JP2008158301A - Signal processing device, signal processing method, reproduction device, reproduction method and electronic equipment - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To attain excellent sound quality by interpolating high frequency signals which are lost, for example, by compression coding. <P>SOLUTION: Self correlation calculation is performed in a frequency band of a prescribed frequency or more, and a correlation maximum shift amount when the calculated correlation value becomes the maximum is calculated, while a ratio of signals between frequency points which are separated from each other by the correlation maximum shift amount, is calculated. Then, the signal of each frequency point to be interpolated after a signal component is lost, is interpolated with a value calculated on the basis of the signal separated from each frequency point to be interpolated by the correlation maximum shift amount, and the ratio. Thereby, the high frequency signals which are lost by coding can be interpolated with a natural form by utilizing correlation. <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention relates to a signal processing apparatus, particularly a signal processing apparatus for correcting a component lost due to a voice compression coding process to improve sound quality and a method thereof. The present invention also relates to a playback apparatus and method, and an electronic device.

  The compression coding process of audio signals includes "quantization (PCM: Pulse Code Moduration)", "time correlation coding" using temporal continuity of audio signals, and "frequency correlation coding using human auditory characteristics" ", It can be realized by combining" entropy coding "using a bias in the probability of occurrence of codes obtained from these coding.

  As a method of compressing and encoding audio signals, MPEG (Moving Pictures Experts Group), ATRAC (Adaptive TRansform Acoustic Coding: registered trademark), AC-3 (Audio Code Number 3: registered trademark), WMA (Windows Media Audio: registered) The encoded signal is currently widely used for digital television broadcasting, digital audio players, WEB streaming, and the like.

Here, among the compression encoding processes, the frequency correlation encoding is an encoding process that greatly affects the compression rate and the sound quality. In the frequency correlation coding, the quantized PCM signal is orthogonally transformed from the time domain to the frequency domain, and a deviation of signal energy in the frequency domain is obtained. Encoding efficiency can be increased by encoding using this deviation. In addition, using the psychoacoustic characteristics for the signal after orthogonal transformation, the frequency band is divided into several bands, and some weighting is performed so as to minimize signal degradation in a band that is more easily perceived by humans. Thus, the overall coding quality can be improved.
In the encoding using the psychoacoustic characteristics, the corrected audible threshold value is obtained using the absolute audible threshold value and the relative audible threshold value determined by the masking effect. Based on this corrected audible threshold, bit allocation is performed for each divided band. A frequency component having a sound pressure equal to or lower than the corrected audible threshold is cut during encoding as a sound that cannot be perceived by humans. In addition, since the absolute audible threshold value increases in the high frequency band (hereinafter, also simply referred to as the high band), more frequency components in the high band are cut compared to the low frequency band (also referred to as the low band). . This is a frequency band limitation in the high band, which is peculiar to audio signal compression coding.

The compression coding method of the audio signal using such psychoacoustic characteristics is actively adopted in the MPEG standard. The tendency of compression encoding of audio signals can be determined by the technical strength of each encoder manufacturer, but for audio signals of digital television broadcasts adopting the MPEG standard, a certain frequency is obtained by the above encoding. It has also been confirmed that all subsequent high-frequency signals are cut off at the border, or that all signals in a certain divided band are cut even within the audible band.
In other words, sound quality deterioration (decrease) occurs due to such lack of signal components.

There are several prior arts for suppressing deterioration in sound quality due to signal loss in the high frequency part due to such compression encoding.
For example, “Frequency interpolation apparatus, frequency interpolation method, and recording medium” described in Patent Document 1 below discloses a method of replicating a high frequency signal using an existing audio signal.
Further, in the “frequency interpolation system, frequency interpolation device, frequency interpolation method, and recording medium” described in Patent Document 2, information of a missing signal is recorded in advance at the time of encoding, and the sound quality is used by using the information at the time of decoding. A method for decoding while maintaining the above is disclosed.

In addition to the lack of the high frequency part as described above, the quantization error associated with the reduction in the number of assigned bits can be cited as one of the sound quality degradations associated with compression coding.
In other words, in the coding using psychoacoustic characteristics, the number of bits allocated to the frequency band that is considered inaudible to be heard is reduced, thereby trying to realize efficient information compression processing while suppressing deterioration in sound quality. Yes. However, it has been confirmed that in such a frequency band to which a low bit number is assigned, the decoding accuracy is deteriorated due to the low bit number and the difference from the original audio signal is increased. Yes. In other words, this tends to cause deterioration (decrease) in sound quality in a band to which a low bit number is assigned.

As a prior art for suppressing the deterioration of sound quality due to the quantization error accompanying the reduction of the number of allocated bits, for example, the following Patent Document 3 can be cited.
In the “quantization error correction method and apparatus and audio information decoding method” described in Patent Document 3, a range of values that can be originally taken by a speech signal to be corrected is calculated, and encoded signals in adjacent frequency bands are calculated. The correction value is calculated using the least square method. A method is disclosed in which if the correction value is within the range, it is replaced with an existing signal, and if it is out of the range, replacement with the existing signal is performed using the minimum value / maximum value of the range.

Furthermore, as a cause of sound quality degradation accompanying compression coding, an inter-band error due to a reduction in the number of allocated bits as described above can be cited. In other words, in a band where the number of allocated bits is reduced, a quantization error occurs between the bands due to the difference in the number of allocated bits at the boundary portion between adjacent bands with a relatively large number of allocated bits. .
As described above, in the portion where the quantization error between the bands occurs, the continuity of the waveform is lost and the sound quality is deteriorated.

JP 2001-356788 A JP 2002-73096 A JP 2001-102939 A

Here, first, in the above-mentioned Patent Document 1 that deals with signal loss in a high frequency part, a reference band that is a generation source of a high frequency signal is specified from a band in which a signal exists, and information based on the reference band is used. A method of generating and adding a high-frequency signal is disclosed.
However, in this method, when specifying the reference band, it is necessary to divide a band in which a signal exists and to create a combination by the number of the divided number, thereby obtaining a correlation.
Further, since the processing amount until the correlation is obtained changes in each frame, the processing amount and the processing time are changed depending on the input signal.

  Also, in the invention described in the above-mentioned Patent Document 2 that also takes measures against signal loss in the high frequency part, a common algorithm is required on the encoder side and the decoder side, and therefore the versatility is poor. Yes.

  In this way, with the conventional method used for interpolating (adding) the high-frequency component lost due to encoding, there is a problem of a bias in processing amount, an increase in processing load, and a problem in versatility. It was something to have.

Further, in the method described in Patent Document 3 for correcting the quantization error due to the reduction in the number of allocated bits, several cases are required depending on the encoded signals in the adjacent frequency bands, and the processing load increases accordingly. To do. Furthermore, in calculating the “correction value”, a quadratic curve is obtained for each signal, but in order to ensure the accuracy of such a correction value, the calculation is performed by sampling a signal over a relatively long time. It is necessary to do this, and the processing amount tends to increase also in this respect.
That is, in the conventional method for correcting the quantization error due to the reduction in the number of allocated bits, it is required to reduce the processing load for correction.

  Furthermore, regarding the above-described quantization error between bands, a technique for correcting it has not been established at present, and sound quality cannot be improved in this respect.

Therefore, in the present invention, first, in view of the above-described problem with the lack of high-frequency signals, the signal processing apparatus is configured as follows.
That is, correlation calculation means is provided for performing autocorrelation calculation when a self-signal is sequentially shifted with respect to the self-signal with respect to a signal in a band of a predetermined frequency or higher in an audio signal subjected to predetermined information compression processing.
Further, a shift amount detecting means for obtaining a maximum correlation shift amount when the correlation is highest based on the result of the autocorrelation calculation is provided.
In addition, a ratio calculation unit that calculates a ratio of each amplitude value at each frequency point separated by an amount based on the correlation maximum shift amount obtained by the shift amount detection unit in a band of the predetermined frequency or higher is provided.
Further, the amplitude value of each interpolation target frequency point where the signal is missing in the band of the predetermined frequency or higher is the amplitude value at the frequency point separated from each of the interpolation target frequency points by an amount based on the maximum correlation shift amount. And interpolation means for interpolating with values calculated based on the ratio.

  In the present invention, autocorrelation calculation is performed in a band of a predetermined frequency or higher, and the shift amount (correlation maximum shift amount) that maximizes the correlation value obtained as a result is obtained. The value of the maximum correlation shift amount indicates that the autocorrelation is the highest in the band above the predetermined frequency when the correlation is shifted by that value. For this reason, in interpolating the signal of each interpolation target frequency point after the signal component is lost, after calculating the ratio of the amplitude value between the frequency points separated by the correlation maximum shift amount as described above, If the signal of each interpolation target frequency point is interpolated with the value calculated based on the amplitude value at the frequency point separated from each interpolation target frequency point by the above correlation maximum shift amount and the above ratio, it is based on the signal of the correlated part. It is possible to interpolate a high frequency with a signal that does not feel strange.

Further, in the present invention, in view of the above-described problem regarding quantization error due to the reduction in the number of allocated bits, the signal processing apparatus is configured as follows.
In other words, the audio signal that has been subjected to the predetermined information compression processing is provided with a prediction signal generation unit that generates a prediction signal obtained by predicting the original signal before compression.
The audio signal further includes error candidate part detection means for detecting a part having an amplitude value of a predetermined value on the frequency axis as an error candidate part.
Furthermore, a replacement means is provided for replacing the amplitude value of the error candidate portion based on the value of the prediction signal based on the result of comparing the amplitude value in the error candidate portion with the value of the prediction signal.

  By detecting a portion having a predetermined value on the frequency axis as described above, and replacing the amplitude value of the portion based on the value of the predicted signal based on the comparison result with the value of the predicted signal, It is possible to appropriately detect a portion where a large quantization error due to the small number of assigned bits occurs and correct the amplitude value of the error portion with a more probable value according to the prediction signal.

Furthermore, in the present invention, in view of the above-described problem of quantization error between bands, the signal processing apparatus is configured as follows.
In other words, the audio signal that has been subjected to the information compression processing that determines the resolution for assigning the number of bits in a predetermined frequency band unit is provided with a prediction signal generation unit that generates a prediction signal that predicts the original signal before compression.
The audio signal further includes boundary portion detection means for detecting a boundary portion between successive frequency bands having different resolution values.
Furthermore, based on the result of comparing the amplitude value at the boundary portion with the value of the prediction signal, a replacement means for replacing the amplitude value of the boundary portion based on the value of the prediction signal is provided.

  As described above, a boundary portion between consecutive frequency bands having different resolution values is detected, and the amplitude value of the boundary portion is replaced based on the value of the prediction signal based on the result of comparison with the prediction signal. If this is the case, it is possible to properly detect a part where the continuity between bands is lost due to a large quantization error between bands due to the difference in the number of allocated bits, and the amplitude value of that part is a more probable value based on the predicted signal. It can be corrected.

As described above, according to the present invention for high-frequency interpolation, based on the autocorrelation calculation result for the remaining high-frequency signal, the lost portion of the high-frequency signal is interpolated. Thus, the lost signal component can be correctly interpolated.
Further, in the present invention, only a relatively simple four arithmetic operations such as autocorrelation calculation and ratio calculation and calculation of a signal value to be added using the maximum correlation shift amount and ratio are performed in the interpolation process. Good. In this regard, the processing burden is reduced as compared with the conventional method of dividing the band in which the signal exists in specifying the reference band, and combining the number of the divided number to obtain the correlation. It can be much lighter.

  Further, according to the interpolation method of the present invention, in the interpolation process, it is only necessary to always perform autocorrelation calculation, ratio calculation, and signal addition based on the correlation maximum shift amount and ratio. That is, the content can be the same as the interpolation processing for each predetermined frame unit. From this point, the processing contents change for each frame as in the prior art, and the inconvenience that the processing amount and processing time change depending on the input signal does not occur.

  In addition, since the high-frequency interpolation method of the present invention can be a process independent of the decoding process, it is not particularly necessary to use a decoding algorithm common to the encoding side, so that the problem of loss of versatility does not occur. Can be.

  Further, according to the present invention for the above-mentioned quantization error correction, a part where a large quantization error due to a small number of allocated bits is detected is properly detected, and the amplitude value of the error part is determined according to the prediction signal. It can be corrected with a certain value. That is, it is possible to effectively suppress sound quality deterioration due to compression coding.

According to the present invention, in order to obtain such a sound quality improvement effect, a predicted signal is generated, and the value is compared with the value of the predicted signal based on the result of comparing the value with the amplitude value of the audio signal. It can only be necessary.
According to this, for example, a range of values that can be originally taken is calculated as in the prior art, a correction value is calculated from encoded signals in adjacent frequency bands using the least square method, and the correction value is a value within the range. Compared with the case where high sound quality is achieved by the method of replacing the existing signal if there is, if it is outside the range, replacing with the existing signal using the minimum and maximum values of the range, The processing burden can be significantly reduced.

  Furthermore, according to the present invention relating to the quantization error between bands, a part where a large quantization error occurs between bands due to the difference in the number of allocated bits and the continuity between bands is impaired is appropriately detected. The amplitude value can be corrected by a more probable value based on the prediction signal. That is, it is possible to effectively improve the discontinuity of the waveform in the interband part caused by the difference in the number of assigned bits, and as a result, it is possible to improve the sound quality.

