JP5326714B2 - Band expanding apparatus, method and program, and quantization noise learning apparatus, method and program - Google Patents

Description

  The present invention relates to a bandwidth extension device, method and program, and a quantization noise learning device, method and program. For example, a bandwidth exceeding a bandwidth upper limit is generated and added to a speech signal whose bandwidth is limited. Applicable when expanding.

  Currently, telephones that are actively used for voice communication are limited in the frequency of voice that can be transmitted. Specifically, only a voice signal of 300 Hz to 3.4 kHz can be transmitted, and the call voice quality is not sufficient. In addition, ease of listening is also hindered.

  In order to solve such a problem, there is an attempt to expand the band of a sound signal whose band is limited to improve sound quality and ease of listening (see Patent Document 1 and Patent Document 2). In the conventional extension method, as shown in FIG. 5, the extension of the band is realized by generating and adding a signal of a band exceeding the band from the audio signal whose band is limited.

JP-A-9-90992 JP-A-8-123495

  According to the conventional audio band extending method, a relatively good extended audio signal can be obtained when the input audio signal is in a clean state (for example, a signal immediately after analog / digital conversion).

  However, when quantization noise due to encoding / decoding processing is superimposed on the input audio signal, the generated extended band is affected by the quantization noise, and the quality deterioration of the extended audio signal cannot be ignored. It has become a thing. The extension of the voice band is intended for high sound quality, but if the quality of the extension band deteriorates, the meaning of the extension is lost.

  Therefore, even if there is quantization noise due to encoding / decoding processing in the input signal, it is possible to obtain a high quality band extension signal, a band extension apparatus, method and program, and such a new band extension apparatus and method. And new technologies applicable to the program are desired.

In order to solve such a problem, the first aspect of the present invention provides (1) a bandwidth extension circuit main body that performs bandwidth extension in a bandwidth extension device that extends a bandwidth extension target signal output from a decoding circuit, and (2) A noise frequency characteristic storage means for storing frequency characteristic information of quantization noise specific to the encoding method related to the decoding circuit , formed by learning using a learning signal, and (3) output from the decoding circuit The quantized noise component mixed in the band extension target signal is reduced based on the frequency characteristic information of the quantization noise stored in the noise frequency characteristic storage means, And a quantization noise reduction means for inputting.

According to a second aspect of the present invention, there is provided a band extension method for band extension of a band extension target signal output from a decoding circuit. (1) The band extension circuit body executes band extension. (2) The noise frequency characteristic storage means is , Storing frequency characteristic information of quantization noise specific to the encoding method according to the decoding circuit , formed by learning using a learning signal, and (3) the quantization noise reduction means includes the decoding circuit The band expansion circuit reduces the quantization noise component mixed in from the band extension target signal output from the base station based on the information on the frequency characteristic of the quantization noise stored in the noise frequency characteristic storage means. It is characterized by having the main body input.

A third aspect of the present invention is a band expansion program for band-expanding a band expansion target signal obtained by decoding processing, comprising: (1) a band expansion circuit main body that performs band expansion; and (2) learning. Noise frequency characteristic storage means for storing frequency characteristic information of the quantization noise specific to the encoding scheme related to the decoding process , formed by learning using a signal, and (3) obtained by the decoding process The quantization noise component mixed from the band extension target signal is reduced based on the frequency characteristic information of the quantization noise stored in the noise frequency characteristic storage means, and input to the band extension circuit body. It functions as a quantization noise reduction means.

  According to a fourth aspect of the present invention, information on frequency characteristics of a band expansion circuit main body that performs band expansion on the band extension target signal output from the decoding circuit and quantization noise specific to the encoding method related to the decoding circuit is provided. The stored noise frequency characteristic storage means and the frequency of the quantization noise stored in the noise frequency characteristic storage means from the band extension target signal output from the decoding circuit, the quantized noise component mixed This is a quantization noise learning device that forms the frequency characteristic information of the quantization noise used by the band expansion device having the quantization noise reduction means that is reduced based on the characteristic information and input to the band expansion circuit main body. (1) learning signal generation means for outputting a learning signal; and (2) a quantization noise-mixed learning signal form that encodes the learning signal according to the encoding method and immediately decodes it. And (3) a subtracting means for subtracting the learning signal output from the quantization noise-mixed learning signal forming means from the learning signal output from the learning signal generating means to obtain a quantized noise signal. And (4) noise learning means for analyzing the frequency characteristics of the obtained quantized noise signal and forming information to be stored in the noise frequency characteristics storage means.

