JP4556866B2 - High efficiency encoding program and high efficiency encoding apparatus - Google Patents

Description

  The present invention relates to a high-efficiency encoding program and a high-efficiency encoding apparatus for encoding speech or audio signals, and in particular, provides a quality that can withstand listening by making the quantization noise spectrum audibly desirable. The present invention relates to a high-efficiency encoding program and a high-efficiency encoding apparatus.

  Conventionally, there is a technique for encoding a voice or an audio signal, and there is the following Patent Document 1 as the conventional technique. Patent Document 1 describes a method for improving the quality of hearing by changing the shape of a quantization noise spectrum in an encoding method using differential encoding and quasi-instantaneous compression. In this method, a difference value between two adjacent samples of a digital audio signal is compressed using range information obtained according to the maximum value of a level in a block composed of a difference value of a certain number of samples. Yes, the filter characteristics determined by the information from the level detection means, the masking characteristic calculation means, and the equal loudness curve generation means are applied to the quantization noise by the noise shaping means, so that the equal loudness characteristics and the masking characteristics are met. The quantization noise spectrum can be changed to a audible noise.

  As another encoding technique, there is a technique described in Non-Patent Document 1 below. This technique is adaptive differential PCM (Pulse Code Modulation) and is widely known. In this technique, a difference value from the current sample is calculated using a value one sample before as a predicted value, and encoding is performed, and a step size (quantization width) at that time is obtained by encoding the previous difference value. It is determined by the sign. An encoding tool based on this method is incorporated in general-purpose audio editing software that operates on a PC (Personal Computer), and can be freely encoded and decoded. Specifically, the high-efficiency encoding device of Non-Patent Document 1 will be described with reference to FIG.

  As shown in FIG. 12, the input audio signal X (n) is a linear PCM with a sampling frequency of 8 kHz and 12-bit accuracy. Since an ordinary CD or the like has an accuracy of 16 bits, the LSB (Least Significant Bit) 4th bit is rounded to 12 bits. In the difference calculation step unit 121, the difference between the current sample X (n) and the previous compressed / extended sample X ′ (n−1) is taken to generate a difference value d (n). Then, in the step size (quantization width) determination step unit 123, the step size ss (n) determined based on the encoding result of the previous difference value is supplied to the compression step unit 122, and the difference value d (n ) Is converted into a 4-bit ADPCM code L (n). In the expansion step unit 124, the ADPCM code L (n) converted by the compression step unit 122 is inversely converted to obtain a difference value d ′ (n) including quantization noise (hereinafter also referred to as quantization error). . Then, the accumulation step unit 125 performs reverse processing of the difference calculation step unit 121 to reconstruct the previous sample X ′ (n−1).

Also, since this technique is an encoding method for audio signals, it assumes a relatively gradual change in signal. As a result, when a sudden change occurs in the input signal, the configuration in which the current step size (quantization width) is determined from the code obtained by encoding the immediately preceding difference value cannot follow the signal and the quantization noise There was a problem that the sound became rough and harsh. In order to solve this problem, consider using the technique described in Patent Document 1 described above. The noise shaping means in the technique described in Japanese Patent Application Laid-Open No. H10-228688 increases the spectral shape so that the quantization noise is increased in a region (for example, 16 kHz or more) inaudible based on equal loudness characteristics and masking characteristics. Is changing.
Japanese Patent No. 3191457 (summary, FIG. 5) “Dialogic ADPCM Algorithm” 1988, Dialogic Cooperation

  However, when the technique described in Patent Document 1 is applied to the technique described in Non-Patent Document 1, the sampling frequency used in the technique of Non-Patent Document 1 is 8 kHz (Nyquist frequency 4 kHz). There is a problem that there is no band for moving quantization noise. Even when based on the masking characteristics, weighting is generally performed to reduce the low-mid range quantization noise when calculating the masking level. There is a problem that there is no destination.

  The present invention is to solve the above problem, and in encoding using a low sampling frequency, the spectral envelope of quantization noise spread uniformly in frequency without any change to the existing format. Can be made closer to the spectral envelope of the audio signal, the quality of the audio signal can be improved, and the bias that the signal-to-quantization noise ratio (SNR) is high only in a specific frequency band can be improved, An object of the present invention is to provide a high-efficiency encoding program and a high-efficiency encoding apparatus that can prevent a quality gap that occurs at a processing frame boundary.

  In order to achieve the above object, according to the present invention, it is possible to change the spectral shape so as to increase the quantization noise in a region that cannot be audibly heard based on the equal loudness characteristic and / or the masking characteristic. A difference calculating step for calculating a difference value between the current sample and the previous sample in the input audio signal sampled at a low sampling frequency lower than the sampling frequency, and performing quantization and encoding on the difference value A compression step, a decompression step that performs decoding and inverse quantization on the output obtained from the compression step, and a quantization of a difference value that is calculated next based on the encoding result of the compression step A quantization width determination step for determining a quantization width of the output, and the output obtained from the expansion step and the difference value when calculating the difference value A high-efficiency encoding program for causing a computer to execute a cumulative step of calculating a previous sample value when a difference value is calculated in a next difference calculation step by adding a previous sample value. A framing step for generating a processing frame by dividing the input audio signal into a predetermined number of samples, a spectrum envelope having a reverse characteristic of a spectrum envelope with respect to the generated processing frame is obtained, and the obtained inverse Calculating a coefficient indicating a characteristic of the spectral envelope that cancels the characteristic of the spectral envelope of the characteristic, a filter characteristic determination step using the calculated coefficient as a filter coefficient, a value input to the compression step, and the expansion step Quantization noise is calculated based on the obtained output value, and the processing is performed on the calculated quantization noise. A high-efficiency encoding program for causing a computer to further execute a noise shaping step of shaping a quantized noise spectrum by applying the corresponding filter coefficient of a filter composed of the filter coefficients calculated for each frame; The characteristic determining step includes a spectral envelope limiting step for performing processing so that an amplitude fluctuation range determined by a difference between a maximum level and a minimum level of the spectrum envelope of the inverse characteristic of the processing frame is within a predetermined limit range. An encoding program is provided. With this configuration, in encoding using a low sampling frequency, the spectral envelope of quantization noise spread uniformly in frequency can be brought close to the spectral envelope of the audio signal without any change to the existing format. Therefore, the quality of the audio signal can be improved. In addition, it is possible to improve the bias that the signal-to-quantization noise ratio (SNR) is high only in a specific frequency band. The quantization noise is also referred to as quantization distortion or quantization error. The same applies to the following. Also, as described above, the low sampling frequency means that the spectral shape is increased so that the quantization noise is increased in a region (for example, 16 kHz or more) inaudible based on equal loudness characteristics and masking characteristics. It is a sampling frequency that cannot be changed, for example, a frequency of 32 kHz or less, and is not limited to 8 kHz described later.