Hereinafter, the best mode for carrying out the invention (hereinafter referred to as an embodiment) will be described.
In the following description of each embodiment (first to third embodiments), ISO / IEC 13818- called MPEG-2 AAC (Moving Pictures Experts Group-2 Advanced Audio Coding) is used as an audio encoding method. As an example, a coding method of 7 standards is adopted and the decoding process is performed.
In the following, MPEG-2 AAC is also simply referred to as AAC.
The ISO is an abbreviation for International Organization for Standardization, and IEC is an abbreviation for International Electrotechnical Commission.

[Outline of AAC encoding processing]
First, the outline of the AAC encoding process will be described as a premise for explaining the embodiment.
The voice encoding process based on the AAC method is a frequency correlation encoding process, and based on psychoacoustics, a sound region that cannot be perceived by humans is not converted into data, thereby increasing the compression effect. According to AAC encoding, for example, in the case of 2-channel stereo sound, a CD (Compact Disc) sound quality can be obtained even with a transmission rate of about 96 kilobits / second, and a compression rate of about 1/15 (1/15). Is obtained.

  In the AAC system, based on the result of the psychoacoustic analysis described above, (1) gain adjustment processing → (2) adaptive block length switching MDCT processing → (3) TNS processing → (4) intensity stereo coding processing → ( 5) Prediction processing → (6) M / S stereo processing → (7) After scaling processing is performed, until (8) quantization processing and (9) Huffman encoding processing fall below the allocated number of bits Iterates to form encoded audio data. Actually, a final encoded audio signal (AAC bit stream) is formed by adding various coefficients to be added in these processing steps.

Specific processing contents are shown below.
First, the input speech signal before the encoding process is gain-adjusted and is blocked for each predetermined number of samples, and this is processed as one frame. In the encoding device, an input frame is subjected to FFT (Fast Fourier Transform) processing in a psychoacoustic analysis unit to obtain a frequency spectrum, and auditory masking is calculated based on the frequency spectrum, and allowable quantization noise for each preset frequency band is calculated. A parameter called PE and Perceptual Entropy (PE) for the frame is obtained.

  Psychological auditory entropy corresponds to the total number of bits required to quantize the frame so that the listener does not perceive noise. In addition, psychological entropy has a characteristic that it takes a large value when the signal level suddenly increases like an attack portion of a voice signal. Therefore, the transform block length of MDCT (Modified Discrete Cosine Transform) is determined based on the sudden change portion of the psychological entropy value.

  The MDCT process converts an audio signal input with a block length determined by the psychoacoustic analysis unit into a frequency spectrum (hereinafter referred to as an MDCT coefficient). The process of adaptively switching the transform block length according to the input signal (adaptive block switching) is a process necessary for suppressing auditory harmful noise called pre-echo.

  The MDCT coefficient formed by the MDCT processing is subjected to TNS (Temporal Noise Shaping) processing. In this TNS process, the MDCT coefficient is viewed as if it is a signal on the time axis, linear prediction is performed, and prediction filtering is performed on the MDCT coefficient. By this TNS process, the quantization noise included in the waveform obtained by inverse MDCT on the decoding process side is gathered at a large signal level.

For MDCT coefficients that have been subjected to TNS processing, intensity stereo coding, that is, the sound in the high frequency region has only one coupling channel that combines the left channel (L channel) and the right channel (R channel). Processing is performed to prevent transmission.
Intensity-stereo-encoded MDCT coefficients are predicted for the current MDCT coefficient from the quantized MDCT coefficients in the past two frames for each MDCT coefficient, and the prediction residual is obtained. This predicted MDCT coefficient is transmitted by M / S stereo processing to transmit the left and right channel sum signal (M = L + R) and difference signal (S = LR), or the left and right channels respectively (L channel and R channel). Each channel) is determined and output.

  The MDCT coefficients output by the M / S stereo process are grouped (scaled) by a plurality of preset frequency bands, and are quantized in units. These groups of MDCT coefficients are called scale factor bands (sfb). This sfb is set to be narrow on the low frequency side and wide on the high frequency side in accordance with the auditory characteristics.

  In the quantization process, quantization is performed with the goal of being below the allowable quantization noise power for each sfb obtained by the psychoacoustic part. The quantized MDCT coefficients are further subjected to Huffman coding to reduce redundancy. This quantization / Huffman coding process is performed in an iterative loop, and is repeated until the amount of code actually generated falls below the number of bits assigned to the frame.

  As described above, the AAC encoding method is based on the result of psychoacoustic analysis. (1) Gain adjustment processing → (2) Adaptive block length switching MDCT processing → (3) TNS processing → (4) Intensity stereo coding → (5) Prediction process → (6) M / S stereo process → (7) After the scaling process, (8) quantization process and (9) Huffman encoding process are assigned to the number of bits. By repeating the process until it falls below, the encoded voice data is formed.

  The above-described AAC speech coding process is described in detail in various documents such as “Introduction to Digital Television Technology” by Yutaka Takada, Satoshi Asami, Yoneda Publishing, pages 112 to 124, or Web pages. Explained.

Further, the gain adjustment process, the TNS process, the intensity / stereo coding process, the prediction process, and the M / S stereo process are optional processes, and are not performed in the entire AAC coding process. That is, the gain adjustment process, the TNS process, the intensity stereo coding process, the prediction process, and the M / S stereo process are processes performed only when the option process is selected. In the embodiment described below, a case where an encoded speech signal that has been subjected to the above-described option processing and is compression-encoded is processed will be described as an example.

<First Embodiment>

FIG. 1 is a block diagram showing an internal configuration of a playback apparatus 1 as a first embodiment of a playback apparatus (electronic device) according to the present invention.
The playback device 1 is configured as an audio player capable of decoding and playing back a compressed and encoded audio signal stored in a storage device such as an HDD (Hard Disc Drive) or a flash memory as the illustrated storage unit 2.
In addition to the storage unit 2, the playback device 1 includes a demodulation unit 3, a compression code decoding unit 4, a DSP (Digital Signal Processor) 5, a bus 6, a system controller 7, an operation unit 8, and a display unit 9. Yes.

  First, in the storage unit 2, a compressed encoded audio signal is stored in a state where predetermined processes such as a run length limited encoding process and an error correction code adding process are performed. The demodulator 3 performs predetermined demodulation processing such as decoding of the run-length limited code and error correction processing based on the error correction code on the read signal from the storage unit 2 to obtain a compressed encoded audio signal.

The compression-encoded audio signal obtained by the demodulation process of the demodulation unit 3 is supplied to the compression code decoding unit 4 where the compression code decoding process is performed. As can be understood from the above description, the present embodiment assumes that the compression-encoded audio signal is an AAC compression-encoded audio signal, and the compression encoding / decoding unit 4 performs decoding corresponding to the AAC system. It is comprised so that a process may be performed. That is, it is configured to decode the AAC compression-encoded audio signal so that the audio can be output.
The internal configuration of the compression code decoding unit 4 will be described later.
In the case of the first embodiment, the compression code decoding unit 4 is provided with a high-frequency interpolation unit 4a as shown in the figure, which will also be described later.

  The audio signal obtained through the decoding process of the compression coding / decoding unit 4 is supplied to the DSP 5, where necessary audio signal processing (for example, volume adjustment, various acoustic effect addition processes, etc.) is performed and then illustrated. It is supplied to the output terminal Tout and output externally.

The system controller 7 is constituted by a microcomputer including, for example, a CPU (Central Processing Unit), a ROM (Read Only Memory), a RAM (Randam Access Memory), and the like, and is connected to the storage unit 2 via the bus 6 shown in the figure. The playback apparatus 1 is controlled in its entirety by exchanging control signals and various data with the demodulator 3, compression code decoder 4, and DSP 5.
For example, the system controller 7 executes signal read control of the storage unit 2 in response to an operation input from the operation unit 8 described later. In addition, for example, processing for setting various parameters of audio signal processing in the DSP 5 according to an operation input from the operation unit 8 or the like is also performed.

  The operation unit 8 is an input device such as an operation key (not shown) provided so as to be exposed on the outer surface of the housing of the playback apparatus 1, for example, and the user performs various operation inputs and data inputs. Information input by the operation unit 8 is transmitted to the system controller 7 as operation or data input information. The system controller 7 performs necessary calculations and control corresponding to the input information.

  The display unit 9 includes a display panel such as a liquid crystal panel, for example, and displays various information to the user. In this case, various information is displayed based on an instruction from the system controller 7.

[Configuration of compression coding / decoding unit]
FIG. 2 shows the internal configuration of the compression code decoding unit 4 shown in FIG.
First, the components of the compression code decoding unit 4 are roughly divided into a format analysis unit 10, an inverse quantization processing unit 11, a stereo processing unit 12, an adaptive block length switching inverse MDCT unit 13, and a gain control unit 14. be able to. In addition, the compression code decoding unit 4 in this case is provided with a high-frequency interpolation unit 4a as shown in FIG.

  The inverse quantization processing unit 11 includes a Huffman decoding unit 11a, an inverse quantization unit 11b, and a rescaling unit 11c. The stereo processing unit 12 includes an M / S stereo processing unit 12a, a prediction processing unit 12b, an intensity / stereo processing unit 12c, and a TNS unit 12d.

First, the encoded audio signal (bit stream) from the demodulator 3 shown in FIG. 1 is supplied to the format analyzer 10. The format analysis unit 10 separates the supplied encoded audio signal into MDCT coefficients and other parameters and control information. The MDCT coefficients are supplied to the Huffman decoding unit 11a in the inverse quantization processing unit 11.
In addition, the format analysis unit 10 forms a control signal for each unit based on parameters and control information extracted from the bit stream of the encoded audio signal, and compresses and decodes the control signal as indicated by a broken line arrow in the figure. By supplying to each part in the part 4, processing in each part is controlled.

The MDCT coefficients separated by the format analysis unit 10 as described above are subjected to a process that is the reverse of the process at the time of the AAC encoding described above, whereby the encoded audio signal is decoded.
Specifically, first, the Huffman decoding unit 11a performs a Huffman decoding process on the MDCT coefficients supplied from the format analysis unit 10. Next, after the inverse quantization process is performed in the inverse quantization unit 11b, the MDCT coefficient before quantization is restored by performing the rescaling process in the rescaling unit 11c.

The MDCT coefficients restored to the pre-quantization state in the inverse quantization processing unit 11 in this way are supplied to the M / S stereo processing unit 12a in the stereo processing unit 12.
In the M / S stereo processing unit 12a, the MDCT coefficients of the left channel (Lch) and the right channel (Rch) are restored. The left and right two-channel MDCT coefficients are processed by the next prediction processing unit 12b to be restored to the MDCT coefficients before data compression by the prediction processing at the time of encoding, and further, the intensity stereo processing unit 12c The city stereo decoding process is performed, and the sound in the high frequency region is also distributed to the MDCT coefficients of the left and right channels. Thereafter, prediction filtering is removed in the TNS unit 12d, and the MDCT coefficient immediately after the MDCT processing at the time of encoding is restored.

In the case of the first embodiment, the MDCT coefficients restored to the state immediately after being subjected to MDCT processing in the stereo processing unit 12 in this way are subjected to high-frequency interpolation by the high-frequency interpolation unit 4a as shown in the figure. Later, the adaptive block length switching inverse MDCT unit 13 is supplied.
As described above, the contents of the high-frequency interpolation processing by the high-frequency interpolation unit 4a and the internal configuration thereof will be described later.

The adaptive block length switching inverse MDCT unit 13 performs inverse MDCT processing on the MDCT coefficient (frequency domain audio signal) supplied via the high frequency interpolating unit 4a as described above, thereby performing a time axis domain audio signal (time Audio signal: an audio signal in a state where audio output is possible), and this is supplied to the gain control unit 14.
The temporal audio signal obtained by the adaptive block length switching inverse MDCT unit 13 is supplied to the DSP 5 shown in FIG. 1 after gain adjustment is performed by the gain control unit 14 at the next stage.

[High-frequency interpolation operation]
As described above, the compression encoding / decoding unit 4 performs a decoding process on the encoded audio signal that is encoded by the AAC method to obtain an audio signal that can be output as an audio signal.
However, in general, since compression compression coding employs a method using psychoacoustic analysis as described above as frequency correlation coding, there is a possibility that a high-frequency sound signal is lost and the sound quality is deteriorated. It is expensive. That is, even if an audio signal is obtained by decoding such a compression-encoded audio signal, there is a high possibility that the sound quality has deteriorated.

  Therefore, in the first embodiment, the sound quality is improved (that is, the sound quality is improved) by interpolating the audio signal of the high frequency part that has been lost due to the encoding. As a configuration for that purpose, the high-frequency interpolation unit 4a shown in FIG. 2 (FIG. 1) is provided.

First, the high-frequency interpolation operation of the first embodiment will be described with reference to FIGS.
The operation described below is performed for the MDCT coefficient of the AAC1 frame (1024 samples) output from the stereo processing unit 12 shown in FIG. 2, and in practice, the operation for each frame is repeated. As a result, the sound quality of the temporal sound signal can be improved.
For example, when the sampling frequency is 44.1 kHz, the MDCT coefficient for one AAC frame is about 0.023 sec (1024/44100 sec) in terms of time.