  According to a fifth aspect of the present invention, there is provided a bandwidth extension circuit main body for performing bandwidth extension on the bandwidth extension target signal output from the decoding circuit, and information on frequency characteristics of quantization noise specific to the coding system related to the decoding circuit The stored noise frequency characteristic storage means and the frequency of the quantization noise stored in the noise frequency characteristic storage means from the band extension target signal output from the decoding circuit, the quantized noise component mixed This is a quantization noise learning method for forming frequency characteristic information of quantization noise used by a band expansion apparatus having quantization noise reduction means that is reduced based on characteristic information and input to the band expansion circuit main body. (1) The learning signal generating means outputs a learning signal, and (2) the quantization noise-mixed learning signal forming means encodes the learning signal according to the above encoding method, and immediately (3) The subtracting unit subtracts the learning signal output from the quantization noise mixed learning signal forming unit from the learning signal output from the learning signal generating unit, and outputs a quantization noise signal. (4) The noise learning means analyzes the frequency characteristic of the obtained quantized noise signal and forms information to be stored in the noise frequency characteristic storage means.

  According to a sixth aspect of the present invention, there is provided a band extension circuit main body that performs band extension on a band extension target signal output from a decoding circuit, and information on frequency characteristics of quantization noise specific to the encoding method related to the decoding circuit. The stored noise frequency characteristic storage means and the frequency of the quantization noise stored in the noise frequency characteristic storage means from the band extension target signal output from the decoding circuit, the quantized noise component mixed This is a quantization noise learning program that forms frequency characteristic information of quantization noise used by a band expansion device that has a quantization noise reduction means that is reduced based on characteristic information and is input to the band expansion circuit main body. And (1) a learning signal generating means for outputting a learning signal, and (2) a quantum that encodes the learning signal according to the encoding method and immediately decodes the learning signal. (3) subtracting the learning signal output from the quantization noise mixing learning signal forming means from the learning signal output from the learning signal generating means, and quantizing (4) Analyzing the frequency characteristics of the obtained quantized noise signal and functioning as noise learning means for forming information to be stored in the noise frequency characteristic storage means .

  According to the present invention, it is possible to obtain a high-quality band extension signal even if the extension target signal includes quantization noise due to encoding / decoding processing.

It is a block diagram which shows the whole structure of the band expansion apparatus which concerns on 1st Embodiment. It is a block diagram which shows the example of an internal structure of the codec noise reduction circuit in 1st Embodiment. It is a block diagram which shows the whole structure of the band expansion apparatus which concerns on 2nd Embodiment. It is explanatory drawing which shows the structure of the learning parameter memory | storage part in 2nd Embodiment. It is explanatory drawing of band expansion.

(A) First Embodiment Hereinafter, a first embodiment of a band extending apparatus, method and program, and a quantization noise learning apparatus, method and program according to the present invention will be described in detail with reference to the drawings. The band extending apparatus and the quantization noise learning apparatus according to the first embodiment are used in apparatuses that handle audio signals (for example, telephone terminals and softphones). The apparatus according to the first embodiment is an embodiment in the case where there is only one type of codec that can be applied.

  In the first embodiment, after obtaining the characteristics of quantization noise generated in the encoding / decoding process in advance and performing noise reduction processing corresponding to the learning characteristics on the audio signal obtained by the decoding process, It is characterized by generating an extended band portion.

(A-1) Configuration of the First Embodiment FIG. 1 is a block diagram showing the overall configuration of the bandwidth extension apparatus according to the first embodiment. For example, even if the device equipped with the bandwidth extension device of the first embodiment is a soft phone and the bandwidth extension device is realized by a CPU and a program executed by the CPU, functionally, It can be represented by the functional block diagram of FIG.

  In FIG. 1, the band extending apparatus 100 according to the first embodiment mainly includes a quantization noise learning unit 110, a learning parameter storage unit 120, and a band extending unit 130. Note that the decoding circuit 101 that provides the decoded voice data to the band extension unit 130 represents a circuit that decodes the encoded voice data provided in the apparatus on which the band extension apparatus 100 of the first embodiment is mounted. Yes.

  The quantization noise learning unit 110 obtains learning parameters used by the band extending unit 130 in the band extending process, and stores them in the learning parameter storage unit 120.

