JP5336942B2 - Encoding method, decoding method, encoder, decoder, program - Google Patents

Description

  The present invention relates to an encoding method and a decoding method for an acoustic signal, and particularly to an encoding method and a decoding method for vector quantization in multiple stages.

  Vector quantization is an effective means for quantizing a signal sequence with 1 bit or less per sample. This method selects an input vector and a vector with the least distortion from the codebook and transmits the number (hereinafter referred to as “index”). A codebook suitable for the quantization target is created. If this is done, quantization with small quantization distortion is possible. However, since the amount of codebook memory and the amount of computation for calculating distortion increase with an exponential function of the number of quantization bits, it is difficult to realize quantization when the number of quantization bits is large. One technique for solving this is multistage vector quantization (Patent Document 1). Further, in recent years, as an efficient vector quantization method with little quantization noise, for example, the frequency component is within a predetermined number of quantization bits such as the Spherical Vector Quantization (SVQ) method (Non-patent Document 1). A vector quantization method that stands as a pulse is widely used.

  FIG. 1 is a diagram illustrating a functional configuration example of an encoder and a decoder using a multistage vector quantization method. FIG. 2 is a diagram showing a processing flow of the encoder and the decoder. FIG. 2A shows a processing flow of the encoder, and FIG. 2B shows a processing flow of the decoder. In multi-stage vector quantization, quantizers are connected in multiple stages, the first stage first quantizer operates on the input signal, and the subsequent stages operate on errors between the input signal and the quantized output up to the previous stage. In this way, it is possible to perform vector quantization while keeping the memory amount and calculation amount within a practical range. The number of stages of multistage vector quantization in the figure is 2 for the sake of clarity, but a configuration using two or more stages may be used. As a quantization method, a method (Non-Patent Document 1) for quantizing a component converted into a frequency domain is used as an example.

  The encoder 100 includes a frequency domain transform unit 101, a normalized value calculation unit 102, a normalized value quantization unit 103, a first vector quantization unit 104, an error calculation unit 105, and a second vector quantization unit 106. The decoder 100 ′ includes a normalized value decoding unit 107, a first vector decoding unit 108, a second vector decoding unit 109, an error correction unit 110, and a time domain conversion unit 111.

  The frequency domain transform unit 101 receives an input signal x (n) as an input, and converts it into a frequency domain signal X (k) that is a signal indicating a frequency component at L points for every L input signals x (n) (for each frame). Conversion is performed (S101). Here, n represents a signal number (discrete time number) in the time domain, and k represents a signal number (discrete frequency number) in the frequency domain. L is, for example, 64 or 80. As a frequency domain transform method, for example, there is MDCT (Modified Discrete Cosine Transform).

The normalized value calculation unit 102 receives the frequency domain signal X (k) as an input and responds to the L point or each subband (a frequency band obtained by further dividing the L point, for example, one subband is formed by 8 points). The square root of the average power of the frequency domain signal X (k) is output as the normalized value X ^- ₀ of the frequency domain signal X (k). Assuming that the normalized value of point L is X ^- ₀ , for example, X ^- ₀ is calculated as in the following equation (S102).

In the following description, the case of using the normalization value X ^- ₀ of the L point instead of the subband normalization will be described.
Normalization value quantizer 103, the normalized value X calculated by the normalization value calculation section 102 ^- as input _0, normalized value X ^- ₀ quantizes the normalization value quantization index C _S, normal quantized normalized values corresponding to reduction value quantization index _{C S} X ^- outputting a (S103).

The first vector quantization unit 104 receives the frequency domain signal X (k) and the quantized normalized value X ⁻ and divides the frequency domain signal X (k) by the quantized normalized value X ⁻ or an inverse number. Is normalized by multiplying to obtain a normalized frequency domain signal. Then, vector quantization of the normalized frequency domain signal more samples collectively, quantized first vector quantization index C ₁ is the index of the quantization representative vectors, the signal corresponding to the first vector quantization index C ₁ normalized value X ^- seeking first quantized signal X ^ (k) by multiplying the outputs first and vector quantization index _{C 1} first quantized signal X ^ a (k) (S104).

The error calculation unit 105 receives the frequency domain signal X (k) and the first quantized signal X ^ (k) as input, and the error between the frequency domain signal X (k) and the first quantized signal X ^ (k). For example,
E (k) = X (k) -X ^ (k) (2)
And an error signal E (k) is output (S105).

Second vector quantization section 106 receives error signal E (k) and quantized normalized value X ⁻ as input, and divides error signal E (k) by quantized normalized value X ⁻ or multiplies the inverse. To obtain a normalized error signal. Then, the normalized error signal a plurality of samples collectively by vector quantization, and outputs the index of the quantization representative vector as a second vector quantization index C ₂ (S106).

Encoder 100, a code included first vector quantization index C ₁ and the second vector quantization index C ₂ and the normalized value quantization index C _S, and sends to the decoder 100 '. The decoder 100 ′ performs the following processing.
The normalized value decoding unit 107 receives the normalized value quantization index C _S as input, obtains a decoded quantized normalized value X ⁻ corresponding to the normalized value quantized index C _S , and decodes the quantized normalized value X ^−. Is output (S107).