First, the outline of MDCT coefficients for one AAC frame will be described with reference to FIG. In FIG. 3, the MDCT coefficient for one frame is schematically shown with the vertical axis representing amplitude (MDCT coefficient value) and the horizontal axis representing frequency.
In the AAC system, when the sampling frequency is 48 kHz or 44.1 kHz, one frame is divided into 49 scale factor bands (sfb). Specifically, sfb numbers are assigned to sfb [0], sfb [1], sfb [2],... Sfb [48] in order from the low frequency side to the high frequency side.
As described above, the bandwidth of each scale factor band is set to be narrower on the low frequency side and wider on the high frequency side based on psychoacoustic characteristics. Specifically, the bandwidth of sfb [0] on the lowest side is “4” (4 MDCT coefficients), and the bandwidth gradually increases toward the high side.
In the case of AAC, the bandwidth of sfb is made constant when the frequency becomes high to some extent. Specifically, the bandwidth gradually increases from sfb [0] to sfb [28], but thereafter sfb [29] (sfb from MDCT coefficient number 320) to sfb [47] ( Up to MDCT coefficient number 928 up to sfb), the bandwidth is fixed at “32”. However, as shown in the figure, only the last sfb [48] has a bandwidth = “96”.

In the high-frequency interpolation operation of the first embodiment, after the bandwidth of sfb becomes constant as described above, a band during which a signal exists (that is, a band until a signal is lost due to encoding) is obtained. As shown in the figure, it is set as a “high frequency sub-band”.
By the way, in the case of the AAC method (bit rate 128 kbps), the signal is lost by encoding in the band around MDCT coefficient number = 650 and after.

In the first embodiment, with respect to the high frequency sub-band set in this way, the signal in the band is interpolated using the signal in the band thereafter.
The outline of the interpolation operation is as follows.
First, autocorrelation calculation is performed within the specified high frequency subband. That is, within the high frequency subband, the degree of correlation with the self signal is determined by calculation to what extent the shift is made in the frequency direction. In this specification, the shift amount when autocorrelation becomes the highest is called the correlation maximum shift amount.
Then, together with the correlation maximum shift amount, a value of Rate (ratio) is calculated as information necessary for generating an additional signal for interpolating a portion where the signal is lost. Specifically, the ratio of the amplitude value of each MDCT coefficient number position separated by the correlation maximum shift amount in the high frequency subband is calculated as the value of Rate.
Then, using the maximum correlation shift amount and the value of Rate, interpolation is performed for the portion where the signal is lost. That is, the MDCT coefficient value at each MDCT coefficient number position (each frequency point) in the band where the signal is lost is changed to the MDCT coefficient at the position returned from the MDCT coefficient number by the correlation maximum shift amount and the value of the Rate. Is interpolated with the value calculated based on.

4 to 7, a specific procedure of the high-frequency interpolation operation as the first embodiment will be described.
First, FIG. 4 schematically shows a peak value detection operation to be performed when performing autocorrelation calculation. 4 also shows the MDCT coefficients for one frame when the vertical axis represents amplitude (MDCT coefficient value) and the horizontal axis represents frequency, as in FIG.

In FIG. 4, first, when detecting the peak value, each sfb in the high frequency sub-band is divided into four equal parts. Then, the MDCT coefficient having the largest value (amplitude) in each band (hereinafter referred to as a divided band) corresponding to eight MDCT coefficients, which is obtained by equally dividing each sfb into four, is divided. It is detected as the peak value Peak of the band.
For the peak value Peak, as shown in the drawing, the peak value Peak [0], the peak value Peak [1], the peak value Peak [2],..., The peak value Peak [n] in order from the lowest band. ]
For confirmation, “n” in Peak [n] indicates the number of the last four equally divided bands with amplitude (that is, the number of the divided band). For example, if the signal is lost on the higher frequency side than sfb [30], the peak value Peak [40] is detected from 320/8 = 40.

When the peak value Peak is detected in this way, autocorrelation calculation within the high frequency sub-band is performed using those values.
FIG. 5 is a schematic diagram for explaining the autocorrelation calculation, and shows the MDCT coefficients for one frame when the vertical axis is amplitude (MDCT coefficient value) and the horizontal axis is frequency as in FIG. Yes.

In FIG. 5, in performing autocorrelation calculation, first, the shift amount j is set. The value of the shift amount j is a value for determining how much the self-signal is shifted and starting the autocorrelation calculation. That is, from the position where the self signal is shifted by the value of the shift amount j with respect to the self signal of the high frequency subband, which is configured by each peak value Peak in the high frequency subband as illustrated. The autocorrelation calculation is started.
In this example, as the value of the shift amount j, for example, j = 20 is set.

The specific contents of the autocorrelation calculation are as follows when the number of the divided band in which the peak value Peak is detected is “i” and the last divided band number in which the signal (MDCT coefficient) exists is “N”. It can be expressed by the following formula 1.

  By this autocorrelation calculation, a correlation value at each shift position is obtained. Among these correlation values, the value of the shift amount j when the value is maximum is the value of the maximum correlation shift amount when the correlation is the highest. Hereinafter, the value of the maximum correlation shift amount is set to “k”.

In the description so far, it is assumed that the high frequency signal component is always included in the original signal. However, in reality, the original signal includes only a relatively low frequency component, and the high frequency component is not included. It can be assumed that it is not included at all. In such a case, if high-frequency interpolation is performed, a signal component of an original part is added unnaturally, which may lead to deterioration of sound quality.
Therefore, in the first embodiment, it is determined whether or not the high frequency component is originally included in the original signal, and based on the result, it is determined whether or not high frequency interpolation described below is executed. Yes.
In this case, the determination whether or not the high frequency component is originally included in the original signal uses the correlation value obtained by the autocorrelation calculation as described above. Specifically, among the calculated correlation values, the top five correlation values having the largest value are selected, and when the sum of them is equal to or greater than a predetermined threshold value (for example, 1.0), If the high frequency component is included, the subsequent operation for high frequency interpolation is executed.
On the other hand, when the sum of the top five correlation values having a large value is less than the above threshold value (1.0), it is assumed that the high frequency was not originally included in the original signal, and the subsequent operation for high frequency interpolation is performed. Do not do.

Next, FIG. 6 shows the rate calculation method. 6 also shows MDCT coefficients for one frame, where the vertical axis represents amplitude (MDCT coefficient value) and the horizontal axis represents frequency, as in FIG.
The value of the maximum correlation shift amount “k” obtained as described above is a value indicating that the correlation with the self becomes the highest when shifted by this value “k” in the high frequency subband. . Therefore, using this maximum correlation shift amount “k”, the ratio (Rate) of the amplitude at the position shifted by “k” to the high frequency side is obtained.

In the case of the present embodiment, the calculation of Rate is performed based on Peak [n] detected in the last division band (division band [n]) in which a signal exists.
Specifically, as shown in the figure, first, Peak [n] of the divided band [n] and the divided band [n] at the position returned from the divided band [n] to the low frequency side by the correlation maximum shift amount “k” [ The amplitude ratio of m] to the peak value Peak [m] is obtained by calculation. That is, Peak [m] / Peak [n] is calculated.
Similarly, the amplitude ratio (Peak [m−1] / Peak) between the peak value Peak [n−1] in the divided band [n−1] and the peak value Peak [m−1] in the divided band [m−1]. [n-1]) is calculated.
Similarly, the same ratio calculation is performed for the predetermined band x up to the divided band [nx]. For example, in this example, the predetermined value x = 5, and the amplitude ratio (Peak [m-5] / Peak [n-5]) between the peak value Peak [n-5] and the peak value Peak [m-5]. Calculate up to.
Then, an average of the five amplitude ratio values to be obtained as a result is calculated as a final Rate value.

When the correlation maximum shift amount “k” and the value of Rate are calculated by the above procedure, as shown in FIG. 7 below, for each MDCT coefficient number position in the band where there is no signal, that is, the band to be interpolated. Interpolate (add) actual MDCT coefficients.
7 also shows MDCT coefficients for one frame when the vertical axis represents amplitude (MDCT coefficient value) and the horizontal axis represents frequency, as in FIG.

First, the value of the correlation maximum shift amount “k” obtained previously is a value obtained on the basis of the peak value Peak, and is thus a shift amount converted in units of divided bands (that is, units of 8 MDCT coefficients). ing. On the other hand, since the actual signal addition (interpolation) is performed for each MDCT coefficient number position, it is necessary to return it to a value in units of MDCT coefficient numbers. Specifically, by setting k × 8, the correlation maximum shift amount in units of MDCT coefficients can be restored.
As a specific interpolation operation, the value of each MDCT coefficient number position in the band where the signal disappears is changed to the value of the MDCT coefficient at the position returned to the low frequency side by k × 8 from each MDCT coefficient number position. Interpolate with the value calculated based on the value of Rate. In other words, the MDCT coefficient is added to each MDCT coefficient number position with the value calculated in this way.
In this case, the value of Rate is m / n as described with reference to FIG. 6, that is, the high frequency side where the amplitude is smaller is used as the denominator. The value of each MDCT coefficient position is interpolated by the value obtained by dividing the value of each MDCT coefficient position by the value of Rate.

[Internal configuration of high-frequency interpolation unit]
FIG. 8 is a block diagram showing a configuration in the high-frequency interpolation unit 4a for realizing the high-frequency interpolation operation of the first embodiment described above.
As shown in the figure, in the high-frequency interpolation unit 4a, a Peak detection unit 21, an autocorrelation calculation unit 22, a correlation maximum shift amount detection unit 23, a rate calculation unit 24, a high-frequency signal addition unit 25, and an interpolation determination unit 26 are included. Is provided.

First, also in FIG. 8, the stereo processing unit 12 and the adaptive block length switching inverse MDCT unit 13 shown in FIG. 2 are shown.
The MDCT coefficient for each AAC 1 frame unit output from the stereo processing unit 12 is input to the Peak detection unit 21 as shown in the figure, and is branched into an autocorrelation calculation unit 22, a rate calculation unit 24, and a high frequency range. It is also supplied to each of the signal adding units 25.

In the Peak detection unit 21, a band from a predetermined sfb (in this case, sfb [29]) to the last sfb having an amplitude is determined as a high frequency sub-band based on the supplied MDCT coefficient for one frame. As described above, each sfb in this high frequency sub-band is divided into four and divided into respective divided bands, and then the peak value Peak of the MDCT coefficient in each divided band is detected. That is, the peak value Peak [0] to peak value Peak [n] as described with reference to FIG. 4 are detected.
Each detected peak value Peak [0] to peak value Peak [n] is supplied to the autocorrelation calculation unit 22 and the rate calculation unit 24.

  The autocorrelation calculation unit 22 inputs the peak value Peak [0] to the peak value Peak [n] supplied from the Peak detection unit 21, and the divided band in which the peak value Peak is detected as described above. When the number is “i” and the number of the last divided band in which the amplitude of the MDCT coefficient exists is “N”, the calculation according to the above-described equation 1 is performed to calculate the correlation value at each shift position.

  The correlation maximum shift amount detection unit 23 inputs the correlation value at each shift position calculated by the autocorrelation calculation unit 22 and information on the shift amount, and correlates the shift amount when the correlation value becomes maximum. It is detected as the maximum shift amount “k”. The detected value of the maximum correlation shift amount “k” is supplied to the rate calculation unit 24 and the high frequency signal addition unit 25.

  Further, the correlation value at each shift position calculated by the autocorrelation calculation unit 22 is also supplied to the interpolation determination unit 26. The interpolation determination unit 26 selects the top five correlation values in descending order of the correlation values from the autocorrelation calculation unit 22, and their sum is equal to or greater than a predetermined threshold value (1.0 in this case). It is determined whether or not. When it is determined that the calculated sum value is equal to or greater than the threshold value, it is determined that the high frequency component is included in the original signal, and the operation is performed on the rate calculation unit 24 and the high frequency signal addition unit 25. A control signal (execution instruction signal) for instructing is supplied. On the other hand, if it is determined that the calculated sum is less than the threshold value, it is determined that the original signal originally did not include a high frequency component, and the rate calculation unit 24 and the high frequency signal addition unit 25 A control signal (non-execution instruction signal) for instructing not to execute the operation is supplied.

The rate calculation unit 24 inputs the peak values Peak [0] to Peak [n] from the peak detection unit 21 and the correlation maximum shift amount “k” from the correlation maximum shift amount detection unit 23, and the above FIG. The value of Rate is calculated by the method described in (1). That is, after calculating the amplitude ratio of the peak value Peak of the divided bands [n] to [n-5] with the peak value Peak in the divided band returned to the low frequency side by the maximum correlation shift amount “k”. Then, the average of the five amplitude ratio values obtained as a result is calculated as the final Rate value.
Further, the rate calculation unit 24 is configured to execute / do not execute the rate calculation operation in accordance with the control signal from the interpolation determination unit 26. That is, the rate calculation operation is performed in response to the execution instruction signal, and the rate calculation operation is not executed in response to the non-execution instruction signal.
The value of Rate calculated by the rate calculating unit 24 is supplied to the high frequency signal adding unit 25.

The high frequency signal adding unit 25 inputs the MDCT coefficient from the stereo processing unit 12, the correlation maximum shift amount “k” from the correlation maximum shift amount detection unit 23, and the rate value from the rate calculation unit 24. The value of the MDCT coefficient at each MDCT coefficient number position in the band where the signal disappears is added by the method described with reference to FIG.
That is, for each MDCT coefficient number position in the band where the signal disappears, a value obtained by dividing the MDCT coefficient value at the position returned to the low frequency side by k × 8 from each MDCT coefficient number position by the value of Rate is added. The process which performs is performed.