  The quantization noise learning unit 110 includes a learning data generation unit 111, an encoding circuit 112, a decoding circuit 113, a subtractor 114, and a codec noise learning circuit 115.

  The learning data generator 111 generates and outputs a learning speech signal (digital signal; hereinafter referred to as learning speech data). For example, it is possible to apply one that generates audio data for learning by reading from a recording medium on which audio data is recorded. Further, for example, an arbitrary audio signal captured by a microphone and analog / digital converted may be used as a learning audio signal.

  The encoding circuit 112 encodes the speech data for learning according to the encoding method (type of codec) employed by the device on which the band extending device 100 according to the first embodiment is installed. The decoding circuit 113 decodes the encoded learning speech data output from the encoding circuit 112. The speech data for learning input to the encoding circuit 112 and the speech data for learning output from the decoding circuit 113 are quantized noise (hereinafter referred to as codec noise as appropriate) due to the encoding and decoding processes. ) Is different by the minute.

  The subtractor 114 subtracts the learning speech data output from the decoding circuit 113 from the learning speech data input to the encoding circuit 112 to thereby generate a code consisting of a quantization noise component accompanying encoding and decoding processing. Decoding noise data is obtained.

  The codec noise learning circuit 115 obtains information (learning parameters) representing the frequency characteristics of the codec noise data and stores the information in the learning parameter storage unit 120. The learning parameter may be a parameter that reflects a long-term average of the codec noise data. For example, the codec noise learning circuit 115 divides the codec noise data into frames for every predetermined time, performs FFT processing on each frame, and stores the average of FFT processing results for each frame as a learning parameter. A plurality of learning speech data may be prepared, and learning parameters obtained by applying each learning speech data may be averaged to obtain a learning parameter to be stored in the learning parameter storage unit 120. .

  The band extending unit 130 is configured to input the voice data obtained by the decoding process by the decoding circuit 101 during voice communication, and extends the band of the voice data. It should be noted that codec noise is mixed in the audio data output from the decoding circuit 101 through the encoding and decoding processes of the encoding circuit (not shown) and the decoding circuit 101 of the opposite device. The encoding circuit 112 of the quantization noise learning unit 110 described above is similar to the encoding circuit (not shown) of the opposite device, and the decoding circuit 113 of the quantization noise learning unit 110 described above is a decoding circuit. 101 is the same.

  The band extension unit 130 includes a codec noise reduction circuit 131, a band extension circuit 132, a high-pass circuit 133, an upsampling circuit 134, and an adder 135.

  The codec noise reduction circuit 131 reduces the codec noise of the speech data output from the decoding circuit 101 based on the learning parameters stored in the learning parameter storage unit 120. For example, frequency subtraction can be used to reduce the codec noise.

  FIG. 2 is a block diagram illustrating an internal configuration example of the codec noise reduction circuit 131. In FIG. 2, the codec noise reduction circuit 131 includes an FFT processing circuit 200, a frequency subtraction circuit 201, and an inverse FFT processing circuit 202.

  The FFT processing circuit 200 performs FFT on the audio data output from the decoding circuit 101. The frequency subtraction circuit 201 subtracts the learning parameter (FFT processing result for codec noise) stored in the learning parameter storage unit 120 from the FFT processing result for the speech data for each corresponding frequency. The inverse FFT processing circuit 202 performs an inverse FFT process on the FFT data output from the frequency subtraction circuit 201.

  The band extension circuit 132 performs band extension processing on the audio data output from the codec noise reduction circuit 131. Any existing method may be applied as a bandwidth extension method by the bandwidth extension circuit 132. Note that when the band expansion circuit 132 is internally divided into an expansion band processing unit and an existing band processing unit, audio data in which codec noise is reduced at least in the former case. Should be input.

  The high-pass circuit 133 extracts the frequency component of the extended portion from the audio data whose band has been extended. FIG. 1 shows a case where an existing circuit that outputs wideband audio data including an existing band part and an extended band part is applied as the band extension circuit 132, and therefore a high-pass circuit 133 is provided. . When a circuit that forms only the extended band portion is applied instead of the band extending circuit 132, the high-pass circuit 133 can be omitted.

  The upsampling circuit 134 increases the sampling rate of the audio data output from the decoding circuit 101 in accordance with a desired wide band.

