of the non-speech signal.
The speech code conversion method according to supplementary note 3, characterized by:
(Supplementary note 5) In the conversion step, the first plurality of element codes are inversely quantized by an inverse quantizer having the same quantization table as that of the first speech encoding method,
Quantize the inverse quantized values of multiple element codes obtained by inverse quantization with a quantizer that has the same quantization table as the second speech encoding method, and convert it to the second multiple element codes. The speech code conversion method according to appendix 2 or appendix 3 or 4, characterized by the above.

(Supplementary Note 6) A first speech code obtained by encoding a speech signal in a speech interval in a frame unit by a first speech encoding method and a non-speech signal in a non-speech interval, with a certain number of samples of the input signal as a frame The first non-speech code obtained by encoding the first non-speech code is transmitted from the transmission side in a mixed manner, and the first non-speech code and the first non-speech code are respectively transmitted to the second non-speech code. The second speech code by the speech coding method and the second non-speech code by the second non-speech coding method are respectively converted, and the second speech code and the second non-speech code obtained by the conversion are converted. In a voice code conversion method in a voice communication system that mixes and transmits to the receiving side,
In a non-voice section, a non-voice code is transmitted only in a predetermined frame, and a non-voice code is not transmitted in other frames.
Add frame type information indicating the distinction between voice frames, non-voice frames, and non-transmission frames that do not transmit codes to the code information in units of frames,
Identify which frame code is based on frame type information,
In the case of non-speech frames and non-transmission frames, the first non-speech code is assigned to the first non-speech code in consideration of the difference in frame length between the first and second non-speech coding methods and the difference in transmission control of the non-speech code. Convert to 2 non-speech code,
A speech code conversion method characterized by the above.

(Appendix 7) (1) The first non-speech encoding method is a method of transmitting a non-speech code averaged for each predetermined number of frames in a non-speech interval and not transmitting a non-speech code in other frames. (2) The second non-speech encoding method transmits a non-speech code only in a frame in which the degree of change of a non-speech signal in a non-speech period is large, and does not transmit a non-speech code in other frames, (3) when the frame length of the first non-voice encoding scheme is twice the frame length of the second non-voice encoding scheme;
Converting the code information of the non-transmission frame in the first non-voice encoding scheme into the code information of two non-transmission frames in the second non-voice encoding scheme;
Converting the code information of the non-voice frame in the first non-voice coding method into two pieces of code information of the non-voice frame and code information of the non-transmission frame in the second non-voice coding method;
The speech code conversion method according to appendix 6, characterized in that:

(Appendix 8) When changing from a voice section to a non-voice section, the first non-voice coding method transmits a voice code by regarding the consecutive n frames including the frame at the change point as a voice frame, If the frame carries frame type information as the first non-voice frame that does not contain a non-voice code,
When the first non-speech frame in the first non-speech encoding method is detected, an inverse quantization value obtained by de-quantizing the speech code of the n previous speech frames in the first speech encoding method Averaging, quantizing the average value to obtain a non-speech code in a non-speech frame of the second non-speech coding scheme;
The speech code conversion method according to supplementary note 7, wherein

(Supplementary note 9) (1) The first non-speech coding method transmits a non-speech code only in a frame where the degree of change of a non-speech signal in a non-speech section is large, and transmits a non-speech code in other frames. And (2) the second non-speech encoding method transmits a non-speech code averaged every predetermined number of frames N in the non-speech interval, and (2) In other frames, a non-speech code is not transmitted. Further, (3) when the frame length of the first non-speech encoding method is half of the frame length of the second non-speech encoding method,
A frame for every N frames in the second non-speech coding method is obtained by averaging the inverse quantization values of the non-speech codes in the continuous 2 × N frames of the first non-speech coding method, and dequantizing the average value. Non-speech code
For frames other than every N frames, the code information of two consecutive frames of the first non-voice coding method is changed to the code information of one non-transmission frame of the second non-voice coding method regardless of the frame type. Convert,
The speech code conversion method according to appendix 6, characterized in that:

(Supplementary Note 10) When the second non-speech encoding method changes from a speech segment to a non-speech segment, the second non-speech encoding scheme transmits a speech code by regarding the consecutive n frames including the change point frame as a speech frame, If the frame carries frame type information as the first non-voice frame that does not contain a non-voice code,
Dequantizing the non-speech code of the first non-speech frame to generate a dequantized value of a plurality of element codes, and simultaneously generating a dequantized value of another predetermined or random element code;
The inverse quantization value of each element code of two consecutive frames is quantized using the quantization table of the second speech coding scheme and converted into a speech code for one frame of the second speech coding scheme,
After outputting the voice code of the second voice coding system for n frames, sending the frame type information of the first non-voice frame not including the non-voice code,
The speech code conversion method according to supplementary note 9, characterized by that.

(Supplementary Note 11) In a speech code conversion device that converts a first speech code obtained by encoding an input signal using a first speech encoding method to a second speech code of a second speech encoding method,
A code separation unit for separating a first non-voice code obtained by encoding a non-voice signal included in an input signal by the non-voice compression function of the first voice coding method into a plurality of first element codes;
An element code conversion unit that converts the first plurality of element codes into the second plurality of element codes constituting the second non-voice code;
A code multiplexing unit that multiplexes each second element code obtained by the conversion and outputs a second non-voice code;
A speech code conversion device comprising:

(Supplementary note 12) The first element code is obtained by dividing a non-speech signal into frames each having a predetermined number of samples and analyzing a feature parameter representing a feature of the non-speech signal obtained by analysis for each frame. It is a code obtained by quantizing using a method-specific quantization table,
The second element code is a code obtained by quantizing the feature parameter using a quantization table unique to the second speech encoding method.
The speech code conversion device according to appendix 11, characterized in that.
(Supplementary Note 13) The element code conversion unit includes:
An inverse quantizer that inversely quantizes each of the first element codes based on the same quantization table as the first speech encoding scheme;
A quantizer that quantizes the dequantized value of each element code obtained by the dequantization based on the same quantization table as the second speech encoding method and converts it into a second element code;
The speech code converter according to appendix 11 or 12, characterized by comprising:

(Supplementary Note 14) A first speech code obtained by encoding a speech signal in a speech interval in a frame unit with a first speech encoding method and a non-speech signal in a non-speech interval, with a certain number of samples of the input signal as a frame The first non-speech code obtained by encoding the first non-speech code is transmitted from the transmission side in a mixed manner, and the first non-speech code and the first non-speech code are respectively transmitted to the second non-speech code. The second speech code by the speech coding method and the second non-speech code by the second non-speech coding method are respectively converted, and the second speech code and the second non-speech code obtained by the conversion are converted. In a voice code conversion apparatus in a voice communication system for transmission to a receiving side,
A frame type identifying unit that identifies a non-transmission frame that does not transmit a non-speech code in a non-speech section based on frame type information added to the code information;
The first non-speech code in the non-speech frame is dequantized based on the same quantization table as the first non-speech encoding method, and the obtained dequantized value is the same as the second non-speech encoding method A non-speech code conversion unit that quantizes and converts to a second non-speech code based on a quantization table;
A conversion control unit for controlling the non-speech code conversion unit in consideration of a difference in frame length in the first and second non-speech encoding schemes and a difference in transmission control of the non-speech code;
A speech code conversion device comprising:

(Supplementary Note 15) (1) The first non-speech coding method is a method of transmitting a non-speech code averaged for each predetermined number of frames in a non-speech interval and not transmitting a non-speech code in other frames. (2) The second non-speech encoding method transmits a non-speech code only in a frame in which the degree of change of a non-speech signal in a non-speech period is large, and does not transmit a non-speech code in other frames, When the frame length of the first non-speech encoding method is twice the frame length of the second non-speech encoding method, and the non-speech code is not continuously transmitted. Non-speech code converter
The code information of the non-transmission frame in the first non-voice encoding method is converted into the code information of the two non-transmission frames in the second non-voice coding method, and the non-voice frame of the first non-voice coding method is converted. 15. The speech code conversion apparatus according to appendix 14, wherein the code information is converted into two pieces of code information of a non-voice frame and code information of a non-transmission frame in the second non-voice coding method.