  Although illustration is omitted here, in practice, a delay circuit is provided in a necessary part so that each part in the high-frequency interpolating unit 4a executes processing for the MDCT coefficient of the common n-th frame. Measures such as adjusting the supply timing of each signal should be taken.

As described above, in the first embodiment, the autocorrelation calculation in the high frequency subband is performed to obtain the maximum correlation shift amount, and then the maximum correlation shift amount is separated in the high frequency subband. The value of Rate is calculated based on the result of calculating the amplitude ratio between each MDCT coefficient number position.
Then, using the correlation maximum shift amount and Rate, the value of the MDCT coefficient at each MDCT coefficient number position in the band where the signal is lost is returned from the MDCT coefficient number position by the correlation maximum shift amount. The position MDCT coefficient is interpolated by the value divided by the value of Rate.
Based on the correlation maximum shift amount obtained from the autocorrelation calculation result and the value of Rate in this way, by interpolating the band in which the signal is lost by encoding, the correlation in the high frequency subband and the high The signal can be added in the form of following the degree of amplitude attenuation toward the band side. As a result, a band in which the signal is lost with a more natural signal can be interpolated. That is, this can improve the sound quality.

In the first embodiment, when obtaining the maximum correlation shift amount, the high frequency sub-band is divided into predetermined units of divided bands, the peak value Peak in each divided band is detected, and this peak is detected. Although the autocorrelation calculation is performed using the value Peak, the following effects can be obtained.
That is, such a peak value Peak is a good representation of the characteristics in the divided bands. Thus, by performing autocorrelation calculation using the peak value Peak, a more reliable autocorrelation is obtained. Therefore, a more reliable value can be obtained as the value of the maximum correlation shift amount “k” obtained based on the autocorrelation calculation result.

In the first embodiment, j = 20 is set as the value of the shift amount j, and the autocorrelation calculation is started from a position shifted by 20 divided bands from the start point of the high frequency subband. As the value of the shift amount j, other values can be set as long as j = 1 is not set. That is, when j = 1, since the autocorrelation calculation value becomes the maximum at that time, it is essential that j ≧ 2 as a main premise.
At this time, if the value of j is small, the amount of shift up to the final shift position increases accordingly, and the amount of autocorrelation calculation increases. Conversely, if it is large, the autocorrelation calculation amount can be reduced.
Considering this point, it is conceivable that the shift amount j is made as large as possible to reduce the processing load of autocorrelation calculation. However, in reality, high-frequency signals tend to generate harmonics. Therefore, when an excessively large value is set as the shift amount j, the reliability of the autocorrelation calculation result may be significantly impaired.
Here, in general terms, the peak value Peak near the MDCT coefficient number = 320, which is the starting point of the high frequency sub-band, is the peak value near the MDCT coefficient number = 640, which is the first integer multiple (twice) of 320. It can be expected that the correlation with Peak will be high. In this embodiment, in view of this point, j = 20 (20 × 8 = 160 shift amounts when converted by the MDCT coefficient number, and the correlation calculation start position is around 320 + 160 = 480) is set. Yes. That is, the correlation calculation is started from a position where a margin of about 160 is obtained from the vicinity of the MDCT coefficient number = 640 predicted to have the highest correlation.
As the margin from the predicted position to the actual calculation start position increases, the detection accuracy of the autocorrelation calculation result, that is, the correlation maximum shift amount can be improved. However, if this margin is taken too much, the value of j will decrease as a result, and the amount of shift up to the final shift position will increase accordingly, and the amount of autocorrelation calculation will increase.
From the above, it can be seen that there is a trade-off relationship between reducing the processing load of autocorrelation calculation by setting the value of j and increasing the detection accuracy of the maximum correlation shift amount. In setting the actual shift amount j, an optimum value may be set in consideration of both the reduction of the processing load of the autocorrelation calculation and the securing of the reliability of the maximum correlation shift amount. As an example of the optimum point, j = 20 is illustrated in this example.

Also, according to the high-frequency interpolation operation of the first embodiment described above, the autocorrelation calculation needs to be performed only once per frame, and the ratio of the existing signal is calculated after calculating the ratio. High-frequency signals can be interpolated only by the process of dividing. In this regard, as in the conventional method, when the reference band is specified, the band in which the signal exists is divided, and in comparison with the case where the processing is performed in which the number of the divided numbers is combined to obtain the correlation, the processing is performed. The burden can be greatly reduced.
Further, as the high-frequency interpolation of the first embodiment, the processing content is the same for each frame (autocorrelation calculation / ratio calculation / addition of high-frequency signal based on correlation maximum shift amount and ratio). As described above, there is no inconvenience that the processing contents change for each frame and the processing amount and processing time change depending on the input signal.
Further, since the processing can be independent from the decoding processing, it is not necessary to use a decoding algorithm that is particularly common to the encoding side, and it is possible to prevent the problem of loss of versatility.

FIG. 9 shows experimental results for verifying the effectiveness of the first embodiment. In FIG. 9, for the audio signal for one AAC frame, FIG. 9 (a) shows the spectrogram of the original signal, and FIG. 9 (b) shows the encoded signal obtained by encoding the original signal by the conventional decoding process. FIG. 9C shows the spectrogram of the decoded signal, FIG. 9C shows the spectrogram of the high-frequency signal generated by the interpolation processing of this example, and FIG. 9D shows the signal after the decoding processing generated by the interpolation processing. The spectrogram of the signal which added the added high frequency signal is each shown.
In each drawing of FIG. 9, the vertical axis represents frequency, the horizontal axis represents time, and the intensity of the amplitude is represented by color intensity. FIG. 9 shows, as an example, a result when decoding / high-frequency interpolation is performed on an audio signal encoded by the AAC method / bit rate = 128 kbps.

First, as can be seen by comparing FIG. 9 (a) and FIG. 9 (b), a high-frequency signal is lost along with encoding. Depending on the high-frequency interpolation of the present example described above, as shown in FIG. 9C, a missing portion signal is generated by encoding, and this is decoded as shown in FIG. 9D. It will be added to the part where the missing part in the later signal has occurred.
From FIG. 9 (d), by performing the high-frequency interpolation of this example, the signal of the part in which the loss has occurred is added by taking advantage of the correlation in the high-frequency part that was not lost by encoding. I understand that.

  In the description so far, the case where the high-frequency interpolation operation of the first embodiment is realized by the hardware configuration as shown in FIG. 8 is exemplified. However, in the first embodiment, The high-frequency interpolation operation can also be realized by software processing. That is, the high-frequency interpolation unit 4a can be configured by, for example, an arithmetic processing unit including a CPU and a memory, and the high-frequency interpolation operation as the above-described embodiment is realized by software processing of the arithmetic processing unit. It is.

The processing operation to be executed in this case is shown in the flowchart of FIG.
The processing operation shown in this figure is executed by the arithmetic processing unit as the high-frequency interpolation unit 4a as described above, for example, according to a program stored in the memory.
First, in step S101, Peak detection processing is executed. That is, based on the MDCT coefficient for one frame supplied from the stereo processing unit 12, a band from a predetermined sfb (in this case, sfb [29]) to the last sfb having an amplitude is a high frequency subband. As described above, each sfb in this high frequency sub-band is divided into four and divided into respective divided bands, and then the peak value Peak of the MDCT coefficient in each divided band is detected. As a result, the peak value Peak [0] to peak value Peak [n] as described above with reference to FIG. 4 are detected.

  In subsequent step S102, an autocorrelation calculation process is executed. That is, using the information of the detected peak value Peak [0] to peak value Peak [n], the divided band number where the peak value Peak is detected is “i”, and the last divided band number where the amplitude exists is “N”. In this case, the correlation value at each shift position is calculated by performing the calculation according to Equation 1 shown above.

In the next step S103, first, for the calculated correlation values, the top five correlation values having the largest values are selected.
Then, in the subsequent step S104, a process for determining whether or not to interpolate is executed. That is, the sum of the selected top five correlation values is compared with a predetermined threshold value (1.0), and whether the sum of correlation values is equal to or greater than the threshold value (whether to interpolate). Is determined.

In step S104, if the sum of the correlation values is not equal to or greater than the threshold value and a negative result indicating that no interpolation is performed is obtained, “RETURN” is used as it is as illustrated.
On the other hand, if the sum of the correlation values is equal to or greater than the threshold value and an affirmative result is obtained, the process proceeds to step S105.

In step S105, as the correlation maximum shift amount “k” detection process, the shift amount when the correlation value calculated in the previous step S102 is maximized is detected as the maximum correlation shift amount “k”.
In step S106, the rate calculation process is executed. That is, the value of Rate is calculated by the method described in FIG. 6 using the peak values Peak [0] to Peak [n] detected in step S101 and the maximum correlation shift amount “k”. . Specifically, with respect to the peak values Peak of the divided bands [n] to [n-5], the amplitude ratio with the peak value Peak in the divided band returned to the low frequency side by the maximum correlation shift amount “k” is calculated. Then, the average of the five amplitude ratio values obtained as a result is calculated as the final Rate value.

Further, in the subsequent step S107, a high frequency signal addition process based on the correlation maximum shift amount “k” and the value of Rate is executed. That is, with respect to each supplied MDCT coefficient for one frame, each MDCT coefficient number position in the band where the signal (amplitude) disappears is a position returned from the MDCT coefficient number position to the low frequency side by k × 8. Processing for adding a value obtained by dividing the value of the MDCT coefficient by the value of Rate is performed.
When the process of step S107 is executed, “RETURN” is obtained as shown in the figure, whereby the processes described above (S101 to S107) are repeatedly executed for each frame.

In FIG. 10, processing for executing / not executing the rate calculation processing / high-frequency signal addition processing depending on whether or not the original signal originally has no high-frequency signal (selection processing in step S103 / step S104 in FIG. 10). Although the case where the determination process) is executed as a continuous set of processes has been illustrated, such processes in steps S103 and S104 are not necessarily executed as a continuous process. For example, the processing can be executed in the order of step S103 → step S105 → step S104.
Further, although the case where the determination process in step S104 is executed before the detection process of the correlation maximum shift amount “k” is illustrated, the determination process in step S104 is more than the rate calculation process and the high frequency signal addition process. Any timing can be used as long as it is before and after the selection processing of the top five correlation values. The selection process in step S103 can be performed at any timing as long as it is after the autocorrelation calculation process and before the determination process for determining whether to interpolate.

  In the description so far, the case where the high-frequency interpolation unit 4a is provided in the compression code decoding unit 4 is exemplified. However, for example, as shown in FIG. On the other hand, a high-frequency interpolation unit 4a may be provided.

In FIG. 11, the high-frequency interpolation unit 4 a is omitted in the compression code decoding unit 4 in this case, and the MDCT coefficients from the stereo processing unit 12 are directly supplied to the adaptive block length switching inverse MDCT unit 13. It is like that.
Then, an MDCT conversion unit 30 that inputs the audio signal (temporal audio signal) output from the compression coding / decoding unit 4 and converts the same into an MDCT by performing MDCT conversion again is added. .
In addition, the audio signal (MDCT coefficient) subjected to time-frequency conversion by the MDCT conversion unit 30 is input to the high-frequency interpolation unit 4a in this case for each AAC1 frame unit.
In this case as well, the operation performed by the high-frequency interpolation unit 4a is the same as that described above, and a description thereof will be omitted.

  Then, the MDCT coefficient to which the high-frequency signal is added by the high-frequency interpolation unit 4a is subjected to inverse MDCT conversion again by the inverse MDCT conversion unit 31 in the drawing so as to return to the time audio signal that can be output. The time audio signal thus obtained is supplied to the DSP 5 shown in FIG.

As described above, the high frequency interpolating unit 4a that performs the high frequency interpolating operation of the first embodiment is provided outside the compression code decoding unit 4 to improve the sound quality of the audio signal that has been decoded. It can also be designed.
However, as can be understood from the above description, in the case where the compression code decoding unit 4 is provided outside in this way, the time audio signal after decoding is converted back to the audio signal in the frequency-converted state again. A configuration (MDCT conversion unit 30) and a configuration (inverse MDCT conversion unit 31) for returning the audio signal in the frequency axis region after high-frequency interpolation back to the audio signal in the time axis region are separately required.
In view of this point, in the reproduction apparatus 1 shown in FIG. 1 above, the high-frequency interpolation unit 4a is provided in the compression code decoding unit 4, and has already been converted into the frequency axis region in the course of the decoding process. By applying high-frequency interpolation to a certain audio signal, the above-described separate configuration is not necessary.

Although the first embodiment has been described above, the high-frequency interpolation according to the present invention should not be limited to the specific examples described so far.
For example, in the description so far, the value of Rate is determined based on the end point side (the highest frequency side) of the high frequency sub-band, but can be determined based on the start point side (the lowest frequency side). That is, peak value Peak [k] / peak value Peak [0], peak value Peak [k + 1] / peak value Peak [1],... Peak value Peak [k + x] / peak value Peak [x] For the peak value Peak of each divided band from the start point of the high frequency sub-band to the predetermined value x, the amplitude ratio with the peak value Peak of the divided band set to + k is calculated, and the average value thereof is calculated as the value of Rate. It is something like that.
However, if the rate calculation is based on the end point as in the previous example, the rate value can be calculated more reliably than the calculation on the start point because the rate can be calculated closer to the part to be interpolated. be able to. That is, by adding a high frequency signal using such a value of Rate, a high frequency signal can be interpolated with a more natural signal.