  The adder 135 adds the up-sampled audio data portion of the existing band and the high-band audio data portion from the high-pass circuit 133 to obtain a band extension signal.

(A-2) Operation of the First Embodiment Next, the operation of the band extension apparatus 100 of the first embodiment is divided into a quantization noise learning operation (quantization noise learning method) and a band extension operation (band extension method). Will be described in the order.

  In the learning operation mode of codec noise (quantization noise), learning speech data is output from the learning data generation unit 111 and provided to the encoding circuit 112 and the subtracter 114.

  The learning speech data is encoded by the encoding circuit 112, immediately decoded by the decoding circuit 113, and given to the subtraction input terminal of the subtractor 114.

  The learning speech data input to the encoding circuit 112 is different from the learning speech data output from the decoding circuit 113 only in the amount of code decoding noise. It is taken out. Information (learning parameters) indicating the frequency characteristics of the extracted codec noise data is obtained by the codec noise learning circuit 115 and stored in the learning parameter storage unit 120.

  In the voice data band expansion operation mode, the voice data obtained by the decoding process by the decoding circuit 101 is supplied to the codec noise reduction circuit 131, and the codec noise mixed in the voice data is stored in the learning parameter storage unit 120. The stored learning parameters are utilized and reduced.

  The audio data whose codec noise has been reduced is expanded in band by the band extension circuit 132, and then only the frequency component of the desired extension portion is extracted by the high-pass circuit 133 and supplied to the adder 135. Also, the audio data obtained by the decoding process by the decoding circuit 101 is given to the adder 135 with the sampling rate increased by the upsampling circuit 134. As a result, the adder 135 adds the up-sampled audio data portion of the existing band and the high-frequency audio data portion from the high-pass circuit 133 to obtain and output a band extension signal.

(A-3) Effect of First Embodiment According to the first embodiment, after removing the quantization noise (code decoding noise) due to the encoding / decoding process, the band extension component is formed. Thus, it is difficult to perceive the codec noise in the extension band portion, and it is possible to obtain a band extension signal with high voice quality and good listening.

  Incidentally, conventionally, codec noise has been made difficult to perceive by masking processing in consideration of auditory characteristics. The existing band portion could be handled by such processing, but the generated extended band portion could not obtain an accurate auditory characteristic, so that effective masking processing could not be performed, and codec noise was not generated. It was perceived.

(B) Second Embodiment Next, a second embodiment of the band extending apparatus, method and program, and quantization noise learning apparatus, method and program according to the present invention will be described in detail with reference to the drawings. The apparatus of the second embodiment is an embodiment when there are two or more types of codecs that can be applied.

  FIG. 3 is a block diagram showing the overall configuration of the bandwidth extension apparatus according to the second embodiment. The same and corresponding parts as those in FIG. 1 according to the first embodiment are assigned the same and corresponding reference numerals. Show.

  In FIG. 3, the band extending apparatus 100A of the second embodiment is mainly composed of a quantization noise learning unit 110A, a learning parameter storage unit 120A, and a band extending unit 130A.

  The quantization noise learning unit 110A of the second embodiment includes a learning data generation unit 111, an encoding circuit 112A, a decoding circuit 113A, a subtractor 114, a codec noise learning circuit 115, and a learning codec type setting unit 116.

  In the case of the second embodiment, each of the encoding circuit 112A and the decoding circuit 113A has a variable configuration capable of executing encoding and decoding according to the codec type (encoding method) instructed by the learning codec type setting unit 116. Is. If the sampling rate of the learning data varies depending on the codec type, the learning data generation unit 111 also outputs the learning data corresponding to the codec type instructed by the learning codec type setting unit 116. It is necessary to apply the configuration.

  The learning codec type setting unit 116 newly provided in the second embodiment causes the encoding circuit 112A and the decoding circuit 113A to sequentially set the codec types applicable to the apparatus, and learns each codec type. Each unit is controlled so that parameters are sequentially formed and stored in the learning parameter storage unit 120A.

  FIG. 4 is an explanatory diagram showing the configuration of the learning parameter storage unit 120A. The learning parameter storage unit 120A of the second embodiment has a configuration in which learning parameters are stored for each codec type.

  The band extension unit 130A of the second embodiment includes a codec noise reduction circuit 131A, a band extension circuit 132, a high-pass circuit 133, an upsampling circuit 134, an adder 135, and an applied codec type setting unit 136.