(Supplementary Note 16) When the speech section changes to the non-speech section, the first non-speech coding scheme transmits the speech code by regarding the consecutive n frames including the change point frame as speech frames, When transmitting the frame type information as a first non-voice frame that does not include a non-voice code, the non-voice code conversion unit includes:
A buffer that holds a dequantized value obtained by dequantizing the speech code of the latest n speech frames in the first speech coding scheme;
An average value calculation unit that averages n dequantized values,
A quantizer for quantizing the average value when the first non-voice frame is detected;
16. The speech code conversion device according to appendix 15, wherein the speech code conversion device outputs a non-speech code in the second non-speech coding method based on an output of the quantizer.

(Supplementary Note 17) (1) The first non-speech coding method transmits a non-speech code only in a frame where the degree of change of a non-speech signal in a non-speech section is large, and transmits a non-speech code in other frames. And (2) the second non-speech encoding method transmits a non-speech code averaged every predetermined number of frames N in the non-speech interval, and (2) In other frames, a non-speech code is not transmitted, and (3) when the frame length of the first non-speech encoding method is half the frame length of the second non-speech encoding method, Non-speech code converter
A buffer that holds an inverse quantization value of each non-voice code in consecutive 2 × N frames of the first non-voice coding method;
An average value calculation unit for calculating the average value of the dequantized values held;
A quantizer that quantizes the average value and converts it into a non-voice code every N frames in the second non-voice coding method;
For frames other than every N frames, the code information of two consecutive frames of the first non-voice coding method is changed to the code information of one non-transmission frame of the second non-voice coding method regardless of the frame type. Means to convert,
15. A speech code conversion device according to supplementary note 14, characterized by comprising:

(Supplementary Note 18) When the speech section changes to the non-speech section, the second non-speech coding scheme transmits the speech code by regarding the consecutive n frames including the frame at the change point as speech frames, If the frame transmits frame type information as the first non-voice frame that does not contain a non-voice code, the non-voice code conversion unit
An inverse quantizer that dequantizes the non-speech code of the first non-speech frame to generate dequantized values of a plurality of element codes;
Means for generating dequantized values of a plurality of predetermined or random element codes, and using the quantization table of the second speech coding method for the dequantized values of the element codes of two consecutive frames Each of them is quantized and converted into a speech code for one frame of the second speech coding method and output, and after a speech code of the second speech coding method for n frames is output, a non-speech code is not included. Sending frame type information of the first non-voice frame;
The speech code conversion device according to supplementary note 17, characterized in that.

As described above, according to the present invention, in communication between two speech communication systems having different non-speech encoding methods, a non-speech code (CN code) encoded by the non-speech encoding method on the transmission side is decoded into a CN signal. Without conversion, it can be converted into a non-speech code (CN code) according to the non-speech encoding method on the receiving side, and high-quality non-speech code conversion can be realized.
Further, according to the present invention, the non-speech code (CN code) on the transmitting side is converted to the non-speech code on the receiving side without decoding into a non-speech signal in consideration of the difference in frame length between the transmitting side and the receiving side and the difference in DTX control. It can be converted into a voice code (CN code), and high-quality non-voice code conversion can be realized.