The first vector decoding unit 108 receives the first vector quantization index C ₁ and the decoded quantization normalized value X ⁻ as input, decodes the first vector quantization index C ₁ , and normalizes the normalized signal in units of frames. determined as the first decoded quantized signal of normalized first decoded quantized signal to the decoding quantized normalization value X ^- inversely normalized by multiplying the outputs as a first decoded quantized signal X ^ (k) (S108).

Second vector decoding section 109 receives second vector quantization index C ₂ and decoded quantization normalized value X ⁻ as inputs, decodes second vector quantization index C ₂ , and normalizes the normalized signal as a normalization error. calculated as the signal, a normalized error signal to the decoding quantized normalization value X ^- inversely normalized by multiplying the outputs as a decoded quantization error signal E ^ (k) (S109) .

The error correction unit 110 receives the first decoded quantized signal X ^ (k) and the decoded quantized error signal E ^ (k) as input, and converts the decoded signal Z (k) into, for example,
Z (k) = X ^ (k) + E ^ (k) (3)
Thus, the decoded signal Z (k) is output (S110).
The time domain conversion unit 111 receives the decoded signal Z (k) as an input, performs time domain conversion using, for example, inverse MDCT for the number of frame points L, and outputs an output signal z (n) for the number of frame points L (S111). ).

Japanese Patent No. 2767474

ITU-T Recommendation G.729.1, SERIES G: TRANSMISSION SYSTEMS AND MEDIA, DIGITAL SYSTEMS AND NETWORKS, Digital terminal equipments-Coding of analogue signals by methods other than PCM, G.729-based embedded variable bit-rate coder: An 8- 32 kbit / s scalable wideband coder bitstream interoperable with G.729, 05/2006.

  In the above-described encoder 100 and decoder 100 ′, if the number of bits necessary to quantize the frequency domain signal is insufficient, the frequency component that should exist in the input signal does not exist in the output signal (frequency Component deficiency) may occur.

  However, the encoder 100 and the decoder 100 ′ have the first quantized signal regardless of whether the value of the first quantized signal (the value of the quantized signal obtained by the preceding vector quantization) is 0 or other than 0. A signal indicating an error between the quantized signal and the signal to be encoded is a target of vector quantization in the subsequent stage. In other words, the conventional technology has a special device for eliminating noise that humans feel sensitive, such as musical noise (noise that humans often feel when the presence or absence of a certain frequency component changes over time). There is no. Therefore, even if the power of error between the original signal and the decoded signal is small in numerical value, it is not an encoding that focuses on errors that are sensitive to human senses, so there is a problem that musical noise tends to remain. is there.

  The present invention has been made in view of such a problem, and even when the number of bits necessary for multistage quantization of a signal to be encoded is insufficient and a signal loss occurs after decoding. An object of the present invention is to provide a multistage vector quantization method capable of reducing musical noise by second-stage and subsequent vector quantization.

  The encoding method of the present invention includes at least a first vector quantization step, an extraction step, and a second vector quantization step. In the first vector quantization step, a signal sample to be encoded for each predetermined number of samples (hereinafter referred to as “first encoding target signal”) or a signal corresponding to the first encoding target signal is collected into a plurality of samples and a vector quantum is collected. To obtain a first vector quantization index and a first quantized signal. As the “first encoding target signal”, an acoustic signal converted into the frequency domain, an acoustic signal in a specific frequency band (for example, 8 kHz to 14 kHz) in the acoustic signal in the frequency domain, or divided into frames There may be time-domain acoustic signals, but it is not necessary to limit them. Further, the first vector quantization step does not need to be a step of performing the first vector quantization on the input signal, and the signal after performing several vector quantizations may be used as the first encoding target signal. Good. The “signal corresponding to the first encoding target signal” is not the first encoding target signal itself, but a signal corresponding to the first encoding target signal on a one-to-one basis (returning to the first encoding target signal) Signal). For example, there is a signal obtained by normalizing the first encoding target signal. In the extraction step, a second encoding target signal is generated using only the first encoding target signal corresponding to the first quantized signal having a value of 0. In the second vector quantization step, the second encoding target signal or a signal corresponding to the second encoding target signal is vector-quantized to obtain a second vector quantization index. The “second encoding target signal” is a signal generated by extracting only the first encoding target signal whose value of the first quantized signal has become 0 as a result of the first vector quantization. Further, the “signal corresponding to the second encoding target signal” is not the second encoding target signal itself, but is a signal corresponding to the second encoding target signal on a one-to-one basis (in the second encoding target signal). A signal that can be returned). For example, there is a signal obtained by normalizing the second encoding target signal.

  The decoding method of the present invention includes a first vector decoding step, an M value calculation step, a second vector decoding step, and a reconstruction step. In the first vector decoding step, a first decoded quantized signal is obtained from a signal corresponding to the first vector quantization index. In the M value calculation step, the number M of first decoded quantized signals having a value of 0 is obtained. The second vector decoding step obtains a second decoded quantized signal from the signal corresponding to the second vector quantization index. The reconstruction step uses the value of the first decoded quantized signal as the decoded signal value when the value of the first decoded quantized signal is non-zero, and performs the second decoding when the value of the first decoded quantized signal is zero. Let the value of the quantized signal be the value of the decoded signal.