  The applied codec type setting unit 136 newly provided in the second embodiment sets the codec type determined to be applied in the negotiation executed before the start of voice communication in the decoding circuit 101A having a variable configuration. In addition, the codec noise reduction circuit 131A functions as an instruction for the type of codec for extracting the learning parameter from the learning parameter storage unit 120A. If the sampling rate of the audio data varies depending on the codec type, the applied codec type setting unit 136 also applies to the band extension circuit 132, the high-pass circuit 133, the upsampling circuit 134, the adder 135, and the like. Informs or indicates the codec type or sampling rate.

  As described above, the second embodiment is configured to be compatible with a plurality of types of codecs, and the operation when the type of codec to be learned is set is the same as that of the first embodiment. The operation after the type of codec is determined is the same as in the first embodiment, and a detailed description of the operation is omitted.

  According to the second embodiment, the apparatus supports a plurality of types of codecs, and the effect of the first embodiment described above can be achieved regardless of which codec is selected during communication.

(C) Other Embodiments In each of the above embodiments, the same apparatus is also included in the quantization noise learning unit in addition to the learning parameter storage unit and the band extension unit. However, only the learning parameter storage unit and the band extension unit are included. It is also possible to construct the apparatus so as not to include the quantization noise learning unit. In this case, the learning parameter obtained by the external device is set in the learning parameter storage unit.

  In each of the above embodiments, the input data to the upsampling circuit is obtained from the input side of the codec noise reduction circuit. It may be obtained from the output side.

  In each of the above embodiments, the band extension target is audio data, but the present invention can also be applied to the case where the band extension target is acoustic data.

100, 100A ... band extension device,
110, 110A ... Quantization noise learning unit,
111 ... Data generator for learning, 112, 112A ... Encoding circuit,
113, 113A ... decoding circuit, 114 ... subtractor,
115: Coded noise learning circuit, 116: Learning codec type setting unit,
120, 120A ... learning parameter storage unit,
130, 130A ... bandwidth extension unit,
131, 131A ... Coded noise reduction circuit, 132 ... Band extension circuit,
133 ... high-pass circuit, 134 ... upsampling circuit,
135... Adder, 136... Applicable codec type setting unit.

Claims (9)