Further, according to the present invention, normal code conversion processing can be performed not only for voice frames but also for SID frames and non-transmission frames based on the non-voice compression function. As a result, it is possible to perform code conversion between speech coding methods having a non-speech compression function, which has been a problem in conventional speech code conversion units.
Further, according to the present invention, it is possible to perform speech code conversion between different communication systems while maintaining the transmission efficiency improvement effect of the non-speech compression function and further suppressing quality degradation and transmission delay. Most voice communication systems such as VoIP and mobile phone systems use a non-voice compression function, and the effect of the present invention is great.

51a Encoder for encoding system 1
51b VAD section
52 Frame type detector
53 Conversion controller
54 Coding method 2 decoder
60 Non-speech code converter
61 Code separator
62 _{1 to} 62n CN code converter
63 Code multiplexer
70 Voice code converter

Claims (3)

A first sample code obtained by encoding a speech signal in a speech section in a frame unit with a first speech coding scheme and a non-speech signal in a non-speech section as a first, with a certain number of samples of the input signal as a frame. A first non-speech code obtained by encoding with a non-speech encoding method is mixed and transmitted from the transmission side, and the first speech code and the first non-speech code are respectively transmitted to the second speech encoding method. Are converted into a second speech code by the second non-speech coding method and a second non-speech code by the second non-speech coding method, and the second speech code and the second non-speech code obtained by the conversion are mixedly received. In a speech code conversion method in a speech communication system that transmits to the side,
In a non-voice section, a non-voice code is transmitted only in a predetermined frame, and a non-voice code is not transmitted in other frames.
Add frame type information indicating the distinction between voice frames, non-voice frames, and non-transmission frames that do not transmit codes to the code information in units of frames,
Identify which frame code is based on frame type information,
In the case of non-speech frames and non-transmission frames, the first non-speech code is changed to the first non-speech code in consideration of the difference in frame length between the first and second non-speech coding methods and the difference in transmission control of the non-speech code. While converting to 2 non-speech code,
In the case of a non-speech frame, if the first non-speech code to be converted cannot be obtained, the second non-speech code is obtained using the speech code of the past first speech frame, and the second non-speech code is obtained. Convert to non-speech code,
A speech code conversion method characterized by the above. A first sample code obtained by encoding a speech signal in a speech section in a frame unit with a first speech coding scheme and a non-speech signal in a non-speech section as a first, with a certain number of samples of the input signal as a frame. A first non-speech code obtained by encoding with a non-speech encoding method is mixed and transmitted from the transmission side, and the first speech code and the first non-speech code are respectively transmitted to the second speech encoding method. Are converted into a second speech code by the second non-speech coding method and a second non-speech code by the second non-speech coding method, and the second speech code and the second non-speech code obtained by the conversion are mixedly received. In a speech code conversion method in a speech communication system that transmits to the side,
In a non-voice section, a non-voice code is transmitted only in a predetermined frame, and a non-voice code is not transmitted in other frames.
Add frame type information indicating the distinction between voice frames, non-voice frames, and non-transmission frames that do not transmit codes to the code information in units of frames,
Identify which frame code is based on frame type information,
In the case of non-speech frames and non-transmission frames, the first non-speech code is changed to the first non-speech code in consideration of the difference in frame length between the first and second non-speech coding methods and the difference in transmission control of the non-speech code. And convert it to 2 non-speech codes,
When the first non-speech code is periodically converted to the second non-speech code, an average value obtained by averaging the received first non-speech code regardless of the presence or absence of the non-speech code Use as a second non-speech code, thereby periodically generating a second non-speech code;
A speech code conversion method characterized by the above. When the second non-speech coding scheme is a scheme for transmitting a speech code by regarding consecutive n frames as a speech frame including a frame at the transition point at a transition point from a speech segment to a non-speech segment, A second speech code using a dequantized value of a plurality of element codes obtained by dequantizing a non-speech code of one non-speech frame, and a dequantized value of another predetermined or random element code Generated in a continuous n-frame speech code of the encoding method, and outputs a second speech encoding speech code for the n frames,
The speech code conversion method according to claim 2.

Images