  According to the encoding method and the decoding method of the present invention, the second encoding target signal does not include the first encoding target signal corresponding to the first quantized signal having a value other than 0, and the value is Only the first encoding target signal corresponding to the first quantized signal of 0 is included. That is, the second vector quantization step is a process of adding a value other than 0 to the first encoding target signal whose value has become 0 in the first vector quantization. Therefore, the second code is a code for reducing the error between the first encoding target signal and the first quantized signal while avoiding the presence or absence of the frequency component for each frame that is one of the causes of musical noise. A vector quantization index can be determined. Therefore, efficient vector quantization with a limited number of bits can be realized.

The figure which shows the function structural example of the encoder and decoder using the conventional multistage vector quantization method. The figure which shows the processing flow of the conventional encoder and decoder. The figure for demonstrating the principle of this invention. FIG. 3 is a diagram illustrating a configuration example of an encoder and a decoder according to the first embodiment. The figure which shows the example of a processing flow of the encoder and decoder of Example 1. FIG. The figure which shows the example of the processing flow for calculating | requiring the 2nd encoding object signal E (m) and the number M. The figure which shows the example of the processing flow which calculates | requires the number M of the 1st decoding quantization signal X ^ (k) whose value is 0. The figure which shows the example of the processing flow which reconfigure | reconstructs the decoding signal Z (k). The figure which shows the structural example of the encoder of Example 2 and a modification of Example 2 and a decoder. The figure which shows the example of a processing flow of the encoder of Example 2 and a modification of Example 2 and a decoder. It illustrates an example of a processing flow for determining the ^- recalculation normalized value E of Example 2. It illustrates another example of processing flow for determining the ^- recalculation normalized value E of Example 2. It illustrates an example of a processing flow for determining the ^- recalculation normalized value E of Example 2 Modification. It illustrates another example of processing flow for determining the ^- recalculation normalized value E of Example 2 Modification. The figure which shows the experimental result when only audio | voice is objected. The figure which shows the experimental result with respect to the audio | voice when music is flowing in the background. The figure which shows the experimental result with respect to the audio | voice when noise is contained.

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

Principle FIG. 3 is a diagram for explaining the principle of the present invention. The horizontal axis represents 64 samples indicating the frequency, and the vertical axis represents the value of the signal at each frequency. A thin line indicates a signal (frequency domain signal) indicating 64 frequency components obtained by converting one frame (64 time domain signals) into the frequency domain. A thick line indicates an example of the first quantized signal obtained when the first vector quantizing unit 104 vector-quantizes the frequency domain signal indicated by the thin line.

  In the conventional encoder 100, an error signal between the frequency domain signal and the first quantized signal is obtained. That is, when the error between the first quantized signal and the frequency domain signal having a value other than 0 is large, the bits of the second vector quantization index are used to correct this error. Therefore, the number of bits of the 2-vector quantization index used for correcting the error of the first quantized signal having a value of 0 is reduced, resulting in a decoded signal having many frequency components having a value of 0. Therefore, if the number of bits necessary to quantize the frequency domain signal is insufficient, the frequency component that should be present in the input signal does not exist in the decoded signal (missing frequency component), resulting in musical noise. It was one of the causes.

  Therefore, in the present invention, a signal to be subjected to the next vector quantization (second encoding target signal) is generated using only the frequency domain signal corresponding to the first quantized signal having a value of 0. In the subsequent stage vector quantization, the second encoding target signal is subjected to vector quantization by collecting a plurality of samples. That is, since only the frequency domain signal corresponding to the first quantized signal having a value of 0 is vector-quantized, the decoded signal has fewer frequency components having a value of 0. Therefore, loss of frequency components is less likely to occur, and musical noise can be reduced.

Specific Example FIG. 4 shows a configuration example of the encoder and decoder of the first embodiment. FIG. 5 shows a processing flow example of the encoder and decoder of the first embodiment. 5A shows the processing flow of the encoder, and FIG. 5B shows the processing flow of the decoder. The encoder 200 includes a frequency domain transform unit 101, a normalized value calculation unit 102, a normalized value quantization unit 103, a first vector quantization unit 104, an extraction unit 212, and a second vector quantization unit 206. The decoder 200 ′ includes a normalized value decoding unit 107, a first vector decoding unit 108, an M value calculation unit 213, a second vector decoding unit 209, a reconstruction unit 214, and a time domain conversion unit 111. The encoder 200 is different from the encoder 100 in the extraction unit 212 and the second vector quantization unit 206. The decoder 200 ′ is different from the decoder 100 ′ in the M-value calculating unit 213, the second vector decoding unit 209, and the reconstruction unit 214.

  The extraction unit 212 receives the frequency domain signal X (k) and the first quantized signal X ^ (k) as inputs, and the frequency domain signal X (k) (value) in which no pulse was generated in the first vector quantization unit 104. The second encoding target signal E (m) extracted only from the first quantized signal whose frequency is 0) and the number M of frequency domain signals that were not pulsed in the first vector quantization unit (The number of first quantized signals having a value of 0) is obtained, and the second encoding target signal E (m) and the number M are output (S212). Here, m is an integer value representing the sequence number.

  FIG. 6 shows an example of a processing flow for obtaining the second encoding target signal E (m) and the number M. First, the process is started with k = 0 and m = 0. It is confirmed whether k is smaller than L (number of frequency domain signals X (k)) (S2121). When step S2121 is Yes, it is confirmed whether X ^ (k) is 0 (S2122). If step S2122 is Yes, E (m) is set to X (k), and the value of m is incremented by 1 (S2123). If step S2122 is No, the process proceeds to step S2124. In step S2124, the value of k is incremented by 1, and the process returns to step S2121. When S2121 is No, the value of m is set to M (S2125), and the process ends.

The second vector quantization unit 206, the second coded signal E (m) and quantization normalized value wherein X ^- and the value as an input the number M of first quantized signal 0. The second vector quantization unit 206 obtains a normalized second encoding target signal by dividing the second encoding target signal E (m) by the quantization normalized value X ⁻ or by multiplying by the inverse number. . Then, the second coded signal normalization and vector quantization in the range of order of multiple Th of M points or quantized vector, and outputs the index of the quantization representative vector as a second vector quantization index C ₂ (S206 ). The order of the quantization vector is, for example, 8. In this case, the multiple Th has 8, 16,..., 64, etc., and the number closest to M, the number closest to M, or the number closest to M may be selected as Th. For example, if the closest number equal to or less than M is selected as Th, vector quantization may be performed for Th out of M normalized second encoding target signals.

Next, the decoder 200 ′ will be described. The M-value calculation unit 213 receives the first decoded quantized signal X ^ (k) obtained by the first vector decoding unit 108, and the first decoded quantized signal X ^ (k) has a value of 0. The number M of one decoded quantized signal X ^ (k) is obtained. FIG. 7 is an example of a processing flow for obtaining the number M of first decoded quantized signals X ^ (k) having a value of 0. The process is started with k = 0 and m = 0. It is confirmed whether k is smaller than L (number of frequency domain signals X (k)) (S2131). When step S2131 is Yes, it is confirmed whether X ^ (k) is 0 (S2132). When step S2132 is Yes, the value of m is increased by 1 (S2133). If step S2132 is No, the process proceeds to step S2134. In step S2134, the value of k is incremented by 1, and the process returns to step S2131. When S2131 is No, the value of m is set to M (S2135), and the process ends. In the normalization of the encoder side quantized normalization value X ^- and that is used, by the decoder side is provided with M value calculation unit 213, the decoder the information indicating the number M from the encoder No need to send (no need to use bits to convey number M).

The second vector decoding unit 209, the second vector quantization index C ₂ and the decoded quantized normalization value X ^- and enter the number M of the value is 0 the first decoded quantized signal X ^ (k). The second vector decoding unit 209, a second vector by decoding the quantization index C ₂ in order of multiple Th of M points or quantized vector to obtain the normalized signal as the normalized second decoded quantized signal, normal The normalized second decoded quantized signal is denormalized by multiplying by the decoded quantized normalized value X ⁻ and output as the second decoded quantized signal E ^ (m) (S209).

  The reconstructing unit 214 is a first decoded quantized signal X ^ (k), a second decoded quantized signal E ^ (m), and a multiple of the order of the quantization vector used when the second vector decoding unit 209 decodes. The number M of the first decoded quantized signals X ^ (k) whose values are Th and 0 is input. When the value of the first decoded quantized signal X ^ (k) is other than 0, the reconstruction unit 214 sets the first decoded quantized signal X ^ (k) as the decoded signal Z (k), and uses the first decoded quantized signal When the value of the signal X ^ (k) is 0, the decoded signal Z (k) is obtained and output by sequentially setting the second decoded quantized signal E ^ (m) as the decoded signal Z (k) (S214). ). FIG. 8 is a diagram illustrating an example of a processing flow for reconstructing the decoded signal Z (k). The process is started with k = 0 and m = 0. It is confirmed whether k is smaller than L (number of frequency domain signals X (k)) (S2141). When step S2141 is Yes, it is confirmed that m is smaller than M (S2142). When step S2142 is Yes, it is confirmed that m is smaller than Th (S2143). When step S2143 is Yes, it is confirmed whether X ^ (k) is 0 (S2144). When step S2144 is Yes, E ^ (m) is set to Z (k), and the value of m is increased by 1 (S2145). When Steps S2142, S2143, and S2144 are No, X ^ (k) is set to Z (k) (S2146). Then, the value of k is incremented by 1 (S2147), and the process returns to step S2141. If S2141 is No, the process ends.

  Other components are the same as those of the encoder 100 and the decoder 100 '. With this configuration, according to the encoding method and the decoding method of the first embodiment, the second encoding target signal includes the first encoding target signal corresponding to the first quantized signal having a value other than 0. Only the first encoding target signal corresponding to the first quantized signal having a value of 0 is included. That is, the second vector quantization step is a process of adding a value other than 0 to the first encoding target signal whose value has become 0 in the first vector quantization. Therefore, the second code is a code for reducing the error between the first encoding target signal and the first quantized signal while avoiding the presence or absence of the frequency component for each frame that is one of the causes of musical noise. A vector quantization index can be determined. Therefore, efficient vector quantization with a limited number of bits can be realized.

  Further, according to the encoding method and the decoding method of the first embodiment, for example, the number of frequency components in one frame is set to 80 points (80 frequency domain signals), and pulses are generated in the first vector quantization unit. If the number of frequency components obtained is 40 points (40 first quantized signals having a value other than 0), the number of points to be vector quantized by the first vector quantization unit 104 is 80 points. Since the number of points to be vector quantized by the vector quantization unit 206 can be halved to 40 points, it can be expected to improve the efficiency of vector quantization.

  In the present embodiment, the signal (first encoding target signal) to be encoded by the first vector quantization unit 104 is assumed to be an acoustic signal converted into the frequency domain. However, it is not necessary to limit to this signal, and an acoustic signal in a specific frequency band (for example, 8 kHz to 14 kHz) among the acoustic signals in the frequency domain may be used as the first encoding target signal or divided into frames. The time domain acoustic signal thus made may be used as the first encoding target signal.

  In this embodiment, both the first vector quantization unit 104 and the second vector quantization unit 206 normalize the signal to be vector quantized and perform vector quantization. However, vector quantization may be performed without normalization. In this case, the normalized value calculation unit 102, the normalized value quantization unit 103, and the normalized value decoding unit 107 are not necessary. In addition, each process does not require division or multiplication with the quantized normalized value.

In the first embodiment, the second vector quantization unit 206 performs normalization using the quantization normalized value X ⁻ . Further, in the second vector decoding unit 209, decoded quantized normalization value X ^- inversely normalized using. However, in the second vector quantization unit, a plurality of samples of the second encoding target signal E (m) configured only by the frequency domain signal corresponding to the first quantized signal X ^ (k) having a value of 0 are obtained. Collectively vector quantization. Therefore, using the normalized value for the second encoding target signal E (m) improves the accuracy of normalization and enables efficient vector quantization. Therefore, in this embodiment, normalization is performed using a normalized value obtained from the second encoding target signal E (m) (hereinafter referred to as “recalculated normalized value E ⁻ ”).

  FIG. 9 shows a configuration example of the encoder and decoder of the second embodiment. FIG. 10 shows a processing flow example of the encoder and decoder of the second embodiment. 10A shows the processing flow of the encoder, and FIG. 10B shows the processing flow of the decoder. The encoder 300 includes a frequency domain transform unit 101, a normalized value calculation unit 102, a normalized value quantization unit 103, a first vector quantization unit 104, an extraction unit 212, a normalized value recalculation unit 315, a second vector. A quantization unit 306 is provided. The decoder 300 ′ includes a normalized value decoding unit 107, a first vector decoding unit 108, an M value calculation unit 213, a normalized value recalculation unit 316, a second vector decoding unit 309, a reconstruction unit 214, and a time domain conversion unit. 111. The encoder 300 is different in a normalized value recalculation unit 315 and a second vector quantization unit 306. The decoder 300 ′ is different from the decoder 200 ′ in a normalized value recalculation unit 316 and a second vector decoding unit 309.

Normalization value recalculation unit 315, a first quantized signal X ^ (k), the quantization normalized value X ^- and, the number M of first quantized signal value is 0, the second vector quantization unit A multiple Th of the order of the quantization vector close to the number M used in 306 is input. The normalized value recalculation unit 315 recalculates the normalized value for the second encoding target signal E (m) and outputs it as a recalculated normalized value E ⁻ (S315). Figure 11 is recalculated normalized value E ^- is an example of a processing flow for determining the. The process is started with k = 0 and tmp = 0. It is confirmed whether k is smaller than L (number of frequency domain signals X (k)) (S3151). If step S3151 is Yes, | X ^ (k) | ² is added to tmp (S3156). The value of k is incremented by 1 (S3157), and the process returns to step S3151. If S3151 is No, the recalculation normalized value E ^- the

(S3158) and the process ends.
Further, FIG. 12 is recalculated normalized value E ^- is an example of another processing flow for determining the. In this flow, the number of first quantized signals having a value of 0 is calculated again. The process starts with k = 0, m = 0, and tmp = 0. It is confirmed whether k is smaller than L (number of frequency domain signals X (k)) (S3151). If step S3151 is Yes, it is confirmed that m is smaller than M (S3152). When step S3152 is Yes, it is confirmed that m is smaller than Th (S3153). When step S3153 is Yes, it is confirmed whether X ^ (k) is 0 (S3154). When step S3154 is Yes, the value of m is increased by 1 (S3155). When Steps S3152, S3153, and S3154 are No, | X ^ (k) | ² is added to tmp (S3156). Then, the value of k is incremented by 1 (S3157), and the process returns to step S3151. If S3151 is No, the recalculation normalized value E ^- the

(S3158 ′) and the process ends. In both the processing flow of FIG. 11 and the processing flow of FIG. 12, the recalculated normalized value E ⁻ multiplies the square of the quantized normalized value X ⁻ by the number L of the frequency domain signals X (k), The power of the first quantized signal X ^ (k) | X ^ (k) | ² is subtracted, the subtraction result is divided by the number M of the first quantized signals having a value of 0, and the square root of the division result is obtained. The recalculated normalization value E ⁻ is used.

The second vector quantization unit 306 is different from the second vector quantization unit 206 of the first embodiment in that the recalculated normalization value E ⁻ is used for normalization, but the others are the same. That is, the second vector quantization unit 306 receives the second encoding target signal E (m), the recalculated normalization value E ^−, and the number M of first quantized signals having a value of 0 as inputs. The second vector quantization unit 306 obtains a normalized second encoding target signal by dividing the second encoding target signal E (m) by the recalculated normalized value E ⁻ or by multiplying by the inverse number. . Then, the second coded signal normalization and vector quantization in the range of order of multiple Th of M points or quantized vector, and outputs the index of the quantization representative vector as a second vector quantization index C ₂ (S306 ).

Next, differences between the decoder 300 ′ will be described. The normalized value recalculator 316 obtains a recalculated normalized value E ⁻ for the decoder 300 ′ by the same process as the normalized value recalculator 315. If the recalculated normalization value E ⁻ is obtained according to the processing flow of FIG. 11 or FIG. 12, the information of the recalculated normalization value E ⁻ need not be transmitted from the encoder to the decoder, and the use of bits is avoided. be able to.

The second vector decoding unit 309 is the same as the second vector decoding unit 209 of the first embodiment except that the recalculated normalization value E ⁻ is used for normalization. That is, the second vector decoding unit 309 receives the second vector quantization index C ₂ , the recalculated normalized value E ^−, and the number M of the first decoded quantized signals X ^ (k) whose values are 0. . The second vector decoding unit 309, a second vector by decoding the quantization index C ₂ in order of multiple Th of M points or quantized vector to obtain the normalized second decoded quantized signal is normalized signal, The normalized second decoded quantized signal is denormalized by multiplying it by the recalculated normalization value E ⁻ and is output as the second decoded quantized signal E ^ (m) (S309).

  Other configurations are the same as those of the encoder 200 and the decoder 200 '. With such a configuration, according to the encoding method and decoding method of the second embodiment, the same effects as the encoding method and decoding method of the first embodiment can be obtained. Since the normalized value for the second encoding target signal E (m) is used, the accuracy of normalization is improved and vector quantization can be performed more efficiently.

[Modification 1]
Quantized normalized values X ^- and the first quantized signal X ^ (k) and the target value ^- if the error between (normalized value of the frequency domain signals X ₀ and the frequency domain signal X (k)) is large, Example the calculation method, recalculated normalized value E ^{- -} 2 recalculation normalized value E a is sometimes not be accurately calculated. In this case, for example, if the recalculated normalization value E ⁻ is estimated too small, the frequency component is lost, and if the recalculated normalization value E ⁻ is estimated too large, the quantization noise is generated. Therefore, in this modified example, the normalized value recalculation units 415 and 416 calculate the recalculated normalized value E ⁻ by the following method.

  FIG. 9 shows a functional configuration of this modification, and FIG. 10 shows a processing flow. The only difference between the encoder 400 and decoder 400 ′ from the encoder 300 and decoder 300 ′ of the second embodiment is the normalized value recalculation units 415 and 416.

The normalized value recalculation units 415 and 416 obtain the recalculated normalized value E ⁻ as follows. Figure 13 is recalculated normalized value E ^- is an example of a processing flow for determining the. The process is started with k = 0, tmp = 0, and tmp2 = 0. It is confirmed whether k is smaller than L (number of frequency domain signals X (k)) (S4151). When step S4151 is Yes, | X ^ (k) | ² is added to tmp, and | X ^ (k) | is added to tmp2 (S4156). The value of k is incremented by 1 (S4157), and the process returns to step S4151. When S4151 is No, the first recalculated value E ^- ₁ and the second recalculated value E ^- ₂ are

(S4158). E ^- ₁ is E ^- large or confirms than ₁ times _{2 A} (S4159). A ₁ is an adjustment constant serving as a reference for selecting which candidate, and for example, A ₁ = 0.5. If step S4159 is Yes, the first recalculated E ^- ₁ recalculates normalized value E ^- and then (S4160), if the step S4159 is No, the second recalculated E ^- ₂ recalculation normalized value E ^- (S4161). It is confirmed whether E ⁻ is smaller than A ₂ times X ⁻ (S4162). A ₂ is a constant for confirming whether E ⁻ has a contradictory value, and for example, A ₂ = 1. Step S4162 is finished processing cases Yes, the case step S4162 is No, the recalculation normalized value E ^- quantized normalized value X ^- replaced (S4163), the process ends.

FIG. 14 is an example of another processing flow for obtaining the recalculated normalization value E ⁻ . In this flow, the number of first quantized signals having a value of 0 is calculated again. The process is started with k = 0, m = 0, tmp = 0, and tmp2 = 0. It is confirmed whether k is smaller than L (number of frequency domain signals X (k)) (S4151). If step S4151 is Yes, it is confirmed that m is smaller than M (S4152). If step S4152 is Yes, it is confirmed that m is smaller than Th (S4153). When step S4153 is Yes, it is confirmed whether X ^ (k) is 0 (S4154). When step S4154 is Yes, the value of m is increased by 1 (S4155). When Steps S4152, S4153, and S4154 are No, | X ^ (k) | ² is added to tmp, and | X ^ (k) | is added to tmp2 (S4156). The value of k is incremented by 1 (S4157), and the process returns to step S4151. When S4151 is No, the first recalculated value E ^- ₁ and the second recalculated value E ^- ₂ are

(S4158 ′). The processing from step S4159 to S4163 is the same as that in FIG. In the processing flow of FIG. 13, in the processing flow of FIG. 14, the first recalculated E ^- ₁ is quantized normalized values X ^- multiplied by the number L of the frequency domain signal X (k) to the square of the multiplication result Is subtracted from the sum of the powers | X ^ (k) | ² of the first quantized signal X ^ (k), and the subtraction result is divided by the number M of first quantized signals having a value of 0, and the square root of the division result Is the first recalculated value E ^- ₁ . The second recalculated E ^- ₂ is quantized normalized values X ^- multiplied by the number L of the frequency domain signal X (k), the absolute value of the first quantized signal X ^ (k) from the multiplication result The total of | X ^ (k) | is subtracted, the subtraction result is divided by the number M of first quantized signals having a value of 0, and the division result is taken as a second recalculated value E ^- ₂ .

Other configurations are the same as those of the encoder 300 and the decoder 300 ′. With such a configuration, according to the encoding method and decoding method of the present modification, the same effects as the encoding method and decoding method of the second embodiment can be obtained. The recalculation normalized value E ^- since the can more accurately calculated, and improves the accuracy of the normalization, can more efficiently vector quantization.

Experimental Results An experiment was conducted to confirm the effect of the method of the second embodiment. FIG. 15 is a diagram showing experimental results when only audio is targeted, FIG. 16 is a diagram showing experimental results with respect to audio when music is flowing in the background, and FIG. 17 is for audio when noise is included. It is a figure which shows an experimental result. In each figure, the horizontal axis indicates time, and the vertical axis indicates the S / N ratio shown in the following equation. The larger the value, the better the quality.

  A dotted line indicates a case where encoding and decoding are performed by the conventional method, and a solid line indicates a case where encoding and decoding are performed by the method of the second embodiment. The first encoding target signal is a frequency domain signal converted into the frequency domain by MDCT (Modified Discrete Cosine Transform). Any experimental result shows that the quality of the present invention is better. This also shows that the present invention can realize efficient vector quantization with a limited number of bits.

The various processes described above, such as a program, are not only executed in time series according to the description, but may also be executed in parallel or individually as required by the processing capability of the apparatus that executes the process. Needless to say, other modifications are possible without departing from the spirit of the present invention.
Further, when the above-described configuration is realized by a computer, processing contents of functions that each device should have are described by a program. The processing functions are realized on the computer by executing the program on the computer.

The program describing the processing contents can be recorded on a computer-readable recording medium. As the computer-readable recording medium, for example, any recording medium such as a magnetic recording device, an optical disk, a magneto-optical recording medium, and a semiconductor memory may be used.
The program is distributed by selling, transferring, or lending a portable recording medium such as a DVD or CD-ROM in which the program is recorded. Furthermore, the program may be distributed by storing the program in a storage device of the server computer and transferring the program from the server computer to another computer via a network.

  A computer that executes such a program first stores, for example, a program recorded on a portable recording medium or a program transferred from a server computer in its own storage device. When executing the process, the computer reads a program stored in its own recording medium and executes a process according to the read program. As another execution form of the program, the computer may directly read the program from a portable recording medium and execute processing according to the program, and the program is transferred from the server computer to the computer. Each time, the processing according to the received program may be executed sequentially. Also, the program is not transferred from the server computer to the computer, and the above-described processing is executed by a so-called ASP (Application Service Provider) type service that realizes the processing function only by the execution instruction and result acquisition. It is good. Note that the program in this embodiment includes information that is used for processing by an electronic computer and that conforms to the program (data that is not a direct command to the computer but has a property that defines the processing of the computer).

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

100, 200, 300, 400 Encoder 100 ′, 200 ′, 300 ′, 400 ′ Decoder 101 Frequency domain transform unit 102 Normalized value calculator 103 Normalized value quantizer 104 First vector quantizer 105 Error Calculation unit 106, 206, 306 Second vector quantization unit 107 Normalized value decoding unit 108 First vector decoding unit 109, 209, 309 Second vector decoding unit 110 Error correction unit 111 Time domain conversion unit 212 Extraction unit 213 M value Calculation unit 214 Reconfiguration unit 315, 316, 415, 416 Normalized value recalculation unit

Claims (9)

A signal sample to be encoded for each predetermined number of samples (hereinafter referred to as “first encoding target signal”) or a first quantized signal obtained by vector quantization of a plurality of samples of signals corresponding to the first encoding target signal. And a first vector quantization step for obtaining a first vector quantization index corresponding to the first quantized signal;
An extraction step of generating a second encoding target signal using only the first encoding target signal corresponding to the first quantized signal having a value of 0;
A second vector quantization step of obtaining a second vector quantization index by vector-quantizing a plurality of samples of the second encoding target signal or a signal corresponding to the second encoding target signal;
An encoding method comprising: The encoding method according to claim 1, comprising:
A normalized value calculating step for obtaining a normalized value of the first encoding target signal;
A normalized value quantization step of quantizing the normalized value to obtain a normalized value quantization index and a quantized normalized value;
The quantization normalized value is X ⁻ , the predetermined number of samples is L, the total power of the first quantized signal is tmp, the number of the first quantized signals having a value of 0 is M, and a recalculated normalized value E ^- when the,

A normalized value recalculation step for obtaining a recalculated normalized value E ⁻ which is
The first vector quantization step normalizes the first encoding target signal using the quantized normalized value to obtain a normalized first encoding target signal, and determines the normalized first encoding target signal as A plurality of samples are vector-quantized to obtain the first vector quantization index and the first quantized signal,
The second vector quantization step normalizes the second encoding target signal using the recalculated normalization value to obtain a normalized second encoding target signal, and determines the normalized second encoding target signal as A coding method, wherein a plurality of samples are vector-quantized to obtain the second vector quantization index. The encoding method according to claim 1, comprising:
A normalized value calculating step for obtaining a normalized value of the first encoding target signal;
A normalized value quantization step of quantizing the normalized value to obtain a normalized value quantization index and a quantized normalized value;
The quantized normalization value X ^-, the predetermined number of samples L, and total power of the first quantized signal tmp, the first tmp2 the sum of the absolute values of the quantized signal, the value of 0 first When the number of 1 quantized signals is M, the first recalculation value is E ^- ₁ and the second recalculation value is E ^- ₂ ,