In a bandwidth extension device that extends the bandwidth extension target signal output from the decoding circuit,
A bandwidth extension circuit body for performing bandwidth extension;
A noise frequency characteristic storage means for storing information on the frequency characteristic of quantization noise specific to the encoding method according to the decoding circuit , formed by learning using a learning signal ;
From the band extension target signal output from the decoding circuit, the mixed quantization noise component is reduced based on the frequency characteristic information of the quantization noise stored in the noise frequency characteristic storage unit, and A bandwidth expansion apparatus comprising: quantization noise reduction means for input to a bandwidth expansion circuit body. A training signal generating means for outputting said training signal,
Encoding the learning signal according to the above encoding method, and immediately decoding the quantization noise mixed learning signal forming means,
Subtracting means for subtracting the learning signal output from the quantization noise mixed learning signal forming means from the learning signal output from the learning signal generating means to obtain a quantized noise signal;
The band extending apparatus according to claim 1, further comprising: a noise learning unit that analyzes frequency characteristics of the obtained quantized noise signal and forms information to be stored in the noise frequency characteristic storage unit. In a bandwidth extension method for extending a bandwidth extension target signal output from a decoding circuit,
The bandwidth extension circuit body performs bandwidth extension,
The noise frequency characteristic storage means stores the frequency characteristic information of the quantization noise specific to the encoding system related to the decoding circuit , formed by learning using the learning signal ,
The quantization noise reduction means converts the quantization noise component mixed from the band extension target signal output from the decoding circuit into information on the frequency characteristics of the quantization noise stored in the noise frequency characteristic storage means. A bandwidth expansion method comprising: reducing the bandwidth based on the bandwidth and inputting the bandwidth to the bandwidth expansion circuit body. The learning signal generating means outputs a learning signal,
The quantization noise-mixed learning signal forming means encodes the learning signal according to the above encoding method, and immediately decodes it,
The subtracting means subtracts the learning signal output from the quantization noise mixed learning signal forming means from the learning signal output from the learning signal generating means, to obtain a quantization noise signal,
The band extending method according to claim 3, wherein the noise learning unit analyzes the frequency characteristic of the obtained quantized noise signal and forms information to be stored in the noise frequency characteristic storage unit. A bandwidth extension program for extending the bandwidth extension target signal obtained by the decoding process,
Computer
A bandwidth extension circuit body for performing bandwidth extension;
A noise frequency characteristic storage means for storing information on frequency characteristics of quantization noise specific to the encoding method according to the decoding process , formed by learning using a learning signal ;
From the band extension target signal obtained by the decoding process, the mixed quantization noise component is reduced based on the frequency characteristic information of the quantization noise stored in the noise frequency characteristic storage means, and A bandwidth expansion program that functions as a quantization noise reduction means that is input to a bandwidth expansion circuit body. Computer
Learning signal generating means for outputting a learning signal;
Encoding the learning signal according to the above encoding method, and immediately decoding the quantization noise mixed learning signal forming means,
Subtracting means for subtracting the learning signal output from the quantization noise mixed learning signal forming means from the learning signal output from the learning signal generating means to obtain a quantized noise signal;
6. The bandwidth expansion program according to claim 5, wherein the bandwidth expansion program functions as noise learning means for analyzing the frequency characteristics of the obtained quantized noise signal and forming information to be stored in the noise frequency characteristics storage means. . A noise frequency characteristic that stores information on the frequency characteristic of a quantization noise peculiar to a coding system related to the above-mentioned decoding circuit and a band expansion circuit main body that performs band expansion on the band extension target signal output from the decoding circuit The quantization noise component mixed in the storage means and the band extension target signal output from the decoding circuit is reduced based on the frequency characteristic information of the quantization noise stored in the noise frequency characteristic storage means. A quantization noise learning device that forms information on frequency characteristics of quantization noise used by a bandwidth expansion device having a quantization noise reduction means that is input to the bandwidth expansion circuit body,
Learning signal generating means for outputting a learning signal;
Encoding the learning signal according to the above encoding method, and immediately decoding the quantization noise mixed learning signal forming means,
Subtracting means for subtracting the learning signal output from the quantization noise mixed learning signal forming means from the learning signal output from the learning signal generating means to obtain a quantized noise signal;
A quantization noise learning device comprising: noise learning means for analyzing frequency characteristics of the obtained quantized noise signal and forming information to be stored in the noise frequency characteristic storage means. A noise frequency characteristic that stores information on the frequency characteristic of a quantization noise peculiar to a coding system related to the above-mentioned decoding circuit and a band expansion circuit main body that performs band expansion on the band extension target signal output from the decoding circuit The quantization noise component mixed in the storage means and the band extension target signal output from the decoding circuit is reduced based on the frequency characteristic information of the quantization noise stored in the noise frequency characteristic storage means. A quantization noise learning method for forming frequency noise information of a quantization noise used by a band expansion device having a quantization noise reduction means to be input to the band expansion circuit body,
The learning signal generating means outputs a learning signal,
The quantization noise-mixed learning signal forming means encodes the learning signal according to the above encoding method, and immediately decodes it,
The subtracting means subtracts the learning signal output from the quantization noise mixed learning signal forming means from the learning signal output from the learning signal generating means, to obtain a quantization noise signal,
A noise learning means analyzes the frequency characteristic of the obtained quantized noise signal and forms information to be stored in the noise frequency characteristic storage means. A noise frequency characteristic that stores information on the frequency characteristic of a quantization noise peculiar to a coding system related to the above-mentioned decoding circuit and a band expansion circuit main body that performs band expansion on the band extension target signal output from the decoding circuit The quantization noise component mixed in the storage means and the band extension target signal output from the decoding circuit is reduced based on the frequency characteristic information of the quantization noise stored in the noise frequency characteristic storage means. A quantization noise learning program for forming information on frequency characteristics of quantization noise used by a band expansion device having quantization noise reduction means to be input to the band expansion circuit body,
Computer
Learning signal generating means for outputting a learning signal;
Encoding the learning signal according to the above encoding method, and immediately decoding the quantization noise mixed learning signal forming means,
Subtracting means for subtracting the learning signal output from the quantization noise mixed learning signal forming means from the learning signal output from the learning signal generating means to obtain a quantized noise signal;
A quantization noise learning program which analyzes the frequency characteristics of the obtained quantized noise signal and functions as noise learning means for forming information to be stored in the noise frequency characteristic storage means.

Images