  Further, according to the present invention, it is lower than the sampling frequency that can change the spectral shape so as to increase the quantization noise in the region that cannot be heard audibly based on the equal loudness characteristic and / or the masking characteristic. A difference calculating step of calculating a difference value between a current sample and a previous sample in an input audio signal sampled at a low sampling frequency, a compression step of performing quantization and encoding on the difference value, and the compression A decompression step that performs decoding and inverse quantization on the output obtained from the step, and a quantization width at the time of quantization of the difference value calculated next is determined based on the encoding result of the compression step A quantization width determination step, an output obtained from the expansion step, and a value of the previous sample when calculating the difference value A high-efficiency encoding program for causing a computer to execute a cumulative step of calculating a value of a previous sample when calculating a difference value in a next difference calculation step, wherein the input audio signal is A framing step for generating a processing frame separately for each predetermined number of samples, a spectrum envelope of a reverse characteristic of the spectral envelope with respect to the generated processing frame is obtained, and a characteristic of the obtained spectral envelope of the reverse characteristic is obtained. A coefficient indicating the characteristic of the spectral envelope that cancels the filter, a filter characteristic determination step using the calculated coefficient as a filter coefficient, an input value to the compression step, and an output value obtained from the expansion step Quantization noise is calculated based on the above, and the calculated quantization noise is calculated for each processing frame. And a noise shaping step of shaping a quantized noise spectrum by applying the corresponding filter coefficient of the filter composed of the filter coefficients, and causing the computer to execute the filter characteristic determining step. If the sample value does not change in the processing frame and the value continues, a DC value search step for counting the number of samples followed by the value, and the quantization noise based on the counted number of samples. There is provided a high-efficiency encoding program having a final filter selection step for selecting whether to apply the filter coefficient to be calculated to the calculated filter coefficient or to store the filter coefficient stored in a predetermined storage area in advance. With this configuration, in encoding using a low sampling frequency, the spectral envelope of quantization noise spread uniformly in frequency can be brought close to the spectral envelope of the audio signal without any change to the existing format. Therefore, the quality of the audio signal can be improved. In addition, it is possible to prevent a quality gap that occurs at the processing frame boundary.

  Further, according to the present invention, it is lower than the sampling frequency that can change the spectral shape so as to increase the quantization noise in the region that cannot be heard audibly based on the equal loudness characteristic and / or the masking characteristic. A difference calculating step of calculating a difference value between a current sample and a previous sample in an input audio signal sampled at a low sampling frequency, a compression step of performing quantization and encoding on the difference value, and the compression A decompression step that performs decoding and inverse quantization on the output obtained from the step, and a quantization width at the time of quantization of the difference value calculated next is determined based on the encoding result of the compression step A quantization width determination step, an output obtained from the expansion step, and a value of the previous sample when calculating the difference value A high-efficiency encoding program for causing a computer to execute a cumulative step of calculating a value of a previous sample when calculating a difference value in a next difference calculation step, wherein the input audio signal is A framing step for generating a processing frame separately for each predetermined number of samples, a spectrum envelope of a reverse characteristic of the spectral envelope with respect to the generated processing frame is obtained, and a characteristic of the obtained spectral envelope of the reverse characteristic is obtained. A coefficient indicating the characteristic of the spectral envelope that cancels the filter, a filter characteristic determination step using the calculated coefficient as a filter coefficient, an input value to the compression step, and an output value obtained from the expansion step Quantization noise is calculated based on the above, and the calculated quantization noise is calculated for each processing frame. And a noise shaping step of shaping a quantized noise spectrum by applying the corresponding filter coefficient of the filter composed of the filter coefficients, and causing the computer to execute the filter characteristic determining step. A spectrum envelope limiting step for performing processing so that an amplitude fluctuation range determined by a difference between a maximum level and a minimum level of the spectrum envelope of the inverse characteristic of the processing frame falls within a predetermined limit width; and a sample of the sample in the processing frame When there is no change in the value and the value continues, a DC value search step for counting the number of samples followed by the value, and a filter coefficient to be applied to the quantization noise is calculated based on the counted number of samples. The filter coefficient is set or stored in a predetermined storage area in advance. And a final filter selection step for selecting whether to use the stored filter coefficient. With this configuration, in encoding using a low sampling frequency, the spectral envelope of quantization noise spread uniformly in frequency can be brought close to the spectral envelope of the audio signal without any change to the existing format. Therefore, the quality of the audio signal can be improved. In addition, it is possible to improve the bias that the signal-to-quantization noise ratio (SNR) is high only in a specific frequency band. In addition, it is possible to prevent a quality gap that occurs at the processing frame boundary.

  In the high-efficiency encoding program according to the present invention, the step of calculating a spectral envelope of the inverse characteristic of the spectral envelope with respect to the generated processing frame in the filter characteristic determining step obtains a linear prediction coefficient by linear prediction analysis. A characteristic of spectral envelope that cancels the characteristic of the obtained spectral envelope of the inverse characteristic, and comprises a spectral envelope calculating step of obtaining a spectral envelope by Fourier transform with respect to the linear prediction coefficient. A second linear prediction for calculating a linear prediction coefficient by performing a linear prediction analysis using a coefficient obtained by performing an inverse Fourier transform on the spectrum envelope obtained by the Fourier transform. Comprising the analysis step is a preferred embodiment of the present invention. With this configuration, the spectral envelope of quantization noise spread uniformly in frequency can be brought close to the spectral envelope of the audio signal.