Further, the value of Rate is a value obtained by averaging the amplitude ratios of a plurality of sets of peak values Peak separated by the maximum correlation shift amount “k”, but each set of peak values separated by k The amplitude ratio of Peak can be used as the value of Rate as it is.
However, if a value obtained by averaging a plurality of sets of amplitude ratios is used, the reliability of the value of the Rate can be improved as compared with the case where the rate is calculated from only one set of amplitude ratios.

  The peak value Peak is a peak value when divided in units of divided bands corresponding to eight MDCT coefficients, but the divided band width for obtaining the peak value Peak is set to a value other than “8”. You can also.

In the above description, the autocorrelation calculation is performed using the peak value Peak as the self signal. However, the autocorrelation calculation is performed using the MDCT coefficient at each MDCT coefficient number position without detecting the peak value Peak. You can also. The calculation of Rate can also be performed using the MDCT coefficient without detecting the peak value Peak.
When autocorrelation calculation is performed using the MDCT coefficient at each MDCT coefficient number position, the maximum correlation shift amount is a value in units of MDCT coefficient numbers. In this case, if the rate calculation is performed using the peak value Peak as in the previous specific example, the correlation maximum shift amount in units of MDCT coefficient numbers is divided by 1/8 (that is, divided). The value of Rate may be calculated based on the result of calculating the amplitude ratio of each peak value Peak separated by that value.
In addition, when autocorrelation calculation is performed using the MDCT coefficient at each MDCT coefficient number position in this way, the value of the maximum correlation shift amount is intentionally multiplied by the value of the divided bandwidth during the high-frequency signal addition operation. Need not be used.
However, if the autocorrelation calculation using the peak value Peak is used, the calculation amount of the autocorrelation calculation can be reduced by that amount, and a highly reliable correlation value can be calculated as described above. The maximum shift amount can be detected.

  Further, in the above description, for autocorrelation calculation, “N” in the previous equation 1 is set to the value of the last divided band number with amplitude, and autocorrelation calculation is performed using the entire range of the high frequency subband as the self signal. Although the case has been illustrated, for example, by setting the value of “N” as the value of the divided band number on the lower frequency side, the autocorrelation calculation can be performed using a part of the high frequency subband as a self signal. By doing so, the amount of autocorrelation calculation can be reduced.

In the description so far, only the case where high-frequency interpolation is performed on an audio signal (audio signal) that has been compression-encoded by the AAC method has been described. However, an audio signal that has been encoded by another audio compression-encoding method. The present invention can also be suitably applied when performing high-frequency interpolation for.
Here, except for the AAC method, only the region of a predetermined frequency or higher is not constant in bandwidth, and the bandwidth may be constant over the entire region. Therefore, in such a case, a predetermined frequency (MDCT coefficient number) is set in advance as the starting point of the high frequency sub-band, and similarly, from the starting point to the last band with amplitude is set as the high frequency sub-band. What should I do? Thereafter, the same effect as in the embodiment can be obtained by performing the same operation.

<Second Embodiment>

Next, a second embodiment will be described.
The second embodiment is for correcting a quantization error.
FIG. 12 is a block diagram showing an internal configuration of the playback apparatus 40 according to the second embodiment.
As shown in the figure, in the reproduction apparatus 40 of the second embodiment, a quantization error correction unit 4b is provided in the compression code decoding unit 4 in place of the previous high frequency interpolation unit 4a.
Note that the playback device 40 of the second embodiment has the same configuration as the playback device 1 of the previous first embodiment, except that this quantization error correction unit 4b is provided. The description is omitted.

FIG. 13 shows the internal configuration of the compression code decoding unit 4 shown in FIG.
As the quantization error correction unit 4b in the case of the second embodiment, the stereo processing unit 12 and the adaptive block length switching inverse MDCT unit are included in the compression code decoding unit 4 as in the high frequency interpolation unit 4a. 13 is provided. That is, as the quantization error correction unit 4b, similarly to the high frequency interpolation unit 4a, the MDCT coefficient obtained by the stereo processing unit 12 and restored to the state immediately after the MDCT processing at the time of encoding is obtained. Enter and process to process it.
Since the other components in the compression code decoding unit 4 are the same as those in the first embodiment, the same reference numerals are given here and the description thereof is omitted.

[Quantization error]
Here, in general, the audio compression coding employs a method using psychoacoustic analysis as the frequency correlation coding as described above. In this encoding using psychoacoustic characteristics, efficient information compression processing is performed while suppressing deterioration in sound quality by reducing the number of bits allocated for frequency bands that cannot be heard due to auditory perception. It is illustrated.
However, in the frequency band to which the low bit number is assigned in this way, the decoding accuracy at the time of decoding is lowered due to the low bit number, and the phenomenon that the difference from the original audio signal becomes large occurs. Such a phenomenon is called a quantization error.

FIG. 14 is a diagram for explaining a quantization error caused by such an allocated bit number.
First, as a premise, in compression coding, the number of bits to be allocated for a required frequency band is reduced as described above. The number of bits allocated for each band is the resolution set for that band. Determined by the value of.
FIG. 14 shows the relationship between original data (original signal data), data before / after quantization by encoding, and data as a decoding result when different resolutions are set. Specifically, FIG. 14A shows a case where resolution = 2 (high resolution) is set, and FIG. 14B shows a case where resolution = 5 (low resolution) is set.

As can be seen with reference to these figures, when the resolution is high, the error between the original data and the decoding result is smaller than when the resolution is low.
Specifically, for example, focusing on the case of the original data = 8 in the leftmost column in the figure, when resolution = 2, the value before quantization is “4” by 8/2, and the quantization result is This can be regarded as an integer, and is the same “4” as shown after quantization in the figure. The decoding result corresponds to a value obtained by returning the quantized value as a resolution value, and becomes “8” by 4 × 2.
On the other hand, when resolution = 5, the value before quantization for original data = 8 is “1.6”, and the value after quantization is converted to an integer and becomes “1”. Then, the decoding result is “5” obtained by returning “1” with the resolution “5”, and the error is larger than the decoding result “8” when the resolution = 2.

  When the resolution is low in this way, the error between the value before quantization and the value after quantization corresponding to the integer value tends to increase, and the decoding result is also based on this. This will cause a large error from the data value. This is why a quantization error is likely to occur in a portion where the number of allocated bits is small.

  Here, for example, if the value that the original data can take is in the range of 0 to 100, the value of the original data is expressed in fine increments by increasing the resolution (decreasing the value). However, it can be seen that more bits need to be allocated accordingly. On the other hand, by reducing the resolution (increasing the value), the step size for expressing the value of the original data can be made coarse, so that the number of bits to be allocated can be reduced. From this, it can be understood that the number of allocated bits for each band is determined by the resolution value set for each band during encoding.

  FIG. 15 is a diagram for explaining an example of an actual form of quantization error. FIG. 15A shows a spectrum distribution when the bit rate is 512 kbps, and FIG. 15B shows a spectrum distribution when the bit rate is 128 kbps. Is shown. In this figure, the distribution of MDCT coefficients for one AAC frame output from the stereo processing unit 12 shown in FIG. 13 (FIG. 2) (black circle in the figure) is shown as the spectrum distribution. Further, in this figure, the distribution of MDCT coefficients after 7 kHz (˜22 kHz), for example, is shown as the middle / high range where the reduction rate of the number of allocated bits is relatively high due to frequency correlation coding.

First, in the case of the high bit rate shown in FIG. 15A, the number of bits allocated to each band can be made relatively large even in the middle and high bands, so that it can be confirmed that a large quantization error hardly occurs. .
On the other hand, in the case of the low bit rate in FIG. 15B, a relatively large quantization error is likely to occur as the reduction rate of the number of allocated bits in the middle / high range increases. This large quantization error portion can be confirmed as a portion where the same amplitude value is continuous, as surrounded by a circle in the figure. This is a phenomenon called flattening due to quantization error.

  Of course, such a flattened portion does not faithfully reproduce the original waveform of the original signal. That is, the sound quality is particularly deteriorated in the portion where the flattening occurs.

[Correction of quantization error]
Therefore, in the second embodiment, it is particularly assumed that sound quality deterioration due to such a flattened portion is prevented and the sound quality is improved (that is, the sound quality is improved). As a configuration for that purpose, the quantization error correction unit 4b shown in FIG. 13 (FIG. 12) is provided.

FIG. 16 is a block diagram showing an internal configuration of the quantization error correction unit 4b.
As shown in the figure, the quantization error correction unit 4b includes a flattened part extraction unit 41, a replacement determination unit 42, a prediction processing unit 43, and a replacement unit 44.
MDCT coefficients for each AAC 1 frame unit output from the stereo processing unit 12 shown in FIG. 13 are input to the respective units in the quantization error correction unit 4b.

First, the prediction processing unit 43 generates a prediction signal obtained by predicting the original signal before compression coding based on the MDCT coefficient for each frame supplied from the stereo processing unit 12.
In this embodiment, a predictor standardized by the AAC Main profile (ISO / IEC13818-7) is used as the predictor included in the prediction processing unit 43. This predictor is the same as that provided in the prediction processing unit 12b shown in FIGS.

Here, FIG. 17 shows a flow of prediction processing by the predictor.
This predictor is a second-order backward adaptive lattice predictor. The predicted value x est (n) is obtained as follows.
x est (n) = x est, 1 (n) + x est, 2 (n)
here,
x est, 1 (n) = bk1 (n) rq, 0 (n-1)
x est, 2 (n) = bk2 (n) rq, 1 (n-1)
And a = b = 0.953125.
rq, 0 (n) = axrec (n)
rq, 1 (n) = a (rq, 0 (n-1) − bk1 (n) eq, 0 (n))
eq, 0 (n) = xrec (n)
eq, 1 (n) = eq, 0 (n) − xest, 1 (n)
km (n + 1), m = 1,2 is obtained by the following equation.
km (n + 1) = CORm (n) / VARm (n)
here,
CORm (n) = αCOR m (n-1) + r q, m-1 (n-1) eq, m-1 (n)
VARm (n) = αVAR m (n-1) +0.5 (r ² q, m-1 (n-1) + e ² q, m-1 (n))
And α = 0.90625.

  In this example, as the prediction processing unit 43, a part for performing the prediction processing by the same method as that of the previous prediction processing unit 12b is separately provided. This is an optional processing of AAC as shown in FIGS. This is because the intensity stereo processing unit 12c and the TNS processing unit 12d shown in FIG. That is, when these optional processes are performed, the processing result in the prediction processing unit 12b cannot be obtained as a prediction signal that can be used in the quantization error correction unit 4b of this example. 43 is provided.

Returning to FIG.
The flattened portion extraction unit 41 is supplied with the MDCT coefficients for one frame from the stereo processing unit 12 as described above, and the scale factor band (sfb) supplied from the format analysis unit 10 shown in FIG. ) For each resolution is input.
Based on the MDCT coefficient for one frame and the resolution information for each sfb, the flattened part extraction unit 41 detects (extracts) a part where the value of the MDCT coefficient is the same as the resolution of the sfb for each sfb. )
In many cases, the portion detected as having the same value as the resolution value in this way is a portion that forms a flattened portion as shown in FIG.

Here, in the second embodiment, the band for performing the correction of the quantization error is limited to the middle / high range where the number of bit allocation is insufficient. For example, in this case, the correction operation is performed only for the band from 7 kHz to 22 kHz as shown in FIG.
In response to this, in the flattened portion extraction unit 41, the MDCT coefficient position where the value of the MDCT coefficient is the same value as the resolution of the sfb only for such sfb of 7 kHz or later among the sfb forming one frame. Is detected.
The flattening band extraction unit 41 supplies the detected MDCT coefficient position information (MDCT coefficient number information) and the resolution information of the sfb to which the MDCT coefficient position belongs to the replacement determination unit 42 together.

The replacement determination unit 42 includes information on the MDCT coefficient number detected by the flattened portion extraction unit 41 and resolution information in the band, the MDCT coefficient from the stereo processing unit 12, and the prediction generated by the prediction processing unit 43. A signal is input and a replacement determination is performed based on the information. That is, based on the value of the prediction signal at the MDCT coefficient position specified by the MDCT coefficient number from the flattened portion extraction unit 41, the value of the MDCT coefficient, and the resolution information of the sfb to which the MDCT coefficient position belongs. , Perform replacement judgment.
Specifically, the replacement determination is performed based on the result of determining whether or not the value of the prediction signal is equal to or less than [MDCT coefficient value (absolute value) + resolution / n]. For example, in this case, n = 2 and it is determined whether or not the value of the prediction signal is equal to or less than the value of the MDCT coefficient + the value of resolution / 2.
When it is assumed that the value of the prediction signal is not equal to or less than the value of the MDCT coefficient + the value of resolution / 2, a determination signal indicating a determination result indicating that the MDCT coefficient position is not replaced is a replacement unit 44 described below. Supply against. On the other hand, when the value of the prediction signal is equal to or less than the value of the MDCT coefficient + the value of resolution / 2, a determination signal indicating a determination result that the MDCT coefficient position is to be replaced is sent to the replacement unit 44. Supply.

The replacement unit 44 performs replacement processing based on the MDCT coefficients supplied from the stereo processing unit 12, the determination signal from the replacement determination unit 42, and the prediction signal from the prediction processing unit 23.
Specifically, the value of the MDCT coefficient is replaced with the value of the prediction signal for the MDCT coefficient position where the determination result indicating that the replacement is performed by the determination signal from the replacement determination unit 42 is indicated.
The MDCT coefficients for one frame subjected to the replacement processing by the replacement unit 44 are sequentially supplied to the adaptive block length switching inverse MDCT unit 13 shown in FIG.