E the first recalculated as a relation ^- ₁ and the second recalculated E ^- ₂ a determined,
A normalized value recalculation step in which any one of the first recalculated value, the second recalculated value, and the quantized normalized value is a recalculated normalized value;
The first vector quantization step normalizes the first encoding target signal using the quantized normalized value to obtain a normalized first encoding target signal, and determines the normalized first encoding target signal as A plurality of samples are vector-quantized to obtain the first vector quantization index and the first quantized signal,
The second vector quantization step normalizes the second encoding target signal using the recalculated normalization value to obtain a normalized second encoding target signal, and determines the normalized second encoding target signal as A coding method, wherein a plurality of samples are vector-quantized to obtain the second vector quantization index. A decoding method for obtaining a decoded signal from a code including a first vector quantization index and a second vector quantization index,
A first vector decoding step of obtaining a first decoded quantized signal from a signal corresponding to the first vector quantization index ;
Before SL signal corresponding to the second vector quantization index and the second vector decoding step of obtaining a second decoded quantized signal,
When the value of the first decoded quantized signal is other than 0, the value of the first decoded quantized signal is the value of the decoded signal. When the value of the first decoded quantized signal is 0, the second decoded quantized signal A decoding method comprising: a reconstruction step in which the value of the coded signal is the value of the decoded signal. The decoding method according to claim 4, wherein
The code also includes a normalized value quantization index,
A normalized value decoding step of obtaining a decoded quantization normalized value corresponding to the normalized value quantization index;
An M value calculation step for obtaining the number M of the first decoded quantized signals having a value of 0;
The decoded quantized normalized value is X ⁻ , the number of the first decoded quantized signals is L, the total power of the first decoded quantized signals is tmp, and the number of the first decoded quantized signals is 0. Is M and the recalculated normalization value E ⁻

A normalized value recalculation step for obtaining a recalculated normalized value E ⁻ which is
The first vector decoding step obtains a normalized first decoded quantized signal corresponding to the first vector quantization index, and uses the normalized first decoded quantized signal and the decoded quantized normalized value. Obtaining a first decoded quantized signal;
The second vector decoding step obtains a normalized second decoded quantized signal corresponding to the second vector quantization index, and uses the normalized second decoded quantized signal and the recalculated normalized value. 2. A decoding method characterized by obtaining a decoded quantized signal. The decoding method according to claim 4, wherein
The code also includes a normalized value quantization index,
A normalized value decoding step of obtaining a decoded quantization normalized value corresponding to the normalized value quantization index;
An M value calculation step for obtaining the number M of the first decoded quantized signals having a value of 0;
The decoded quantization normalized value is X ⁻ , the number of the first decoded quantized signals is L, the total power of the first decoded quantized signals is tmp, and the total absolute value of the first quantized signals is tmp 2 , the number of the first decoded quantized signal values are 0 M, the first recalculated E ^- _1, the second recalculated E ^- when a _2,

E the first recalculated as a relation ^- ₁ and the second recalculated E ^- ₂ a determined,
A normalized value recalculation step in which any one of the first recalculated value, the second recalculated value, and the decoded quantized normalized value is a recalculated normalized value;
The first vector decoding step obtains a normalized first decoded quantized signal corresponding to the first vector quantization index, and uses the normalized first decoded quantized signal and the decoded quantized normalized value. Obtaining a first decoded quantized signal;
The second vector decoding step obtains a normalized second decoded quantized signal corresponding to the second vector quantization index, and uses the normalized second decoded quantized signal and the recalculated normalized value. 2. A decoding method characterized by obtaining a decoded quantized signal. A signal sample to be encoded for each predetermined number of samples (hereinafter referred to as “first encoding target signal”) or a first quantized signal obtained by vector quantization of a plurality of samples of signals corresponding to the first encoding target signal. And a first vector quantization unit for obtaining a first vector quantization index corresponding to the first quantized signal;
An extraction unit that generates a second encoding target signal only with the first encoding target signal corresponding to the first quantized signal having a value of 0;
A second vector quantization unit for obtaining a second vector quantization index by vector-quantizing a plurality of samples of the second encoding target signal or a signal corresponding to the second encoding target signal;
An encoder comprising: A decoder for obtaining a decoded signal from a code including a first vector quantization index and a second vector quantization index,
A first vector decoding unit for obtaining a first decoded quantized signal from a signal corresponding to the first vector quantization index ;
Before SL signal corresponding to the second vector quantization index and the second vector decoding unit for obtaining a second decoded quantized signal,
When the value of the first decoded quantized signal is other than 0, the value of the first decoded quantized signal is the value of the decoded signal. When the value of the first decoded quantized signal is 0, the second decoded quantized signal A decoder comprising: a reconstruction unit that uses the value of the coded signal as the value of the decoded signal.   A program that causes a computer to execute each step of the method according to claim 1.

Images