  In the high-efficiency encoding program of the present invention, it is a preferable aspect of the present invention that the processing frame configured in the framing step is a processing frame obtained by combining the immediately preceding processing frame. With this configuration, it is possible to prevent a quality gap that occurs at the processing frame boundary.

  Further, according to the present invention, it is lower than the sampling frequency that can change the spectral shape so as to increase the quantization noise in the region that cannot be heard audibly based on the equal loudness characteristic and / or the masking characteristic. Difference calculating means for calculating a difference value between the current sample and the previous sample in the input audio signal sampled at a low sampling frequency, compression means for performing quantization and encoding on the difference value, and the compression A decompression unit that performs decoding and inverse quantization on the output from the unit, and a quantum that determines a quantization width in the quantization of the difference value calculated next based on the encoding result of the compression unit And adding the output from the decompression means and the value of the previous sample when calculating the difference value, and then the difference calculation means A high-efficiency encoding device comprising accumulation means for calculating a value of a previous sample at the time of calculation, wherein the input audio signal is divided into predetermined samples to generate a processing frame And calculating a coefficient indicating the characteristic of the spectrum envelope for canceling the characteristic of the spectrum envelope of the reverse characteristic obtained, and calculating a spectrum envelope of the reverse characteristic of the spectrum envelope for the generated processing frame A quantization characteristic is calculated on the basis of a filter characteristic determination unit using the calculated coefficient as a filter coefficient, a value input to the compression unit, and a value output from the decompression unit, and the calculated quantization noise The quantization noise spectrum is multiplied by the filter coefficient corresponding to the filter composed of the filter coefficients calculated for each processing frame. Noise shaping means for shaping the noise, and the filter characteristic determination means has a predetermined limit width for an amplitude fluctuation range determined by a difference between a maximum level and a minimum level of the spectrum envelope of the inverse characteristic of the processing frame. There is provided a high-efficiency encoding device having spectrum envelope limiting means for performing processing so as to be contained within. With this configuration, in encoding using a low sampling frequency, the spectral envelope of quantization noise spread uniformly in frequency can be brought close to the spectral envelope of the audio signal without any change to the existing format. Therefore, the quality of the audio signal can be improved. In addition, it is possible to improve the bias that the signal-to-quantization noise ratio (SNR) is high only in a specific frequency band.

  Further, according to the present invention, it is lower than the sampling frequency that can change the spectral shape so as to increase the quantization noise in the region that cannot be heard audibly based on the equal loudness characteristic and / or the masking characteristic. Difference calculating means for calculating a difference value between the current sample and the previous sample in the input audio signal sampled at a low sampling frequency, compression means for performing quantization and encoding on the difference value, and the compression A decompression unit that performs decoding and inverse quantization on the output from the unit, and a quantum that determines a quantization width in the quantization of the difference value calculated next based on the encoding result of the compression unit And adding the output from the decompression means and the value of the previous sample when calculating the difference value, and then the difference calculation means A high-efficiency encoding device comprising accumulation means for calculating a value of a previous sample at the time of calculation, wherein the input audio signal is divided into predetermined samples to generate a processing frame And calculating a coefficient indicating the characteristic of the spectrum envelope for canceling the characteristic of the spectrum envelope of the reverse characteristic obtained, and calculating a spectrum envelope of the reverse characteristic of the spectrum envelope for the generated processing frame A quantization characteristic is calculated on the basis of a filter characteristic determination unit using the calculated coefficient as a filter coefficient, a value input to the compression unit, and a value output from the decompression unit, and the calculated quantization noise The quantization noise spectrum is multiplied by the filter coefficient corresponding to the filter composed of the filter coefficients calculated for each processing frame. A noise shaping means for shaping a filter, wherein the filter characteristic determination means counts the number of samples followed by the value when the sample value remains unchanged in the processing frame. And a filter coefficient to be applied to the quantization noise is selected as the calculated filter coefficient or a stored filter coefficient stored in a predetermined storage area based on the number of samples counted. There is provided a high-efficiency encoding device having final filter selection means. With this configuration, in encoding using a low sampling frequency, the spectral envelope of quantization noise spread uniformly in frequency can be brought close to the spectral envelope of the audio signal without any change to the existing format. Therefore, the quality of the audio signal can be improved. In addition, it is possible to prevent a quality gap that occurs at the processing frame boundary.

  Further, according to the present invention, it is lower than the sampling frequency that can change the spectral shape so as to increase the quantization noise in the region that cannot be heard audibly based on the equal loudness characteristic and / or the masking characteristic. Difference calculating means for calculating a difference value between the current sample and the previous sample in the input audio signal sampled at a low sampling frequency, compression means for performing quantization and encoding on the difference value, and the compression A decompression unit that performs decoding and inverse quantization on the output from the unit, and a quantum that determines a quantization width in the quantization of the difference value calculated next based on the encoding result of the compression unit And adding the output from the decompression means and the value of the previous sample when calculating the difference value, and then the difference calculation means A high-efficiency encoding device comprising accumulation means for calculating a value of a previous sample at the time of calculation, wherein the input audio signal is divided into predetermined samples to generate a processing frame And calculating a coefficient indicating the characteristic of the spectrum envelope for canceling the characteristic of the spectrum envelope of the reverse characteristic obtained, and calculating a spectrum envelope of the reverse characteristic of the spectrum envelope for the generated processing frame A quantization characteristic is calculated on the basis of a filter characteristic determination unit using the calculated coefficient as a filter coefficient, a value input to the compression unit, and a value output from the decompression unit, and the calculated quantization noise The quantization noise spectrum is multiplied by the filter coefficient corresponding to the filter composed of the filter coefficients calculated for each processing frame. Noise shaping means for shaping the noise, and the filter characteristic determination means has a predetermined limit width for an amplitude fluctuation range determined by a difference between a maximum level and a minimum level of the spectrum envelope of the inverse characteristic of the processing frame. Spectrum envelope limiting means for performing processing so as to fall within, DC value search means for counting the number of samples followed by the value when there is no change in the value of the sample in the processing frame, and the counted Final filter selection means for selecting whether the filter coefficient to be applied to the quantization noise is the calculated filter coefficient or the stored filter coefficient stored in a predetermined storage area based on the number of samples. A high-efficiency encoding device is provided. With this configuration, in encoding using a low sampling frequency, the spectral envelope of quantization noise spread uniformly in frequency can be brought close to the spectral envelope of the audio signal without any change to the existing format. Therefore, the quality of the audio signal can be improved. In addition, it is possible to improve the bias that the signal-to-quantization noise ratio (SNR) is high only in a specific frequency band. In addition, it is possible to prevent a quality gap that occurs at the processing frame boundary.