  In the second embodiment as described above, the MDCT coefficient position that is the resolution value of the band (sfb) on the frequency axis is detected for the audio signal (audio signal) that has been compression-encoded. The amplitude value at that position is replaced with the value of the prediction signal based on the comparison result with the value of the prediction signal.

Here, as described above, in this example, the portion having the same value as the resolution on the frequency axis is detected, but as can be understood from the description of FIGS. 14 and 15 above, A flattened portion that causes deterioration in sound quality is likely to occur where the resolution is the same. In other words, when the wave number is rounded down by integerization as the quantization, the portion with the same value as the resolution (resolution value × 1) is likely to have a relatively high rounding rate, and the resulting quantization The error rate tends to increase, and a large quantization error tends to occur.
Therefore, in this example, as described above, first, a portion having the same value as the resolution value is detected as a replacement candidate portion, and whether a large quantization error has occurred in the candidate portion from the predicted signal value, that is, replacement is performed. In the case where it is determined to be valid after determining whether it is appropriate, the MDCT coefficient of that portion is replaced with the value of the prediction signal.
As a result, it is possible to appropriately detect a portion where a large quantization error is caused due to the small number of assigned bits and correct the amplitude value of the error portion with a more probable value according to the prediction signal. As a result, it is possible to effectively suppress deterioration in sound quality due to compression coding and to improve sound quality.

According to the correction operation as the second embodiment, when the sound quality is improved, the prediction signal is generated based on the result of generating the prediction signal and comparing the value with the amplitude value of the audio signal. It is only necessary to perform substitution with the value of.
According to this, for example, a range of values that can be originally taken is calculated as in the prior art, a correction value is calculated from encoded signals in adjacent frequency bands using the least square method, and the correction value is a value within the range. Compared with the case where high sound quality is achieved by the method of replacing the existing signal if there is, if it is outside the range, replacing with the existing signal using the minimum and maximum values of the range, The processing burden can be significantly reduced.

  In particular, in the case of this example that employs a second-order backward adaptive grid type predictor as described above in generating a prediction signal, the prediction algorithm may be called once per frame as the prediction algorithm. As in the case where a quadratic curve or the like appropriate for each band is obtained as described above, it is not necessary to perform processing over a plurality of frames, and the processing load can be remarkably reduced. Furthermore, the processing load can be made constant regardless of the sound source, and the processing can be stabilized.

In the second embodiment, for example, the correction operation is performed only for the middle and high frequency bands after 7 kHz. However, this may replace the unnecessary portion. It can be effectively prevented.
For example, the invention described in Patent Document 3 described above is based on the premise that correction is performed for the entire frequency band. Where the bit allocation is sufficient, the obtained correction value may be far from the original amplitude value, and as a result, the sound quality may not be improved.
On the other hand, if the band to be corrected is limited to the middle / high band where bit allocation is insufficient, as in this example, correction can be performed only for the necessary band, effectively preventing the occurrence of such problems. Can be prevented.

  In the description so far, the case where the quantization error correction operation as the second embodiment is realized by the hardware configuration as shown in FIG. 16 is exemplified. The correction operation of this form can also be realized by software processing as in the case of the first embodiment. That is, also in this case, the quantization error correction unit 4b is constituted by an arithmetic processing unit including, for example, a CPU and a memory, and the second embodiment described above is performed by software processing of the quantization error correction unit 4b as the arithmetic processing unit. The correction operation is realized.

The processing operation to be executed in this case is shown in the flowchart of FIG.
Note that also in the processing operation shown in this figure, the quantization error correction unit 4b as the arithmetic processing unit as described above is executed according to a program stored in the memory, for example.
Further, although not illustrated, the quantization error correction unit 4b serving as the arithmetic processing unit is based on the MDCT coefficients from the stereo processing unit 12 in parallel with the processing operation shown in FIG. It is assumed that a process of generating a prediction signal by performing a prediction process based on the description is performed.

  First, in step S201, resolution information is acquired. That is, the resolution information for each sfb from the format analysis unit 10 shown in FIG. 13 is acquired.

  In a succeeding step S202, a process for extracting a flattened portion is performed. That is, the MDCT coefficient for one frame supplied from the stereo processing unit 12 is input, and among the sfb forming this one frame, for example, only the sfb of 7 kHz or higher is set as the MDCT coefficient value and the resolution of the sfb. The MDCT coefficient position having the same value is detected.

Further, in the next step S203, replacement determination is performed for each extracted portion based on the prediction signal.
That is, for each MDCT coefficient position detected in step S202, replacement determination is performed based on the value of the MDCT coefficient, the resolution value of the sfb to which the MDCT coefficient position belongs, and the value of the prediction signal. Specifically, as described above, regarding the MDCT coefficient value, the predicted signal value, and the resolution value, whether the predicted signal value is equal to or less than the MDCT coefficient value + resolution value / 2. Based on the determination result, replacement determination is performed.

Then, in the subsequent step S204, a process for replacing the MDCT coefficient of the replacement required part based on the prediction signal is performed. That is, only the MDCT coefficient position determined from the replacement determination in step S203 that the predicted signal value is equal to or less than the MDCT coefficient value + resolution value / 2 is replaced with the predicted signal value.
When this step S204 is executed, “RETURN” is obtained as shown in the figure. For confirmation, the processing operation shown in this figure should be executed for each AAC1 frame.

Here, in the description so far, the case where the quantization error correction unit 4b is provided in the compression code decoding unit 4 is exemplified. However, even in this case, for example, as shown in FIG. A quantization error correction unit 4b may be provided outside the decoding unit 4.
As shown in the figure, the configuration in this case corresponds to a configuration in which the quantization error correction unit 4b is provided instead of the high-frequency interpolation unit 4a in the configuration shown in FIG. However, in the case of the quantization error correction unit 4b, the resolution information for each sfb is used, so the resolution information for each sfb from the format analysis unit 10 is also input as indicated by the broken line arrow in the figure. To do.
In this case as well, the operation performed by the quantization error correction unit 4b is the same as that described above, and a description thereof will be omitted.

In the modification shown in FIG. 19, the resolution information from the format analysis unit 10 is input to the quantization error correction unit 4b (flattened portion extraction unit 41) thus provided externally. Although configured, input of resolution information from the format analysis unit 10 is not essential.
When the resolution information from the format analysis unit 10 is not input, the MDCT coefficient obtained through the compression code decoding unit 4 → MDCT conversion unit 30 is supplied to the flattened partial extraction unit 41. In addition, in this case, the flattened portion extraction unit 41 detects the minimum amplitude value in each sfb as the resolution value of the sfb, and in the detection of the flattened portion and the subsequent replacement determination, The detected value is used as the resolution value. Alternatively, an amplitude value that frequently occurs for each sfb can be regarded as a resolution value of the sfb and used.

Although the second embodiment has been described above, the quantization error correction of the present invention should not be limited to the specific examples described so far.
For example, in the description so far, the case where the correction process is performed only for a predetermined band has been illustrated, but the correction process may be performed for the entire band.

In the above description, an example of using a second-order backward adaptive grid type predictor is used for generating a prediction signal. However, a predictor of another method may be used. Alternatively, the prediction signal can be generated by a prediction process using a multidimensional function such as an interpolation polynomial or a multidimensional approximate expression.
However, as described above, in terms of reducing the processing load, it is preferable to use a second-order backward adaptive grid type predictor as employed in the embodiment.

Further, in the description so far, the case where it is determined whether or not the value of the prediction signal of the error candidate portion is equal to or less than the value of the MDCT coefficient + the value of resolution / 2 is used as a criterion for determining the appropriateness of the replacement. However, such a determination criterion may be at least resolution / n. However, because the quantization error is originally less than the resolution, a value that makes the resolution / n value less than the resolution value should be selected as the value of n.
Alternatively, the validity of the replacement may be determined by determining whether the value of the prediction signal is equal to or less than the value of the MDCT coefficient ± the resolution value / n. Alternatively, it is possible to determine whether or not the difference between the amplitude value of the error candidate portion and the value of the predicted signal is resolution / n.

<Third Embodiment>

FIG. 20 is a block diagram illustrating an internal configuration of the playback device 50 according to the third embodiment.
The third embodiment is for correcting the quantization error between bands.
As shown in the figure, in the reproduction apparatus 50 according to the third embodiment, inter-band quantization is performed in the compression code decoding unit 4 in place of the previous high-frequency interpolation unit 4a (or quantization error correction unit 4b). An error correction unit 4c is provided.
Note that the playback device 50 of the third embodiment has the same configuration as the playback device 1 of the previous first embodiment, except that such an interband quantization error correction unit 4c is provided. Therefore, the same reference numerals are given here and the description thereof is omitted.

FIG. 21 shows an internal configuration of the compression code decoding unit 4 shown in FIG.
Similarly to the high-frequency interpolation unit 4a, the interband quantization error correction unit 4c in the case of the third embodiment also includes the stereo processing unit 12 and the adaptive block length switching inverse in the compression code decoding unit 4. Provided with the MDCT unit 13. That is, the inter-band quantization error correction unit 4c is also restored to the state corresponding to the state immediately after the MDCT processing obtained by the stereo processing unit 12 as in the previous high-frequency interpolation unit 4a and quantization error correction unit 4b. The processed MDCT coefficients are input and processed.
In this case as well, the other components in the compression code decoding unit 4 are the same as those in the first embodiment, so the same reference numerals are given and the description thereof is omitted.

[Quantization error between bands]
Here, the causes of sound quality degradation associated with compression coding include not only high-frequency signal loss and flattening due to quantization errors described above, but also inter-band quantization associated with a reduction in the number of allocated bits. An error can also be mentioned.

22 and 23 are diagrams for explaining such an inter-band quantization error.
First, in FIG. 22, only a part of MDCT coefficient positions of continuous scale factor bands (fsb [n], fsb [n + 1]) are extracted and shown on the frequency axis. In this figure, the horizontal axis indicates the MDCT coefficient number, and the left side in the figure is the low frequency side and the right side is the high frequency side.
In this figure, it is assumed that the amplitude value shown at each MDCT coefficient position indicates the value of the original data (original signal), not the value of the MDCT coefficient.

As shown in the drawing, in the sfb [n] on the lower side of the continuous sfb, the amplitude of the MDCT coefficient number [1] is “12”, the amplitude of the MDCT coefficient number [2] is “17”, and the MDCT coefficient number The amplitude of [3] is “11”.
Further, sfb [n + 1] adjacent to the high frequency side of sfb [n] starts from the MDCT coefficient number [4] as shown in the figure. Therefore, the MDCT coefficient number [4] is a boundary portion between these sfb [n] and sfb [n + 1]. In sfb [n + 1], the amplitude of the MDCT coefficient number [4] is “8”, the amplitude of the MDCT coefficient number [5] is “10”, the amplitude of the MDCT coefficient number [6] is “13”, and the MDCT coefficient number [7] ] Is shown to be “18”.

FIG. 23 shows the MDCT coefficient numbers [1] to [3] in sfb [n] and the MDCT coefficient numbers [4] to [7] in sfb [n + 1] shown in FIG. It is the figure which tabulated and showed a mode that the data before and after quantization with respect to the original data, and a decoding result change with the value of the resolution set by sfb.
Here, for example, suppose that the same resolution = 2 is set for sfb [n] and sfb [n + 1] as shown in the figure as case: A when compressed at a high bit rate (for example, 512 kbps). On the other hand, it is assumed that resolution: 2 is set for sfb [n] and resolution = 5 is set for sfb [n + 1] as case: B when compressed at a low bit rate (for example, 128 kbps).

Focusing on the portion between the bands of sfb [n] and sfb [n + 1], the original data is “11” at the position of the MDCT coefficient number [3] on the sfb [n] side, and the MDCT coefficient number on the sfb [n + 1] side. The position [4] is “8”.
In the case of high bit rate case A, the same resolution = 2, so the position of MDCT coefficient number [3] on the sfb [n] side and the position of MDCT coefficient number [4] on the sfb [n + 1] side Then, the original data “11” and “8” are respectively scaled to become numerical values “5.5” and “4”, and these are rounded to the numerical values “5” and “4” by rounding the decimal point by quantization. Become. The decoding result is a numerical value “10” or “8” in which these quantized values are returned based on the resolution values.
Thus, in case of high bit rate case A, the amplitude between the bands is the decoding results “10” and “8” with respect to the original data “11” and “8”, and the quantization error thereof is “1”.

  On the other hand, in case: B in the case of a low bit rate, the resolution changes from 2 to 5 between sfb. For this reason, at the position of the MDCT coefficient number [4] on the fsb [n + 1] side where the resolution is 5, the quantization error is larger than the case: A, and the decoding result is the same as in the previous case: A. The numerical value “8” changes to “5” as shown in the figure. That is, the quantization error between bands in case: A was “1”, whereas in case: B, this increased to “4”.

  The fact that a large quantization error is likely to occur due to low resolution has already been described in the second embodiment. However, when attention is paid between the bands as described above, the decoding result is caused by the difference in resolution between the bands. The possibility that the continuity of the waveform deteriorates increases. In this way, a large quantization error between bands causes discontinuity and promotes deterioration of sound quality.