  The high-efficiency encoding program and the high-efficiency encoding apparatus of the present invention have the above-described configuration, and in encoding using a low sampling frequency, the frequency is uniformly spread without changing any existing format. The spectrum envelope of the quantization noise can be brought close to the spectrum envelope of the audio signal, and the quality of the audio signal can be improved. In addition, it is possible to improve the bias that the signal-to-quantization noise ratio (SNR) is high only in a specific frequency band. In addition, it is possible to prevent a quality gap that occurs at the processing frame boundary.

  Hereinafter, a high-efficiency encoding apparatus and a high-efficiency encoding program according to embodiments of the present invention will be described with reference to FIGS. FIG. 1 is a configuration diagram showing a configuration of a high-efficiency encoding apparatus according to an embodiment of the present invention. FIG. 2 is a diagram for explaining an example of processing of the framing step unit in the high efficiency coding apparatus according to the embodiment of the present invention. FIG. 3 is a flowchart for explaining an example of processing of the filter characteristic determination step unit in the high-efficiency encoding device according to the embodiment of the present invention. FIG. 4 is a diagram showing a spectrum envelope (hereinafter also simply referred to as a spectrum envelope) having the inverse characteristics of the data of the processing frame in the high efficiency coding apparatus according to the embodiment of the present invention. FIG. 5 is a diagram for explaining an example of processing of the spectrum envelope limiting step unit in the high-efficiency encoding device according to the embodiment of the present invention.

  FIG. 6 is a diagram for explaining a problem caused by the presence of DC data in the start frame in the high efficiency coding apparatus according to the embodiment of the present invention. FIG. 7 is a diagram for explaining a problem caused by the presence of DC data in the end frame in the high efficiency coding apparatus according to the embodiment of the present invention. FIG. 8 is a flowchart showing an example of processing of the DC value search step unit in the high efficiency encoding device according to the embodiment of the present invention. FIG. 9 is a flowchart showing a part of processing (start frame searching step) as an example of processing of the DC value searching step unit in the high-efficiency encoding device according to the embodiment of the present invention. FIG. 10 is a flowchart showing a part of processing (end frame searching step) as an example of processing of the DC value searching step unit in the high efficiency encoding device according to the embodiment of the present invention. FIG. 11 is a conceptual diagram showing the operation of the noise shaping step unit in the high efficiency coding apparatus according to the embodiment of the present invention.

  First, a high-efficiency encoding device according to an embodiment of the present invention will be described. As shown in FIG. 1, the high-efficiency encoding apparatus according to the embodiment of the present invention is different from the conventional high-efficiency encoding apparatus in that a framing step unit 16, a filter characteristic determination step unit 17, and a noise shaping step unit. 18 is added. The input audio signal X (n) is a linear PCM with a sampling frequency of 8 kHz and 12-bit accuracy. In the difference calculation step unit 11, the difference between the current sample X (n) and the previous sample X ′ (n−1) that has been compressed and expanded is taken to generate a difference value d (n). The step size (quantization width) determination step unit 13 determines the step size ss (n) based on the encoding result of the previous difference value, and the determined step size ss (n) is the compression step unit. 12 and the compression step unit 12 converts the difference value d (n) into a 4-bit ADPCM code L (n). In the decompression step unit 14, the ADPCM code L (n) transformed by the compression step unit 12 is inversely transformed to obtain a difference value d ′ (n) including quantization noise. Then, the accumulation step unit 15 performs the reverse process of the difference calculation step unit 11 to reconstruct the previous sample X ′ (n−1).

  The input audio signal X (n) is framed in units of 1024 samples in the framing step unit 16. Then, the accumulated 1024 samples of the previous frame and the 1024 samples of the current frame are combined in time series to form a processing frame of 2048 samples (also referred to as overlapping processing).

  The filter characteristic determination step unit 17 calculates linear prediction coefficients and orders having spectral envelope characteristics using linear prediction (LPC (Linear Predictive Coding)) analysis on the processing frames that have been subjected to the overlapping process as described above. I do. In the noise shaping step unit 18, the quantization error e (n) (obtained from the difference value d (n) immediately before being input to the compression step unit 12 and the difference value d ′ (n) after the expansion step unit 14. For e (n) = d (n) −d ′ (n)), a FIR (Finite Impulse Response) filter configured using linear prediction coefficients is applied. The quantization error ns (n−1) that has been subjected to the filter processing is fed back to the difference calculation step unit 11 so that the approximate shape of the quantization error spectrum is shaped according to the filter characteristics.

  Next, an example of the process of the framing step unit 16 described above will be described with reference to FIG. The input audio signal X (n) is divided in units of 1024 samples (also referred to as frames). The processing frame is configured in a time series manner with past frame data stored in a storage area (not shown). The first processing frame 0 is normally configured using 1024 samples of zero data as past frame data. However, if the first sample of the audio signal X (n) is not zero data, the past processing frame 0 has the same value as that of the first sample. Frame data may be embedded.

  Then, the filter characteristic determination step unit 17 calculates a linear prediction coefficient using this processing frame. The reason for using such overlapped data is to smoothly change the linear prediction coefficient so that a data gap due to filtering does not occur at the joint of frames. The filter coefficient obtained in the processing frame 0 is applied to 1024 samples of the quantization error data in the frame 0. Similarly, the filter coefficient obtained in the processing frame 1 is applied to 1024 samples of the quantization error data in the frame 1. Therefore, it is necessary to smoothly connect the joints of frame 0 and frame 1. Although the frame structure is adopted here, since the input to the difference calculation step unit 11 is every sample, it is input to the difference calculation step unit 11 in units of one sample from the head of the frame.

  Next, an example of the processing of the filter characteristic determination step unit 17 described above will be described with reference to FIG. As shown in FIG. 3, in the processing step S31 in the first linear prediction analysis step unit, linear prediction is performed on the data of the input processing frame to obtain a linear prediction coefficient. In the processing step S32 in the spectrum envelope calculation step unit, a power spectrum is calculated by multiplying the obtained linear prediction coefficient by a forward FFT (Fast Fourier Transform), and the process shown in FIG. A spectral envelope with the inverse characteristics of the frame data is obtained. In processing step S33 in the spectrum envelope limiting step unit, a ratio of the limiting width to the amplitude fluctuation width is calculated in order to keep the amplitude fluctuation width of the spectrum envelope characteristic within a predetermined limiting width, and processing for multiplying the power spectrum is performed.

  The noise shaping technique is not to reduce the amount of quantization error, but to modify the shape of the quantization error spectrum. It will be quite bad. However, uniform quality can be achieved by correcting the spectral envelope in the spectral envelope limiting step S33. In the processing step S34 in the second linear prediction analysis step unit for the corrected power spectrum obtained in this way, the linear prediction is performed by using the autocorrelation function calculated by the reverse FFT, so that the forward characteristic is obtained. A linear prediction coefficient having a spectral envelope is calculated.

  In processing step S35 in the DC (Direct Current) search step unit, using the input processing frame data and the linear prediction coefficient (current filter coefficient) calculated in processing step S34 in the second linear prediction analysis step unit, The number of samples in which the data of the processing frame continues with a certain value (zero or non-zero value) without any change is counted. Then, it is determined whether the current filter coefficient is used or the filter coefficient stored in a storage area (not shown) is used, and the result is output as a current filter use flag. In processing step S36 in the final filter selection step unit, the current filter coefficient and the saved filter coefficient are switched according to the current filter use flag, and the final filter coefficient is determined. The stored filter coefficients include fixed filter coefficients prepared in advance, current filter coefficients calculated in past frames, and the like.

  Here, the details of the processing step S33 in the above-described spectrum envelope limiting step unit will be described with reference to FIGS. By the processing step S31 in the first linear prediction analysis step unit, a linear prediction coefficient that cancels the outline of the spectrum envelope of the original sound around the average level is calculated. The outline is the dotted-line spectral envelope shown in FIG. Then, in the processing step S34 in the second linear prediction analysis step unit, a linear prediction coefficient that cancels the obtained spectral envelope characteristic again is obtained. Therefore, the finally obtained spectral envelope characteristic is shown by a solid line in FIG. This corresponds to the spectrum envelope of the original sound indicated by. For example, when the spectrum envelope obtained as a result of the processing step S31 in the first linear prediction analysis step unit is the spectrum envelope before the restriction shown in FIG. 5, the characteristics in the low band are considerably reduced. If the processing step S34 in the second linear prediction analysis step unit is applied as it is, a linear prediction coefficient that considerably increases the low-range quantization error is calculated.

  In order to prevent this, in the processing step S33 in the spectrum envelope restriction step unit, a process for keeping the amplitude fluctuation width of the spectrum envelope within a predetermined restriction width is performed. For example, when the average level is 0 dB, the amplitude fluctuation range (diff_max) is obtained by the absolute distance between the minimum level (min_val) and the maximum level (max_val) of the spectrum envelope as in the following equation (1). Then, the ratio is calculated from the following formulas (2) and (3) in comparison with a predetermined limit width (diff). Furthermore, in order to apply it to the power spectrum, it is converted from decibel to power as shown in the following equation (4).

  Next, problems that occur when processing a processing frame will be described with reference to FIGS. FIG. 6 shows an example of a problem due to the presence of DC data in the start frame. The DC data is zero data. In the conventional example, even when zero data is input, a constant quantization error occurs. Therefore, the difference value from the previous sample X ′ (n−1) that has been compressed and expanded also repeats constant increase and decrease. 6 repeats a constant waveform as shown in FIG. 6 (quantization error level generated according to the standard). The processing frame 0 is the value of the filter coefficient (filter coefficient 0) for data that is all zero, but the processing frame 1 calculates the value of the filter coefficient (filter coefficient 1) for a filter that contains data other than zero. As a result, different filter coefficients are applied to a certain quantization error, which causes a click sound. FIG. 7 is a diagram showing that a similar click sound is generated in the end frame. FIG. 7 shows that two click sounds are generated because the filter coefficients (filter coefficients 0, 1 and 2) relating to a fixed quantization error corresponding to the zero data portion are different from each other.

  In order to solve such a problem, the following processes shown in FIGS. 8 to 10 are performed. FIG. 8 is a diagram showing in detail the flowchart of the processing step S35 in FIG. 3 in the DC value search step unit. First, it is determined whether it is the first frame of the audio signal (step S81). If it is the first frame, the first sample value is stored as a DC value in a storage area (not shown) (step S82). Then, it is checked whether the start frame search has been completed (start frame search completion flag) (step S83). If the start frame search has been completed, an end frame search step, which will be described later with reference to FIG. 10, is performed (step S85). If the start frame search is not completed, a start frame search step described later with reference to FIG. 9 is performed (step S84). If it is not the first frame in step S81, the process proceeds to step S83.

  Details of the start frame search step (step S84) will be described below with reference to FIG. First, it is determined whether or not the DC value stored in a storage area (not shown) is zero data (step S91). If the DC value is zero data, are all the values of the linear prediction coefficients (current filter coefficients) zero? It is checked whether or not (step S92). If all are zero, the current filter use flag is turned off (OFF) so as to select a filter coefficient stored in a storage area (not shown) (step S94). In this case, since there is a possibility that zero data still continues, the start frame search completion flag is not set. On the other hand, if all the current filter coefficients are not zero in step S92, it is determined that the continuation of zero data is interrupted, and a start frame search completion flag is set (step S93). Then, the current filter use flag is turned off (step S94).

  In step S91, if the stored DC value is not zero (non-zero value), the number of samples in which the non-zero value continues from the first sample of the processing frame (the earliest sample in time) is counted ( Step S95). Next, it is determined whether or not the count value is less than the frame length (half of the processing frame) (step S96). If it is less, a start frame search completion flag is set (step S97). On the other hand, when the count value is equal to or longer than the frame length, the start frame search completion flag is not set because there is a possibility that the next frame is continued as in the case of zero data. In the DC value reliability determination (step S98), it is determined whether the DC value stored in the storage area (not shown) is really correct. For example, the DC value is corrected based on the number of continuous counts or the average value of the samples. Do. Then, the current filter use flag is turned off (step S99).

  Details of the end frame search step (step S85) described above with reference to FIG. 8 will be described below with reference to FIG. First, it is checked whether or not the coefficients of the linear prediction coefficient (current filter coefficient) are all zero (step S101). If it is zero, the DC value count is set to the frame length (step S103). On the other hand, if the coefficient of the linear prediction coefficient (current filter coefficient) is not zero, the sample of the DC value stored in the storage area (not shown) from the sample at the end of the processing frame (the latest sample in time) continues. The number is counted (step S102). Then, it is determined whether or not the count number is equal to or greater than a predetermined value (step S104). If it is equal to or greater than the predetermined value, the current filter use flag is turned off (OFF) (step S106). If it is less than the predetermined value, the current filter use flag is turned on (step S105).

  As described above, in the noise shaping step unit 18 with respect to the result processed by the framing step unit 16 and the filter characteristic determination step unit 17 shown in FIG. n) and the quantization error e (n) (e (n) = d (n) −d ′ (n)) obtained from the difference value d ′ (n) after the expansion step unit 14 described above. The FIR filter configured using the linear prediction coefficient is applied. The quantization error ns (n−1) that has been subjected to the filter processing is fed back to the difference calculation step unit 11 so that the approximate shape of the quantization error spectrum is shaped according to the filter characteristics.

  FIG. 11 is a conceptual diagram showing the operation of the noise shaping step unit 18. As shown in FIG. 11, the quantization error (conventional quantization error) when compression / expansion is performed by the conventional technique has a flat spectrum that is spread uniformly in frequency. Since a normal audio signal has a gentle slope as shown in FIG. 11, the audio signal-to-quantization error ratio (SNR) is high in the low-mid range and worsens toward the high range. However, by applying the noise shaping step unit 18 of the present invention, the flat spectrum of the quantization error is shaped to resemble the spectrum envelope of the audio signal. As a result, the SNR tends to be uniform in the entire band, so that the quantization error can be reduced audibly.

  As described above, according to the present invention, the spectral envelope of the quantization error spread uniformly in frequency can be brought close to the spectral envelope of the audio signal, and the quantization error can be reduced aurally. it can. Further, in the present invention, in the process of limiting the spectral envelope so that the fluctuation range of the spectral envelope characteristic falls within a predetermined range, the fluctuation range of the spectral envelope of the quantization error is limited to a specific frequency. The bias that the signal-to-quantization error ratio (SNR) is high only in the band can be improved. Further, when the input audio signal does not change and a constant value continues, the number of samples is counted. If the input audio signal is equal to or larger than a predetermined value, the stored filter coefficient is used instead of the filter coefficient calculated from the spectrum envelope characteristic. In such a case, the quality gap that occurs at the frame boundary can be prevented. In addition, since the present invention does not change any existing format, a conventional playback player and playback device can be used.

  The present invention includes a program for causing a computer to realize the functions of the above-described high-efficiency encoding apparatus. This program may be read from a recording medium and loaded into a computer, or may be transmitted via a communication network and loaded into a computer.

  The high-efficiency encoding program and the high-efficiency encoding apparatus according to the present invention are a quantization noise spectrum that spreads uniformly in frequency without any change to an existing format in encoding using a low sampling frequency. The envelope can be brought close to the spectrum envelope of the audio signal, and the quality of the audio signal can be improved. In addition, it is possible to improve the bias that the signal-to-quantization noise ratio (SNR) is high only in a specific frequency band. In addition, it is possible to prevent a quality gap that occurs at the processing frame boundary. Therefore, the present invention relates to a high-efficiency encoding program and a high-efficiency encoding apparatus for encoding speech or audio signals, and in particular, high-efficiency that provides a quality that can withstand listening by making the quantization noise spectrum audibly desirable. This is useful for an encoding program and a high-efficiency encoding device.

It is a block diagram which shows the structure of the highly efficient encoding apparatus which concerns on embodiment of this invention. It is a figure for demonstrating an example of the process of the framing step part in the highly efficient encoding apparatus which concerns on embodiment of this invention. It is a flowchart for demonstrating an example of the process of the filter characteristic determination step part in the highly efficient encoding apparatus which concerns on embodiment of this invention. It is a figure which shows the spectrum envelope which has the reverse characteristic of the data of the process frame in the highly efficient encoding apparatus which concerns on embodiment of this invention. It is a figure for demonstrating an example of the process of the spectrum envelope restriction | limiting step part in the highly efficient encoding apparatus which concerns on embodiment of this invention. It is a figure for demonstrating the problem by having DC data in the start frame in the highly efficient encoding apparatus which concerns on embodiment of this invention. It is a figure for demonstrating the problem by DC data existing in the end frame in the high-efficiency encoding apparatus which concerns on embodiment of this invention. It is a flowchart which shows an example of the process of the DC value search step part in the highly efficient encoding apparatus which concerns on embodiment of this invention. It is a flowchart which shows a part of process (start frame search step) of an example of a process of the DC value search step part in the high efficiency encoding device which concerns on embodiment of this invention. It is a flowchart which shows a part of process (end frame search step) of an example of a process of the DC value search step part in the high efficiency encoding device which concerns on embodiment of this invention. It is the conceptual diagram which showed the effect | action by the noise shaping step part in the highly efficient encoding apparatus which concerns on embodiment of this invention. It is a block diagram which shows the structure of the conventional high-efficiency encoding apparatus.

Explanation of symbols

0, 1, 2 Processing frame 11, 121 Difference calculation step unit (difference calculating means)
12, 122 Compression step unit (compression means)
13, 123 Step size (quantization width) determination step unit (quantization width determination means)
14, 124 Extension step part (extension means)
15, 125 Accumulation step part (accumulation means)
16 Framing step section (framing means)
17 Filter characteristic determining step unit (filter characteristic determining means)
18 Noise shaping step (noise shaping means)

Claims (8)

Input sampled at a lower sampling frequency compared to a sampling frequency that can change the spectral shape to increase quantization noise in areas that are not audibly audible based on equal loudness and / or masking characteristics A difference calculating step for calculating a difference value between the current sample and the previous sample in the audio signal;
A compression step of quantizing and encoding the difference value;
A decompression step that performs decoding and inverse quantization on the output obtained from the compression step;
Based on the encoding result of the compression step, a quantization width determination step for determining a quantization width at the time of quantization of the difference value calculated next;
Add the output obtained from the decompression step and the value of the previous sample when calculating the difference value, and the value of the previous sample when calculating the difference value in the next difference calculation step A highly efficient encoding program for causing a computer to execute a cumulative step to be calculated,
A framing step of generating a processing frame by dividing the input audio signal into a predetermined number of samples;
Obtain a spectrum envelope of the reverse characteristic of the spectral envelope for the generated processing frame, calculate a coefficient indicating the characteristic of the spectral envelope that cancels the characteristic of the obtained spectral envelope of the reverse characteristic, the calculated A filter characteristic determination step using a coefficient as a filter coefficient;
Quantization noise is calculated based on the input value to the compression step and the output value obtained from the decompression step, and the calculated quantization noise is calculated for each processing frame. A high-efficiency encoding program that causes a computer to further execute a noise shaping step of shaping a quantized noise spectrum by applying the corresponding filter coefficient of a filter composed of filter coefficients,
The filter characteristic determining step includes a spectral envelope limiting step for performing processing so that an amplitude fluctuation range determined by a difference between a maximum level and a minimum level of the spectrum envelope of the inverse characteristic of the processing frame falls within a predetermined limiting range. High-efficiency encoding program. Input sampled at a lower sampling frequency compared to a sampling frequency that can change the spectral shape to increase quantization noise in areas that are not audibly audible based on equal loudness and / or masking characteristics A difference calculating step for calculating a difference value between the current sample and the previous sample in the audio signal;
A compression step of quantizing and encoding the difference value;
A decompression step that performs decoding and inverse quantization on the output obtained from the compression step;
Based on the encoding result of the compression step, a quantization width determination step for determining a quantization width at the time of quantization of the difference value calculated next;
Add the output obtained from the decompression step and the value of the previous sample when calculating the difference value, and the value of the previous sample when calculating the difference value in the next difference calculation step A highly efficient encoding program for causing a computer to execute a cumulative step to be calculated,
A framing step of generating a processing frame by dividing the input audio signal into a predetermined number of samples;
Obtain a spectrum envelope of the reverse characteristic of the spectral envelope for the generated processing frame, calculate a coefficient indicating the characteristic of the spectral envelope that cancels the characteristic of the obtained spectral envelope of the reverse characteristic, the calculated A filter characteristic determination step using a coefficient as a filter coefficient;
Quantization noise is calculated based on the input value to the compression step and the output value obtained from the decompression step, and the calculated quantization noise is calculated for each processing frame. A high-efficiency encoding program that causes a computer to further execute a noise shaping step of shaping a quantized noise spectrum by applying the corresponding filter coefficient of a filter composed of filter coefficients,
The filter characteristic determining step includes a DC value search step for counting the number of samples followed by the value when there is no change in the value of the sample in the processing frame and the value continues.
A final filter that selects whether the filter coefficient to be applied to the quantization noise is the calculated filter coefficient or the saved filter coefficient stored in a predetermined storage area based on the number of the counted samples A high-efficiency encoding program comprising a selection step. Input sampled at a lower sampling frequency compared to a sampling frequency that can change the spectral shape to increase quantization noise in areas that are not audibly audible based on equal loudness and / or masking characteristics A difference calculating step for calculating a difference value between the current sample and the previous sample in the audio signal;
A compression step of quantizing and encoding the difference value;
A decompression step that performs decoding and inverse quantization on the output obtained from the compression step;
Based on the encoding result of the compression step, a quantization width determination step for determining a quantization width at the time of quantization of the difference value calculated next;
Add the output obtained from the decompression step and the value of the previous sample when calculating the difference value, and the value of the previous sample when calculating the difference value in the next difference calculation step A highly efficient encoding program for causing a computer to execute a cumulative step to be calculated,
A framing step of generating a processing frame by dividing the input audio signal into a predetermined number of samples;
Obtain a spectrum envelope of the reverse characteristic of the spectral envelope for the generated processing frame, calculate a coefficient indicating the characteristic of the spectral envelope that cancels the characteristic of the obtained spectral envelope of the reverse characteristic, the calculated A filter characteristic determination step using a coefficient as a filter coefficient;
Quantization noise is calculated based on the input value to the compression step and the output value obtained from the decompression step, and the calculated quantization noise is calculated for each processing frame. A high-efficiency encoding program that causes a computer to further execute a noise shaping step of shaping a quantized noise spectrum by applying the corresponding filter coefficient of a filter composed of filter coefficients,
The filter characteristic determining step includes a spectrum envelope limiting step for performing processing so that an amplitude fluctuation range determined by a difference between a maximum level and a minimum level of the spectrum envelope of the inverse characteristic of the processing frame falls within a predetermined limit range;
A DC value search step for counting the number of samples followed by the value if the value of the sample does not change in the processing frame and the value continues;
A final filter that selects whether the filter coefficient to be applied to the quantization noise is the calculated filter coefficient or the saved filter coefficient stored in a predetermined storage area based on the number of the counted samples A high-efficiency encoding program comprising a selection step. The step of calculating the spectral envelope of the inverse characteristic of the spectral envelope for the generated processing frame in the filter characteristic determination step includes a first linear prediction analysis step for obtaining a linear prediction coefficient by linear prediction analysis, and the linear prediction coefficient A spectral envelope calculation step for obtaining a spectral envelope by Fourier transform,
The step of calculating a coefficient indicating the characteristic of the spectral envelope that cancels the characteristic of the spectral envelope of the reverse characteristic obtained is obtained by performing an inverse Fourier transform on the spectral envelope obtained by the Fourier transform. The high-efficiency encoding program according to any one of claims 1 to 3, comprising a second linear prediction analysis step in which linear prediction analysis is performed using a coefficient to obtain a linear prediction coefficient.   The high-efficiency encoding program according to any one of claims 1 to 4, wherein the processing frame configured in the framing step is a processing frame obtained by combining the immediately preceding processing frames. Input sampled at a lower sampling frequency compared to a sampling frequency that can change the spectral shape to increase quantization noise in areas that are not audibly audible based on equal loudness and / or masking characteristics A difference calculating means for calculating a difference value between the current sample and the previous sample in the audio signal;
Compression means for performing quantization and encoding on the difference value;
Decompression means for performing decoding and inverse quantization on the output from the compression means;
Based on the encoding result of the compression means, quantization width determination means for determining a quantization width at the time of quantization of the difference value calculated next;
The output from the decompression means and the value of the previous sample when the difference value is calculated are added, and then the value of the previous sample when the difference value is calculated by the difference calculation means is calculated. A high-efficiency encoding device comprising accumulating means,
Framing means for generating a processing frame by dividing the input audio signal into a predetermined number of samples;
Obtain a spectrum envelope of the reverse characteristic of the spectral envelope for the generated processing frame, calculate a coefficient indicating the characteristic of the spectral envelope that cancels the characteristic of the obtained spectral envelope of the reverse characteristic, the calculated Filter characteristic determining means using the coefficients as filter coefficients;
Quantization noise is calculated based on an input value to the compression means and an output value from the decompression means, and the filter coefficient calculated for each processing frame with respect to the calculated quantization noise Noise shaping means for shaping the quantized noise spectrum by applying the corresponding filter coefficient of the filter comprising:
The filter characteristic determining unit includes a spectral envelope limiting unit that performs processing so that an amplitude fluctuation range determined by a difference between a maximum level and a minimum level of a spectrum envelope of the inverse characteristic of the processing frame falls within a predetermined limit range. High-efficiency encoding device. Input sampled at a lower sampling frequency compared to a sampling frequency that can change the spectral shape to increase quantization noise in areas that are not audibly audible based on equal loudness and / or masking characteristics A difference calculating means for calculating a difference value between the current sample and the previous sample in the audio signal;
Compression means for performing quantization and encoding on the difference value;
Decompression means for performing decoding and inverse quantization on the output from the compression means;
Based on the encoding result of the compression means, quantization width determination means for determining a quantization width at the time of quantization of the difference value calculated next;
The output from the decompression means and the value of the previous sample when the difference value is calculated are added, and then the value of the previous sample when the difference value is calculated by the difference calculation means is calculated. A high-efficiency encoding device comprising accumulating means,
Framing means for generating a processing frame by dividing the input audio signal into a predetermined number of samples;
Obtain a spectrum envelope of the reverse characteristic of the spectral envelope for the generated processing frame, calculate a coefficient indicating the characteristic of the spectral envelope that cancels the characteristic of the obtained spectral envelope of the reverse characteristic, the calculated Filter characteristic determining means using the coefficients as filter coefficients;
Quantization noise is calculated based on an input value to the compression means and an output value from the decompression means, and the filter coefficient calculated for each processing frame with respect to the calculated quantization noise Noise shaping means for shaping the quantized noise spectrum by applying the corresponding filter coefficient of the filter comprising:
The filter characteristic determining means, when there is no change in the sample value in the processing frame and the value continues, DC value search means for counting the number of samples followed by the value;
A final filter that selects whether the filter coefficient to be applied to the quantization noise is the calculated filter coefficient or the saved filter coefficient stored in a predetermined storage area based on the number of the counted samples A high-efficiency encoding device having a selection unit. Input sampled at a lower sampling frequency compared to a sampling frequency that can change the spectral shape to increase quantization noise in areas that are not audibly audible based on equal loudness and / or masking characteristics A difference calculating means for calculating a difference value between the current sample and the previous sample in the audio signal;
Compression means for performing quantization and encoding on the difference value;
Decompression means for performing decoding and inverse quantization on the output from the compression means;
Based on the encoding result of the compression means, quantization width determination means for determining a quantization width at the time of quantization of the difference value calculated next;
The output from the decompression means and the value of the previous sample at the time of calculating the difference value are added, and then the value of the previous sample at the time of calculating the difference value by the difference calculation means is calculated. A high-efficiency encoding device comprising accumulating means,
Framing means for generating a processing frame by dividing the input audio signal into a predetermined number of samples;
Obtain a spectrum envelope of the reverse characteristic of the spectral envelope for the generated processing frame, calculate a coefficient indicating the characteristic of the spectral envelope that cancels the characteristic of the obtained spectral envelope of the reverse characteristic, the calculated Filter characteristic determining means using the coefficients as filter coefficients;
Quantization noise is calculated based on an input value to the compression unit and an output value from the decompression unit, and the filter coefficient calculated for each processing frame with respect to the calculated quantization noise Noise shaping means for shaping a quantized noise spectrum by applying the corresponding filter coefficient of a filter comprising:
The filter characteristic determining means is a spectrum envelope limiting means for performing processing so that an amplitude fluctuation range determined by a difference between a maximum level and a minimum level of the spectrum envelope of the inverse characteristic of the processing frame falls within a predetermined limit width;
DC value search means for counting the number of samples followed by the value when the value of the sample does not change in the processing frame and the value continues,
A final filter that selects whether the filter coefficient to be applied to the quantization noise is the calculated filter coefficient or the stored filter coefficient stored in a predetermined storage area based on the number of the counted samples A high-efficiency encoding device having a selection unit.

Images