[Correction of quantization error between bands]
Therefore, in the third embodiment, such a large quantization error between bands is corrected to achieve high sound quality. As a configuration for that purpose, the interband quantization error correction unit 4c shown in FIG. 21 (FIG. 20) is provided.

FIG. 24 is a block diagram showing an internal configuration of the interband quantization error correction unit 4c.
As shown in the figure, in the quantization error correction unit 4c, an error boundary extraction unit 51, a replacement determination unit 52, a prediction processing unit 53, and a replacement unit 54 are provided.
Of the units constituting the quantization error correction unit 4c, the replacement determination unit 52, the prediction processing unit 53, and the replacement unit 54 are each AAC 1 frame unit output from the stereo processing unit 12 shown in FIG. MDCT coefficients are input.
In this case as well, the prediction processing unit 53 generates a prediction signal that predicts the original signal before compression encoding based on the MDCT coefficient for each frame supplied from the stereo processing unit 12. As the prediction processing unit 53 in the third embodiment, a second-order backward adaptive lattice predictor that is standardized by the AAC Main profile as described above is used.

First, the error boundary extraction unit 51 receives resolution information for each sfb supplied from the format analysis unit 10 shown in FIG.
Based on the resolution information for each sfb, the error boundary extraction unit 51 detects (extracts) a boundary portion between consecutive bands having different resolution values.
Here, the correction of the interband quantization error in the case of the third embodiment is not limited to the band unlike the second embodiment, and the correction operation is performed for the entire band. For this reason, the error boundary extraction unit 51 detects a boundary portion in which the resolution value differs between consecutive sfb for the entire band sfb forming one frame.
The error boundary extraction unit 51 supplies information on the MDCT coefficient position (MDCT coefficient number information) as the detected boundary part and information on the resolution of the sfb to which the MDCT coefficient position belongs to the replacement determination unit 52.

The replacement determination unit 52 generates the MDCT coefficient number information and resolution information detected by the error boundary extraction unit 51, the MDCT coefficients supplied from the stereo processing unit 12 as described above, and the prediction processing unit 43. A replacement determination is performed based on the predicted signal. That is, replacement is performed based on the value of the prediction signal at the MDCT coefficient position specified by the MDCT coefficient number from the error boundary extraction unit 51, the value of the MDCT coefficient, and the resolution information of the sfb to which the MDCT coefficient position belongs. Make a decision.
Specifically, with respect to the MDCT coefficient value, the predicted signal value, and the resolution value, it is determined whether or not the predicted signal value is equal to or less than [MDCT coefficient value (absolute value) + resolution / n]. Based on the result, replacement determination is performed. For example, also in this case, n = 2, and it is determined whether or not the value of the prediction signal is [MDCT coefficient value + resolution / 2 or less.
When the value of the prediction signal is not equal to or less than the value of the MDCT coefficient + resolution / 2, a determination signal indicating a determination result indicating that the MDCT coefficient position is not replaced is supplied to the next replacement unit 54. To do. On the other hand, when the value of the prediction signal is equal to or less than the value of the MDCT coefficient + resolution / 2, a determination signal indicating the determination result that the MDCT coefficient position is to be replaced is supplied to the replacement unit 54. To do.

The replacement unit 54 performs replacement processing based on the MDCT coefficients for one frame supplied from the stereo processing unit 12, the determination signal from the replacement determination unit 52, and the prediction signal from the prediction processing unit 53.
Specifically, the value of the MDCT coefficient is replaced with the value of the prediction signal for the MDCT coefficient position where the determination result indicating that the replacement is performed by the determination signal from the replacement determination unit 52 is shown.
The MDCT coefficients for one frame subjected to the replacement processing by the replacement unit 54 are sequentially supplied to the adaptive block length switching inverse MDCT unit 13 shown in FIG.

  As described above, in the third embodiment, the frequency axis of the audio signal (audio signal) that has been compression-encoded and subjected to the information compression processing that determines the resolution for allocating the number of bits in a predetermined frequency band unit. The boundary portion between successive frequency bands having different resolution values is detected, and the amplitude value of the boundary portion is replaced with the predicted signal value based on the result of comparison with the predicted signal. According to this, a part in which a large quantization error occurs between bands due to the difference in the number of allocated bits and the continuity between the bands is impaired is properly detected, and the amplitude value of the part is more likely based on the prediction signal. The value can be corrected. As a result, it is possible to improve the sound quality deterioration due to the discontinuity between the bands caused by the difference in the number of assigned bits, and to improve the sound quality.

  Further, in the method according to the third embodiment, a portion where an error is likely to occur is specified and only the portion is corrected. As in the case where correction is performed, it is possible to avoid a situation in which the correction is performed up to a portion where correction is not necessary, and the sound quality is deteriorated. This also applies to the second embodiment described above.

  In the description so far, the case where the correction operation of the interband quantization error as the third embodiment is realized by the hardware configuration as shown in FIG. The correction operation of the third embodiment can also be realized by software processing as in the previous embodiments. That is, also in this case, the interband quantization error correction unit 4c is configured by an arithmetic processing unit including, for example, a CPU and a memory, and the third processing described above is performed by software processing of the interband quantization error correction unit 4c as the arithmetic processing unit. The correction | amendment operation | movement as embodiment of this is implement | achieved.

The processing operation to be executed in this case is shown in the flowchart of FIG.
The processing operation shown in this figure is also executed by the interband quantization error correction unit 4c as the arithmetic processing unit as described above, for example, according to a program stored in the memory.
Also in this case, the interband quantization error correction unit 4c as the arithmetic processing unit predicts based on the MDCT coefficient from the stereo processing unit 12 in parallel with the processing operation shown in FIG. Assume that processing for generating a prediction signal is performed.

  First, in step S301, resolution information is acquired for each band. That is, the resolution information for each sfb from the format analysis unit 10 shown in FIG. 21 is acquired.

  In subsequent step S302, processing for extracting a resolution difference band boundary portion is performed. That is, based on the resolution information for each band acquired in step S301, the boundary portion (MDCT coefficient position) between consecutive sfb having different resolution values for all sfb forming one frame. Is detected.

Furthermore, in the next step S303, replacement determination is performed for each extracted portion based on the prediction signal.
That is, for each MDCT coefficient position as the boundary portion detected in step S302, replacement determination is performed based on the value of the MDCT coefficient, the resolution value of the sfb to which the MDCT coefficient position belongs, and the value of the prediction signal. Specifically, as described above, regarding the MDCT coefficient value, the predicted signal value, and the resolution value, whether the predicted signal value is equal to or less than the MDCT coefficient value + resolution value / 2. Based on the determination result, replacement determination is performed.

In the subsequent step S304, a process for replacing the MDCT coefficient of the required replacement portion based on the prediction signal is performed. In other words, only the MDCT coefficient position determined from the replacement determination in step S303 that the predicted signal value is equal to or less than the MDCT coefficient value + resolution value / 2 is replaced with the predicted signal value.
When this step S304 is executed, “RETURN” is obtained as shown in the figure. The processing operation shown in this figure should also be executed for each AAC1 frame.

Here, also in the third embodiment, the case where the interband quantization error correction unit 4c is provided in the compression code decoding unit 4 is exemplified, but in this case as well, for example, as shown in FIG. It can also be provided outside the compression code decoder 4.
That is, as shown in the figure, the configuration in this case corresponds to the configuration shown in FIG. 11 in which an interband quantization error correction unit 4c is provided instead of the high-frequency interpolation unit 4a. However, since the interband quantization error correction unit 4c also uses resolution information for each sfb, it also inputs resolution information for each sfb from the format analysis unit 10 as indicated by the broken line arrows in the figure.
In this case as well, the operation performed by the interband quantization error correction unit 4c is the same as that described above, and a description thereof will be omitted.

  In the modified example shown in FIG. 26, as in the modified example of the second embodiment, the minimum amplitude value in each sfb is detected as the resolution value of the sfb, or each sfb If the amplitude value that occurs frequently every time is detected as the resolution value of the sfb, the boundary portion extraction operation and the replacement determination can be performed without inputting resolution information from the format analysis unit 10.

Here, the inter-band quantization error correction of the present invention should not be limited to the specific example as the third embodiment described so far.
For example, in the third embodiment, an example in which a second-order backward adaptive grid type predictor is used for generating a prediction signal has been described. However, in this case also, a predictor using another method or an interpolation polynomial is used. In addition, a prediction signal can be generated by a prediction process using a multidimensional function such as a multidimensional approximate expression.

Also in the third embodiment, whether or not the value of the prediction signal of the detected error candidate portion is equal to or less than the value of the MDCT coefficient + the value of resolution / 2 is used as a criterion for determining the validity of replacement. Although the case of determination is illustrated, the determination criterion may be at least resolution / n even in this case. However, in this case as well, due to the nature of the quantization error, a value that makes the resolution / n value less than the resolution value should be selected as the value of n.
In this case as well, the validity of the replacement may be determined by determining whether the value of the prediction signal is within the value of the MDCT coefficient ± the value of resolution / n. Alternatively, it is possible to determine whether or not the difference between the amplitude value of the error candidate portion and the value of the predicted signal is resolution / n.

As mentioned above, although each embodiment of this invention was described, as this invention, it should not be limited to the specific example demonstrated so far.
For example, in the description so far, the case where the present invention is applied to a playback apparatus that plays back an audio signal stored in, for example, an HDD or a flash memory as the storage unit 2 is exemplified. For example, the present invention can also be applied to a reproducing apparatus that uses a magneto-optical disk such as MD (Mini Disc (registered trademark)) or an optical disk such as CD (Compact Disc) or DVD (Digital Versatile Disc) as a recording medium.
In addition to the case where the audio signal stored in the recording medium is reproduced as described above, various electronic devices having a function of performing the decoding process on the compressed encoded audio signal included in the digital television broadcast signal, The present invention can also be suitably applied to various electronic devices having a function of performing a decoding process on a compression-encoded audio signal included in Web stream data.

  In the description so far, the present invention exemplifies the configuration corresponding to the Lch (channel) / Rch 2ch audio signal. For example, a multi-ch audio signal such as 5.1ch or a monaural audio signal is improved in sound quality. The present invention can also be suitably applied to the processing for this.

  In the above description, only the case where the present invention performs the correction process on the audio signal (audio signal) that has been compression-encoded by the AAC method has been described. However, the audio signal was encoded by another audio compression-encoding method. The present invention can also be suitably applied to processing for improving the sound quality of an audio signal.

It is the block diagram shown about the internal structure of the reproducing | regenerating apparatus (electronic device) as the 1st Embodiment of this invention. It is the block diagram shown about the internal structure of the compression encoding / decoding part with which the inside of the reproducing | regenerating apparatus of 1st Embodiment is provided. It is a figure for demonstrating the outline | summary of the MDCT coefficient for AAC1 frame. It is a figure for demonstrating peak value detection. It is a figure for demonstrating autocorrelation calculation. It is a figure for demonstrating calculation of Rate. It is a figure for demonstrating addition of a high region signal. It is the block diagram shown about the internal structure of the signal processing apparatus (high frequency interpolation part) of 1st Embodiment. It is the figure which showed the experimental result for demonstrating about the effectiveness of the high region interpolation operation | movement of 1st Embodiment. It is the flowchart shown about the processing operation which should be performed in order to implement | achieve the high region interpolation operation | movement as 1st Embodiment by software processing. It is the block diagram which showed the structure of the modification of the reproducing | regenerating apparatus (electronic device) of 1st Embodiment. It is the block diagram shown about the internal structure of the reproducing | regenerating apparatus (electronic device) as the 2nd Embodiment of this invention. It is the block diagram shown about the internal structure of the compression encoding / decoding part with which the inside of the reproducing | regenerating apparatus of 2nd Embodiment is provided. It is a figure for demonstrating a quantization error. It is a figure for demonstrating the planarization by a quantization error. It is the block diagram shown about the internal structure of the signal processing apparatus (quantization error correction | amendment part) of 2nd Embodiment. It is the figure which showed the processing flow of the predictor used by 2nd, 3rd Embodiment. It is the flowchart shown about the processing operation which should be performed in order to implement | achieve correction | amendment operation | movement as 2nd Embodiment by software processing. It is the block diagram which showed the structure of the modification of the reproducing | regenerating apparatus (electronic device) of 2nd Embodiment. It is the block diagram shown about the internal structure of the reproducing | regenerating apparatus (electronic device) as the 3rd Embodiment of this invention. It is the block diagram shown about the internal structure of the compression encoding / decoding part with which the inside of the reproducing | regenerating apparatus of 3rd Embodiment is provided. It is a figure for demonstrating the quantization error between bands. Similarly, it is a figure for demonstrating the quantization error between bands. It is the block diagram shown about the internal structure of the signal processing apparatus (quantization error correction | amendment part) of 3rd Embodiment. It is the flowchart shown about the processing operation which should be performed in order to implement | achieve correction | amendment operation | movement as 3rd Embodiment by software processing. It is the block diagram which showed the structure of the modification of the reproducing | regenerating apparatus (electronic device) of 3rd Embodiment.

Explanation of symbols

  1,40,50 playback device, 2 storage unit, 3 demodulation unit, 4 compression code decoding unit, 4a high frequency interpolation unit, 5 DSP, 6 bus, 7 system controller, 8 operation unit, 9 display unit, 10 format analysis 11, 11 inverse quantization processing unit, 11a Huffman coding unit, 11b inverse quantization unit, 11c rescaling unit, 12 stereo processing unit, 12a M / S stereo processing unit, 12b prediction processing unit, 12c intensity stereo processing Unit, 12d TNS unit, 13 adaptive block length switching inverse MDCT unit, 14 gain control unit, 21 Peak detection unit, 22 autocorrelation calculation unit, 23 correlation maximum shift amount detection unit, 24 Rate calculation unit, 25 high frequency signal addition unit 26 Interpolation determination unit, 30 MDCT conversion unit, 31 Inverse MDCT conversion unit, 41 Flattened part extraction unit, 42, 52 Determination unit, 43,53 Prediction processing unit, 44,54 replacement unit, 51 Error boundary extraction unit

Claims (28)

Correlation calculation means for performing autocorrelation calculation when the self-signal is sequentially shifted with respect to the self-signal with respect to a signal in a band of a predetermined frequency or higher in the audio signal subjected to predetermined information compression processing;
Based on the result of the autocorrelation calculation, a shift amount detecting means for obtaining a maximum correlation shift amount when the correlation is highest,
A ratio calculation means for calculating a ratio of each amplitude value at each frequency point separated by an amount based on the correlation maximum shift amount obtained by the shift amount detection means in a band of the predetermined frequency or higher;
The amplitude value of each frequency point to be interpolated in which a signal is missing in a band equal to or higher than the predetermined frequency, and the amplitude value at a frequency point that is separated from each frequency point to be interpolated by an amount based on the maximum correlation shift amount An interpolation means for interpolating with a value calculated based on the ratio;
A signal processing apparatus comprising: The correlation calculation means is
After detecting a peak value of an amplitude value in each divided band when a band of the predetermined frequency or higher is divided in units of divided bands including n frequency points, each peak value is detected by the self signal. Autocorrelation calculation as
The signal processing apparatus according to claim 1. The ratio calculation means is
The ratio of each amplitude value at each frequency point of a plurality of sets separated by an amount based on the maximum correlation shift amount is averaged and calculated as the value of the ratio.
The signal processing apparatus according to claim 1. The band above the predetermined frequency is a frequency band after the subband in which the bandwidth is constant among the subbands divided in the process of frequency correlation encoding by the AAC scheme.
The signal processing apparatus according to claim 1. Furthermore, a determination means for determining whether or not a signal is present in a band of the predetermined frequency or higher in the original signal before encoding is provided,
At least the interpolation means
As a result of the determination by the determination means, an interpolation operation is performed only when a signal is present in the band of the predetermined frequency or higher in the original signal.
The signal processing apparatus according to claim 1. The determination means is
Based on the autocorrelation calculation result by the correlation calculation means, it is determined whether or not there is a signal in a band of the original signal above the predetermined frequency,
The signal processing apparatus according to claim 5. A correlation calculation procedure for performing autocorrelation calculation when the self-signal is sequentially shifted with respect to the self-signal with respect to a signal of a band of a predetermined frequency or higher in the audio signal subjected to predetermined information compression processing;
Based on the result of the autocorrelation calculation, a shift amount detection procedure for obtaining a maximum correlation shift amount when the correlation is highest,
A ratio calculation procedure for calculating a ratio of each amplitude value at each frequency point separated by an amount based on the correlation maximum shift amount obtained by the shift amount detection procedure in a band of the predetermined frequency or higher;
The amplitude value of each frequency point to be interpolated in which a signal is missing in a band equal to or higher than the predetermined frequency, and the amplitude value at a frequency point that is separated from each frequency point to be interpolated by an amount based on the maximum correlation shift amount An interpolation procedure to interpolate with the value calculated based on the ratio;
A signal processing method comprising: A playback device for playing back at least an audio signal recorded on a recording medium,
Audio signal acquisition means for reading from the recording medium and obtaining an audio signal subjected to predetermined information compression processing;
Correlation calculation means for performing autocorrelation calculation when the self-signal is sequentially shifted with respect to the self-signal with respect to a signal in a band of a predetermined frequency or higher in the audio signal obtained by the audio signal acquisition means;
Based on the result of the autocorrelation calculation, a shift amount detecting means for obtaining a maximum correlation shift amount when the correlation is highest,
A ratio calculation means for calculating a ratio of each amplitude value at each frequency point separated by an amount based on the correlation maximum shift amount obtained by the shift amount detection means in a band of the predetermined frequency or higher;
The amplitude value of each frequency point to be interpolated in which a signal is missing in a band equal to or higher than the predetermined frequency, and the amplitude value at a frequency point that is separated from each frequency point to be interpolated by an amount based on the maximum correlation shift amount An interpolation means for interpolating with a value calculated based on the ratio;
A playback apparatus comprising: A reproduction method for reproducing at least an audio signal recorded on a recording medium,
An audio signal acquisition procedure for obtaining an audio signal that has been subjected to predetermined information compression processing by reading from the recording medium;
A correlation calculation procedure for performing autocorrelation calculation when the self-signal is sequentially shifted with respect to the self-signal with respect to a signal in a band of a predetermined frequency or higher in the audio signal obtained by the audio signal acquisition procedure;
Based on the result of the autocorrelation calculation, a shift amount detection procedure for obtaining a maximum correlation shift amount when the correlation is highest,
A ratio calculation procedure for calculating a ratio of each amplitude value at each frequency point separated by an amount based on the correlation maximum shift amount obtained by the shift amount detection procedure in a band of the predetermined frequency or higher;
The amplitude value of each frequency point to be interpolated in which a signal is missing in a band equal to or higher than the predetermined frequency, and the amplitude value at a frequency point that is separated from each frequency point to be interpolated by an amount based on the maximum correlation shift amount An interpolation procedure to interpolate with the value calculated based on the ratio;
A playback method comprising: Obtaining means for obtaining an audio signal subjected to predetermined information compression processing;
Correlation calculation means for performing autocorrelation calculation when the self-signal is sequentially shifted with respect to the self-signal with respect to a signal having a band of a predetermined frequency or higher in the audio signal acquired by the acquisition means;
Based on the result of the autocorrelation calculation, a shift amount detecting means for obtaining a maximum correlation shift amount when the correlation is highest,
A ratio calculation means for calculating a ratio of each amplitude value at each frequency point separated by an amount based on the correlation maximum shift amount obtained by the shift amount detection means in a band of the predetermined frequency or higher;
The amplitude value of each frequency point to be interpolated in which a signal is missing in a band equal to or higher than the predetermined frequency, and the amplitude value at a frequency point that is separated from each frequency point to be interpolated by an amount based on the maximum correlation shift amount An interpolation means for interpolating with a value calculated based on the ratio;
An electronic device comprising: Prediction signal generation means for generating a prediction signal obtained by predicting an original signal before compression for an audio signal subjected to predetermined information compression processing;
With respect to the audio signal, an error candidate part detecting means for detecting, as an error candidate part, a part whose amplitude value is a predetermined value on the frequency axis;
Substitution means for replacing the amplitude value of the error candidate part based on the value of the prediction signal based on the result of comparing the amplitude value in the error candidate part and the value of the prediction signal;
A signal processing apparatus comprising:   12. The signal processing apparatus according to claim 11, wherein the error candidate part detection means detects the error candidate part by limiting a band.   12. The signal processing apparatus according to claim 11, wherein the error candidate part detecting means detects a part having the same value as a resolution value set in the information compression process. The replacement means is:
Based on the result of comparing the amplitude value of the error candidate portion with the value of the prediction signal based on the resolution value set at the time of the information compression process, the amplitude value of the error candidate portion is changed to the value of the prediction signal. Replace based on,
The signal processing apparatus according to claim 11.   The signal processing apparatus according to claim 11, wherein the prediction unit generates the prediction signal by a prediction process using a second-order backward adaptive lattice type.   The signal processing apparatus according to claim 11, wherein the prediction unit generates the prediction signal by an approximate expression or an interpolation polynomial. A prediction signal generation procedure for generating a prediction signal obtained by predicting an original signal before compression for an audio signal subjected to predetermined information compression processing;
For the audio signal, an error candidate part detection procedure for detecting, as an error candidate part, a part whose amplitude value is a predetermined value on the frequency axis;
A replacement procedure for replacing the amplitude value of the error candidate part based on the value of the prediction signal based on the result of comparing the amplitude value in the error candidate part and the value of the prediction signal;
A signal processing method comprising: A playback device for playing back at least an audio signal recorded on a recording medium,
Audio signal acquisition means for reading from the recording medium and obtaining an audio signal subjected to predetermined information compression processing;
About the audio signal obtained by the audio signal acquisition means, a prediction signal generation means for generating a prediction signal obtained by predicting the original signal before compression;
With respect to the audio signal, an error candidate part detecting means for detecting, as an error candidate part, a part whose amplitude value is a predetermined value on the frequency axis;
Substitution means for replacing the amplitude value of the error candidate part based on the value of the prediction signal based on the result of comparing the amplitude value in the error candidate part and the value of the prediction signal;
A playback apparatus comprising: A reproduction method for reproducing at least an audio signal recorded on a recording medium,
An audio signal acquisition procedure for obtaining an audio signal that has been subjected to predetermined information compression processing by reading from the recording medium;
For the audio signal obtained by the audio signal acquisition procedure, a prediction signal generation procedure for generating a prediction signal predicting the original signal before compression;
For the audio signal, an error candidate part detection procedure for detecting, as an error candidate part, a part whose amplitude value is a predetermined value on the frequency axis;
A replacement procedure for replacing the amplitude value of the error candidate part based on the value of the prediction signal based on the result of comparing the amplitude value in the error candidate part and the value of the prediction signal;
A playback method comprising: Obtaining means for obtaining an audio signal subjected to predetermined information compression processing;
For the audio signal acquired by the acquisition unit, a prediction signal generation unit that generates a prediction signal obtained by predicting the original signal before compression;
With respect to the audio signal, an error candidate part detecting means for detecting, as an error candidate part, a part whose amplitude value is a predetermined value on the frequency axis;
Substitution means for replacing the amplitude value of the error candidate part based on the value of the prediction signal based on the result of comparing the amplitude value in the error candidate part and the value of the prediction signal;
An electronic device comprising: Prediction signal generation means for generating a prediction signal that predicts an original signal before compression of an audio signal that has been subjected to information compression processing that determines resolution for bit number allocation in a predetermined frequency band unit;
Boundary part detection means for detecting a boundary part between successive frequency bands having different resolution values for the audio signal,
A replacement means for replacing the amplitude value of the boundary portion based on the value of the prediction signal based on the result of comparing the amplitude value at the boundary portion and the value of the prediction signal;
A signal processing apparatus comprising: The replacement means is:
Based on the result of comparing the amplitude value of the boundary portion with the value of the prediction signal based on the resolution value set for the frequency band to which the boundary portion belongs, the amplitude value of the boundary portion is replaced based on the value of the prediction signal To
The signal processing apparatus according to claim 21, wherein:   The signal processing apparatus according to claim 21, wherein the prediction unit generates the prediction signal by a prediction process using a second-order backward adaptive lattice type.   The signal processing apparatus according to claim 21, wherein the prediction means generates the prediction signal by an approximate expression or an interpolation polynomial. A prediction signal generation procedure for generating a prediction signal that predicts an original signal before compression of an audio signal that has been subjected to information compression processing that determines resolution for bit number allocation in a predetermined frequency band unit;
For the audio signal, a boundary part detection procedure for detecting a boundary part between successive frequency bands each having a different resolution value;
A replacement procedure for replacing the amplitude value of the boundary portion based on the value of the prediction signal based on the result of comparing the amplitude value at the boundary portion and the value of the prediction signal;
A signal processing method comprising: A playback device for playing back at least an audio signal recorded on a recording medium,
Audio signal acquisition means for performing reading from the recording medium and obtaining an audio signal subjected to information compression processing for determining a resolution for bit number allocation in a predetermined frequency band unit;
About the audio signal obtained by the audio signal acquisition means, a prediction signal generation means for generating a prediction signal obtained by predicting the original signal before compression;
Boundary part detection means for detecting a boundary part between successive frequency bands having different resolution values for the audio signal,
A replacement means for replacing the amplitude value of the boundary portion based on the value of the prediction signal based on the result of comparing the amplitude value at the boundary portion and the value of the prediction signal;
A playback apparatus comprising: A reproduction method for reproducing at least an audio signal recorded on a recording medium,
An audio signal acquisition procedure for performing reading from the recording medium and obtaining an audio signal subjected to information compression processing for determining a resolution for bit number allocation in a predetermined frequency band unit;
For the audio signal obtained by the audio signal acquisition procedure, a prediction signal generation procedure for generating a prediction signal predicting the original signal before compression;
For the audio signal, a boundary part detection procedure for detecting a boundary part between successive frequency bands each having a different resolution value;
A replacement procedure for replacing the amplitude value of the boundary portion based on the value of the prediction signal based on the result of comparing the amplitude value at the boundary portion and the value of the prediction signal;
A playback method comprising: An acquisition means for acquiring an audio signal subjected to information compression processing for determining a resolution for bit number allocation in a predetermined frequency band unit;
For the audio signal acquired by the acquisition unit, a prediction signal generation unit that generates a prediction signal obtained by predicting the original signal before compression;
Boundary part detection means for detecting a boundary part between successive frequency bands having different resolution values for the audio signal,
A replacement means for replacing the amplitude value of the boundary portion based on the value of the prediction signal based on the result of comparing the amplitude value at the boundary portion and the value of the prediction signal;
An electronic device comprising: