JP2005345993A - Device and method for sound decoding - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a device and method for sound decoding that detect errors occurring in sample data and corrects errors, when errors are detected. <P>SOLUTION: The sample data multiplexed with an encoded stream are decoded and converted into frequency data of respective bands, and an error in the converted frequency data is detected, by comparing the absolute value of the difference between two successive frequency data with the threshold. When an error detection part 8 detects the error, frequency data of it are corrected by a sample correcting part 9. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to an acoustic decoding apparatus and an acoustic decoding method for decoding encoded audio data.

  Various encoding (compression) techniques have been developed in order to efficiently store and transmit audio data. As a representative method of compressing audio data with high efficiency and high sound quality, the audio data is converted into a frequency band by subband coding or MDCT (Modified Discrete Cosine Transform), and a human auditory model There is a method of reducing the data size by reducing information that cannot be heard by human ears by using, and further reducing the redundancy of information by Huffman coding or the like. Typical examples are MPEG1, 2, 4 audio, which has become an ISO (International Organization for Standardization) / IEC (International Electro for Standardization) standard by MPEG (Moving Picture Experts Group), Dolby Digital (trademark) developed by Dolby. And ATRAC (trademark) developed by Sony Corporation. There are a plurality of MPEG audio standards, and the MPEG-1 and MPEG-2 audio layer 3 are commonly used as MP3 in personal computers and portable audio players. MPEG-2 AAC is adopted as a compression method for digital broadcasting. Dolby Digital (trademark) is adopted as an audio compression method for DVD (Digital Versatile Disc). ATRAC (trademark) is mainly used in portable audio players such as Mini Disk (trademark) and Memory Stick Walkman (trademark).

  Here, the MPEG-1 audio layer 2 system widely used in the field of video CD (Compact Disc) and embedded devices will be described. One frame in the MPEG-1 audio layer 2 encoded stream is roughly divided into a header including information indicating the head of the frame, and bit allocation including information on how much information is allocated to each frequency band. Describes information, scale factor selection information that contains information about which subband the scale factor when frequency data was normalized to, and the index value of the multiplication coefficient when frequency data was normalized Scale data, sample data in which compressed acoustic information is described as a frequency index value, and ancillary data are multiplexed.

  FIG. 9 is a block diagram showing a conventional acoustic decoding device 101. The acoustic decoding device 101 is separated into a frame buffer 102 that stores an encoded stream, a data separator 103 that separates each data from the stored encoded stream, and a header analyzer 104 that analyzes the separated header. A side information decoding unit 105 that decodes bit allocation information, scale factor selection information, and scale factor, a sample dequantization unit 106 that decodes sample data and performs inverse quantization using the bit allocation information, and a scale factor A sample denormalization unit 107 that calculates frequency data and a band synthesis unit 108 that synthesizes audio data from a plurality of bands of frequency data are provided.

  Next, the operation of the acoustic decoding device 101 will be described. The encoded stream input to the acoustic decoding device 101 is held in the frame buffer 102. The data separation unit 103 separates each data multiplexed in the encoded stream from the encoded stream held in the frame buffer 102. Since one frame of the encoded stream of the MPEG-1 audio layer 2 system includes the above-described data in a specified fixed bit length and in a specified order, the data separation unit 103 Each data is separated by sequentially taking out the specified bits sequentially from the head of the.

  The header separated by the data separation unit 103 is analyzed by the header analysis unit 104, and the sync word indicating the head of the data, the number of channels, and the mode information indicating the compression mode are extracted. Here, when CRC information is included in the header, the header analysis unit 104 adds CRC data calculated from the header, bit allocation information, and scale factor selection information, and is added to the frame in advance during encoding. The cyclic redundancy check is performed by comparing with the CRC data.

  If an error is detected in the cyclic redundancy check, the entire frame is determined to be an error frame because it cannot be decoded. The error frame is, for example, muted or interpolated with a normal previous frame.

  When no error is detected, the side information decoding unit 105 decodes the bit allocation information, the scale factor selection information, and the scale factor separated by the data separation unit 103.

  The sample data separated by the data separation unit 103 is decoded by the sample inverse quantization unit 106, and the decoded sample data is inversely quantized using the bit allocation information decoded by the side information decoding unit 105.

  The sample data dequantized by the sample dequantization unit 106 is multiplied by the scale factor decoded by the side information decoding unit 105 by the sample denormalization unit 107 and converted into frequency data. The MPEG-1 audio layer 2 format data at the stage of conversion to frequency data has a structure in which 32 groups of subband data (frequency data) consisting of three subframes continue in 12 groups in the time axis direction. .

  The band synthesizing unit 108 synthesizes audio data by performing subband synthesis or inverse MDCT operation on the frequency data converted by the sample denormalization unit 107. In the MPEG-1 audio layer 2 system, 32 samples of PCM (Pulse Code Modulation) data is output by calculating 32 samples of frequency data, and this is repeated 3 sub-frames x 12 groups per channel per channel. 1152 samples of PCM data are output.

JP-A-11-355145 JP-A-6-349198

  However, in the error detection based on the CRC information as described above, bit errors of data other than the header, bit allocation information, and scale factor selection information that are detection targets cannot be detected. That is, when a bit error or data destruction occurs in the scale factor or sample data, the error cannot be detected and noise is generated.

  Patent Document 1 describes a method of comparing adjacent partial areas with respect to an error in a scale factor, but does not describe any countermeasures regarding the error of sample data because its influence is small. However, if data destruction occurs in the sample data, a minute noise is generated only in the data portion, and there is a possibility that the listener recognizes it as an irritating sound. In addition, since most of the encoded stream is occupied by the sample data, the possibility that an error will occur in this portion of data is considerably higher than that of the header and the bit allocation information portion.

  Patent Document 2 proposes a method of inserting an error correction signal into ancillary data. However, this method requires additional processing for inserting an error correction signal on the encoding device side. In addition, an acoustic decoding device using this method cannot perform error correction on an encoded stream in which no correction signal is inserted. Furthermore, as a result of using extra bits for the error correction signal portion, the amount of information assigned to the sample data may be reduced, and sound quality may be impaired.

  The present invention has been proposed in view of such conventional circumstances, and detects an error that occurs in sample data, and if an error is detected, an acoustic decoding device that corrects the error. It is another object of the present invention to provide an acoustic decoding method.

  Therefore, in order to achieve the above-described object, the acoustic decoding device according to the present invention divides audio data divided at predetermined time intervals into frequency data of a plurality of frequency bands, and the frequency of the plurality of frequency bands. In an acoustic decoding apparatus that decodes an encoded stream obtained by dividing data into a plurality of groups so that each group includes a certain number of frequency data in time series for each frequency band, the encoded stream is A decoding unit that decodes and restores the frequency data divided into the plurality of groups, compares an absolute value of a difference between the two frequency data restored by the restoring unit and a threshold value, and the absolute value of the difference is An error detection means for determining one of the two frequency data obtained by the difference as an error when the threshold is greater than the threshold; and the error And correction means for correcting the determined frequency data, is characterized by having a synthesizing means for synthesizing the audio data from the divided frequency data to the plurality of groups.

  Also, the audio decoding method according to the present invention divides audio data divided at predetermined time intervals into frequency data of a plurality of frequency bands, and each group has frequency data of the plurality of frequency bands for each frequency band. In an acoustic decoding method for decoding an encoded stream obtained by dividing into a plurality of groups so as to include a certain number of frequency data in time series order, the encoded stream is decoded and divided into the plurality of groups Comparing the restored frequency data and the absolute value of the difference between the two frequency data restored by the restoring step and the threshold value, and the absolute value of the difference is greater than the threshold value, An error detection step of determining one of the two frequency data obtained as a difference as an error, and correcting the frequency data determined as the error And correcting step is characterized by having a synthesis step of synthesizing the audio data from the divided frequency data to the plurality of groups.

  According to the present invention, it is possible to detect an error in sample data multiplexed in an encoded stream by comparing an absolute value of a difference between two consecutive frequency data and a threshold value.

  Further, when an error is detected, by selecting a correction method according to the frequency data, it is possible to effectively suppress noise generated by bit errors or data destruction in the sample data.

  The acoustic decoding device and the acoustic decoding method shown as specific examples of the present invention detect an error included in sample data at the stage where the sample data is dequantized and denormalized into frequency data, and an error is detected. Is a correction process. In this embodiment, the audio data of the MPEG-1 audio layer 2 method is used for explanation, but the present invention is not limited to this.

  Hereinafter, specific embodiments to which the present invention is applied will be described in detail with reference to the drawings. FIG. 1 is a diagram showing a format of an MPEG-1 audio layer 2 encoded stream. One frame of the encoded stream is roughly divided into a header, bit allocation information, scale factor selection information, scale factor, sample data, and ancillary data. Each data will be described below.

  The header includes a sync word indicating the head of data, a layer number, a bit rate indicating the amount of transfer data, a mode indicating the number of channels and a compression mode, a sampling frequency, CRC (Cyclic Redundancy Check) information, and the like.

  When this CRC information is included, CRC data calculated from the header, bit allocation information, and scale factor selection information at the time of encoding is added to the frame. That is, the bit error protection targets by the CRC information are header, bit allocation information, and scale factor selection information. This CRC data is used for cyclic redundancy check which is an error check at the time of decoding.

  If CRC information is not included, errors cannot be detected at the time of decoding, so if a bit error or data corruption occurs in the encoded stream, 1 frame (24 msec for 48 KHz sampling, 32 KHz sampling) Noise is generated for 32 msec).

  The bit allocation information includes information on how much information is allocated to each separated frequency band.

  In the scale factor, an index value of a multiplication coefficient when the frequency data is normalized is described. The scale factor selection information includes information on which subband the scale factor when normalized is used.

  In the sample data, encoded audio data is described as a frequency band index value.

  FIG. 2 is a block diagram showing a configuration of the acoustic decoding device 1 according to the present embodiment. The acoustic decoding device 1 is separated into a frame buffer 2 that holds an input encoded stream, a data separator 3 that separates the encoded stream into data, and a header analyzer 4 that analyzes the separated header. Side information decoding unit 5 that decodes the bit allocation information, scale factor selection information, and scale factor, and a sample dequantization unit that decodes the sample data separated from the encoded stream and performs inverse quantization using the bit allocation information 6, a sample denormalization unit 7 that calculates frequency data using a scale factor, an error detection unit 8 that detects an error in frequency data, and correction of frequency data when an error is detected in the error detection unit 8 A sample correction unit 9 for performing audio data synthesis from frequency data of a plurality of bands And a band synthesis portion 10 that.

  The frame buffer 2 holds the input encoded stream in units of frames.

  The data separation unit 3 separates each data multiplexed in the encoded stream from the encoded stream held in the frame buffer 2. As shown in FIG. 1, a fixed bit in which a header, bit allocation information, scale factor selection information, scale factor, sample data, and ancillary data are defined in one frame of an MPEG-1 audio layer 2 encoded stream The data separation unit 3 sequentially takes out prescribed bits from the head of the encoded stream held in the frame buffer 2 and separates each data.

  The header analysis unit 4 analyzes the sync word and mode in the header separated by the data separation unit 3. Here, when CRC information is included in the header, the header analysis unit 4 adds CRC data calculated from the header, bit allocation information and scale factor selection information, and is added to the frame in advance during encoding. The cyclic redundancy check is performed by comparing with the CRC data.

  The side information decoding unit 5 decodes the bit allocation information, the scale factor selection information, and the scale factor separated by the data separation unit 3.

  The sample dequantization unit 6 decodes the sample data separated by the data separation unit 3 and dequantizes the sample data using the bit allocation information decoded by the side information decoding unit 5.

  The sample denormalization unit 7 calculates the frequency data by multiplying the sample data dequantized by the sample dequantization unit 6 by the scale factor decoded by the side information decoding unit 5. As shown in FIG. 3A, the data of the MPEG-1 audio layer 2 method at the stage where the frequency data is calculated includes subband data (frequency data) of 32 bands including three subframes in the time axis direction. This is a structure that continues to 12 groups. This frequency data is assigned subband numbers from 0 to 31 in ascending order of frequency. Hereinafter, the frequency data is represented by S [k] [t] [n] using the sample number k (0 to 11), the subframe number (0 to 2) t, and the subband number n (0 to 31). I will do it. Note that the subband boundary values shown in FIG. 3A are the frequency bands (subbands) encoded at the time of encoding audio data and the frequency bands (subbands) that were not encoded. Indicates the boundary.

  The error detection unit 8 detects an error in the frequency data calculated by the sample denormalization unit 7. The error is detected by comparing the absolute value of the frequency data difference with a threshold value. For example, the absolute value of the difference between two frequency data with consecutive subband numbers (| S [k] [t] [n + 1]-S [k] [t] [n] |) is set in advance. If the absolute value of the difference is greater than the threshold value, the frequency data S [k] [t] [n] and S [k] [t] [n + 1 ] With the larger absolute value is judged as an error.

  The sample correction unit 9 corrects the frequency data in which an error is detected by the error detection unit 8. For example, the frequency data in which the error is detected is corrected so that the absolute value of the difference between two frequency data having consecutive subband numbers becomes a threshold value.

  The band synthesizing unit 10 synthesizes audio data by performing subband synthesis or inverse MDCT (Modified Discrete Cosine Transform) on the frequency data of each band. In the MPEG-1 audio layer 2 system, 32 samples of PCM (Pulse Code Modulation) data are output by calculating the frequency data S of 32 samples as shown in FIG. Repeats 12 groups, and outputs 1152 samples of PCM data per channel in one frame.

  Next, the error detection unit 8 of the acoustic decoding device 1 in the present embodiment will be described in detail. The frequency data before subband synthesis is related to the continuous frequency domain direction and the time axis direction. For example, in the case of a sound source that is normally heard, such as a human voice or instrument sound, the sound does not consist of only one sample of frequency data, and the value is spread both in the frequency domain direction and in the time axis direction. A series of sounds. However, when noise occurs due to a part of the frequency data being broken, a large change occurs momentarily in the frequency domain direction or the time axis direction. In the present invention, noise is detected by examining a sudden change in value in the frequency band.

  FIG. 4 is a block diagram showing a configuration of the error detection unit 8 to which the present invention is applied. The error detection unit 8 compares the frequency data comparison unit 82 that compares two frequency data having consecutive subband numbers based on the frequency threshold value table 81, and the sample number stored in the previous sample storage unit 83 is one smaller. A time axis data comparison unit 85 that compares the frequency data and the current frequency data based on the time axis threshold value table 84 is provided.

  The error detection unit 8 detects frequency data determined as an error as error data in either the frequency data comparison unit 82 or the time axis data comparison unit 85. The frequency data input to the error detection unit 8 is, for example, subband data (frequency data) of 32 bands including three subframes as in the MPEG-1 audio layer 2 system shown in FIG. It is configured to continue 12 groups in the axial direction. Here, each group includes three frequency data in time series order for each frequency band, and corresponds to a sample number k (0 to 11).

The frequency data comparison unit 82 includes two band frequency data S [k] [t] [n] and S [k] [t] [n + 1] that are continuous in the frequency domain direction calculated by the sample denormalization unit 7. ] Is calculated as shown in the equation (1).
Diff [n] = | S [k] [t] [n + 1] S [k] [t] [n] | (1)
That is, according to the above equation (1), the difference between the frequency data S of consecutive subband numbers is obtained with the same sample number and subframe number.

  Then, the frequency data comparison unit 82 compares the calculated absolute value Diff [n] of the difference with the threshold value f [n] of the frequency threshold value table 81, and the absolute value Diff [n] of the difference is obtained. If it is larger than the threshold value f [n], the frequency data S [k] [t] [n] and S [k] [t] [n + 1] with the larger absolute value is determined as error data. The error information of the sample number k, subframe number t, and subband number n of the frequency data is output. Further, the frequency data S [k] [t] [n] and S [k] [t] [n + 1] for which the difference is obtained are output to the time axis data comparison unit 85.

  The frequency threshold value table 81 stores, for example, a threshold value f [n] corresponding to the subband number n as shown in FIG. Usually, the amount of change in the value of the frequency data is large in the low range, and the amount of change in the value of the frequency data is small in the high range, so it is preferable to create the frequency threshold value table 81 in consideration of this. The threshold shown in FIG. 5 corresponds to 26 subbands from 0 to 25. This is because the 26 subbands from 0 to 25 out of the 32 subbands shown in FIG. This is an example in which a band is an effective subband. That is, the subband boundary value shown in FIG.

The time axis data comparison unit 85 has the same frequency band and the frequency data S [k] [t] [n] stored in the previous sample storage unit 83 and the next frequency data S [k + 1] in time. The absolute value Diff_t [n] of the difference between the frequency data S and [t] [n] is calculated as shown in equation (2).
Diff_t [n] = | S [k + 1] [t] [n] S [k] [t] [n] | (2)
That is, according to the above equation (2), the subband number and the subframe number are the same, and the difference is obtained from the frequency data S having consecutive sample numbers.

  Then, when compared with the threshold value t [n] of the time axis threshold value table 84, if this Diff_t [n] is larger than the threshold value t [n], the frequency data S [k] [t] [n ] And S [k + 1] [t] [n], the one with the larger absolute value is determined as error data, and error information of the frequency data sample number k, subframe number t, and subband number n is output. To do.

  For example, the time axis threshold value table 84 stores a threshold value t [n] corresponding to the subband number n as shown in FIG. Similar to the frequency threshold value table 81 shown in FIG. 5, the change amount of the frequency data value is usually large in the low frequency range, and the change amount of the frequency data value is small in the high frequency range. It is preferable to create a threshold table 84. Further, the thresholds in the time axis direction corresponding to the 26 subbands 0 to 25 shown in FIG. 6 are the same as the threshold values in the frequency domain direction shown in FIG. This is an example in which 26 subbands from 0 to 25 are effective subbands.

  It should be noted that the frequency data to be corrected is the larger of the two frequency data determined as an error by the frequency data comparison unit 82 or the time axis data comparison unit 85, but the absolute value is small. It's also good with you.

  In addition, the error detection unit 8 detects an error when either the frequency data comparison unit 82 or the time axis data comparison unit 85 determines that an error has occurred, but the frequency data comparison unit 82 and the time axis data An error may be detected when both comparators 85 determine that an error has occurred.

  In addition, the error detection unit 8 includes an absolute value of a difference between frequency data consecutive in the frequency domain direction, an absolute value of a difference between frequency data in the time axis direction, and a difference between frequency data adjacent in the frequency domain direction and the time axis direction. These may be combined with an absolute value and compared with a preset threshold value. Here, the absolute value of the difference between the frequency data in the diagonal direction is obtained from the frequency data S [k] [t] [n] and the frequency data S [k + 1] [t] [n + 1] in the diagonal direction. Calculated.

  As described above, the error detection unit 8 uses the absolute value of the difference between the frequency data continuous in the frequency domain direction, the time axis direction, and the like, and compares the frequency data of each band with a previously calculated threshold value. An error can be detected in units.

  Next, the sample correction unit 9 of the acoustic decoding device 1 in the present embodiment will be described. The sample correction unit 9 corrects the frequency data in which an error is detected by the error detection unit 8 and performs processing for reducing noise in the sample data.

  FIG. 7 is a block diagram showing the configuration of the sample correction unit 9. The sample correction unit 9 selects a method for correcting the frequency data based on the error information input from the error detection unit 8, and the frequency data by the correction method selected by the correction method selection unit 91. Frequency data correction units 92, 93, and 94 for correction are provided.

  Here, the correction method performed in each frequency data correction part 92,93,94 is demonstrated first. Further, the correction of the frequency data is performed by the same method whether the frequency data comparison unit 82 determines that an error has occurred or the time axis data comparison unit 85 determines that an error has occurred. Hereinafter, frequency data continuous in the frequency domain direction or the time axis direction are represented by S [m] and S [m + 1], respectively, and | S [m + 1] |> | S [m] |. That is, the correction target is described as S [m + 1].

(Correction method 1)
The frequency data correction unit 92 corrects S [m + 1] so that the absolute value of the difference between successive frequency data becomes a threshold value as shown in equation (3). Note that Th [m] is the threshold value f [n] stored in the frequency threshold value table 81 or the threshold value t [n] stored in the time axis threshold value table 84.
if (S [m + 1]> 0) S [m + 1] = S [m] + Th [m]
else S [m + 1] = S [m] −Th [m] (3)
Thus, by correcting the frequency data S [m + 1] to be corrected so that the absolute value of the difference from the frequency data S [m] that is not the correction target becomes the threshold value Th [m], Abrupt changes between frequency data can be eliminated, and noise anxious during listening can be reduced. This method is particularly effective in a low frequency region where the amount of change in the frequency data value is large even during normal times.

(Correction method 2)
The frequency data correction unit 93 replaces data having a large absolute value with data having a small absolute value as shown in the equation (4).
S [m + 1] = S [m] (4)
By replacing the frequency data with data having a small absolute value and correcting it, the absolute value of the difference between the two can be made zero. As a result, it is possible to suppress noise that is a concern during listening.

(Correction method 3)
The frequency data correction unit 94 replaces data having a large absolute value with 0 as shown in the equation (5).
S [m + 1] = 0 (5)
By setting the frequency data to 0, the frequency component is eliminated, thereby suppressing noise that is a concern during listening. This method is particularly effective in the high frequency portion where the frequency data value is close to zero.

  Next, the operation of the correction method selection unit 91 that selects the above three correction methods will be described. The correction method selection unit 91 selects a correction method based on the error information input from the error detection unit 8.

(Operation example 1)
The correction method selection unit 91, as shown below, out of the continuous frequency domain direction or time axis direction frequency data S [m] and S [m + 1] absolute values that are not determined as errors by the error detection unit 8 The correction methods 1 and 2 are switched in accordance with the magnitude relationship between the small S [m] and the preset setting value th_num. The set value th_num is preferably a different value depending on the frequency domain direction or the time axis direction.
If S [m] <th_num, use correction method 2.
If th_num ≤ S [m], use correction method 1
In the case of S [m] <th_num, it is assumed that the frequency data S [m + 1] in which the error is detected suddenly changes from the frequency data S [m] smaller than the set value th_num. The correction method 2 for replacing the detected frequency data S [m + 1] with S [m] is selected. If the frequency data S [m] is 0, it is replaced with 0.

  In addition, when th_num ≦ S [m], it is assumed that the frequency data S [m] has changed from a large one, so the correction method 1 that suppresses the amount of change in the frequency data value to the threshold is selected. To do.

(Operation example 2)
As operation example 2, the correction method selection unit 91 switches between the correction methods 1, 2, and 3 according to the band to which the frequency data determined to be an error by the error detection unit 8, that is, the subband number n. The subband number of the frequency data in which the error is detected is included in the error information output from the error detection unit 8.

The correction selection unit 91 performs, for example, the following selection according to a subband number n assigned to each band and a table in which boundary values th_band1 and th_band2 in which the correction method changes are determined in advance. This boundary value is set based on the sampling frequency and the bit rate.
If n <th_band1, use correction method 1.
If th_band1 ≤ n <th_band2, use correction method 2.
If th_band2 ≤ n, use correction method 3.
In the case of n <th_band1, since it is assumed that an error occurs in a low-frequency portion where the amount of change in the frequency data value is large, the correction method 1 that suppresses the amount of change to the threshold value is selected.

  Further, in the case of th_band1 ≦ n <th_band2, it is assumed that an error occurs in the middle to high range where the change amount of the frequency data value is small. Therefore, the correction method 2 for changing the change amount to 0 is selected.

  In addition, when th_band2 ≦ n, it is normally assumed that an error occurs in the high frequency part where the frequency data often becomes 0, so the correction method 3 for replacing the frequency data with 0 is selected.

  As described above, the sample correction unit 9 can select a correction method based on the error information input from the error detection unit 8, and thus can perform correction suitable for the frequency data in which the error is detected.

  Next, the processing of the band synthesizing unit 10 will be described with reference to the flowchart shown in FIG. The band synthesizing unit 10 receives frequency data of each band including frequency data of the band corrected by the sample correcting unit 9. In the MPEG system, inverse quantization is performed for each frequency data of 32 bands, and band synthesis processing is performed.

  In step S1, in preparation for the 64 new V vectors to be generated, the previous 64 of the 1024 V vector groups are shifted according to equation (6).

  In step S2, 64 new V vectors are calculated from the input 32 frequency data by the equation (7).

  1024 V vectors are prepared by combining the 64 V vectors newly calculated in step S2 and the past 15 V vectors (64 × 15 = 980).

  In step S3, 512 U vector groups are generated from 1024 V vector groups according to equation (8).

  In step S4, the U vector group is multiplied by a coefficient Di, which is a coefficient used for window processing, by equation (9), and window processing is performed to generate a W vector group.

  The 32 W vector groups generated in step 4 are added in a predetermined direction, and the audio data PCM (Pulse Code Modulation) for one frame is expanded as shown in equation (10) (step S5).

  As described above, the acoustic decoding device according to the present embodiment detects an error in the sample data multiplexed in the encoded stream, and can correct the frequency data when an error is detected, as described above. Noise generated by bit errors and data destruction in sample data can be suppressed, and high-quality audio data can be provided.

  Although the present embodiment has been described using MPEG-1 audio layer 2 format audio data, the present invention is not limited to the MPEG-1 audio layer 2 format and can be applied to audio data of other formats.

It is a figure which shows the audio frame of the encoding stream of an MPEG-1 audio layer 2 system. It is a block diagram which shows the structure of the audio decoding apparatus in this Embodiment. It is a figure for demonstrating decoding of the sample data in an MPEG-1 audio layer 2 system. It is a block diagram which shows the structure of the error detection part in this Embodiment. It is a figure for demonstrating the frequency threshold value table in this Embodiment. It is a figure for demonstrating the time-axis threshold value table in this Embodiment. It is a block diagram which shows the structure of the sample correction part in this Embodiment. It is a flowchart which shows operation | movement of the zone | band synthetic | combination part in this Embodiment. It is a block diagram which shows the structure of the conventional acoustic decoding apparatus.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Acoustic decoding apparatus, 2 Frame buffer, 3 Data separation part, 4 Header analysis part, 5 Side information decoding part, 6 Sample dequantization part, 7 Sample denormalization part, 8 Error detection part, 9 Sample correction part, 10 Band synthesis unit, 81 frequency threshold table, 82 frequency data comparison unit, 83 previous sample storage unit, 84 time axis threshold table, 85 time axis data comparison unit, 91 correction method selection unit, 92, 93, 94 frequency Data correction unit, 101 acoustic decoding device, 102 frame buffer, 103 data separation unit, 104 header analysis unit, 105 side information decoding unit, 106 sample dequantization unit, 107 sample denormalization unit, 108 band synthesis unit

Claims (22)

Audio data divided at predetermined time intervals is divided into frequency data of a plurality of frequency bands, and each group includes a certain number of frequency data in time series for each frequency band. In an acoustic decoding device that decodes an encoded stream obtained by dividing into a plurality of groups and encoding,
Decoding means for decoding the encoded stream and restoring the frequency data divided into the plurality of groups;
When the absolute value of the difference between the two frequency data restored by the restoration means is compared with a threshold value, and the absolute value of the difference is larger than the threshold value, Error detection means for determining one of the errors,
Correction means for correcting the frequency data determined to be the error;
A sound decoding apparatus comprising: a synthesizing unit that synthesizes audio data from the frequency data divided into the plurality of groups.   2. The acoustic decoding device according to claim 1, wherein the absolute value of the difference is calculated from two frequency data having continuous frequency bands at the same time point in time series.   2. The acoustic decoding device according to claim 1, wherein the absolute value of the difference is calculated from two frequency data in which groups are continuous in the same frequency band.   The acoustic decoding device according to claim 1, wherein the absolute value of the difference is calculated from two frequency data in which a frequency band and a group are continuous.   The correction means compares the value of the frequency data that is not determined to be an error by the error detection means and a preset setting value, and selects a correction method for the frequency data determined to be the error. The acoustic decoding device according to claim 1.   In the correcting means, when the value of the frequency data not determined to be an error by the error detecting means is equal to or greater than the set value, the absolute value of the difference from the frequency data not determined to be the error becomes the threshold value. 6. The acoustic decoding apparatus according to claim 5, wherein the frequency data determined as the error is corrected.   When the frequency data value that has not been determined to be an error by the error detection means is smaller than the set value, the correction means determines that the frequency data value that has been determined to be an error is the frequency data value that has not been determined to be the error. The acoustic decoding device according to claim 5, wherein the acoustic decoding device is corrected by being replaced with the above.   2. The sound according to claim 1, wherein the correction means selects a correction method for correcting the frequency data determined to be an error according to a frequency band of the frequency data determined to be an error by the error detection means. Decoding device.   When the frequency band of the frequency data determined to be an error by the error detection means is smaller than a predetermined boundary value, the correction means has an absolute value of a difference from the frequency data not determined to be the error as the threshold value. The sound decoding apparatus according to claim 8, wherein the frequency data determined as the error is corrected.   When the frequency band of the frequency data determined to be an error by the error detection means is equal to or higher than the first boundary value and less than the second boundary value, the correction means sets the value of the frequency data determined to be the error as the value. 9. The acoustic decoding device according to claim 8, wherein the acoustic decoding device is corrected by replacing with a value of frequency data not determined to be an error.   When the frequency band of the frequency data determined as an error by the error detection means is equal to or greater than a predetermined boundary value, the correction means replaces the value of the frequency data determined as the error with 0 to correct it. The acoustic decoding device according to claim 8. Audio data divided at predetermined time intervals is divided into frequency data of a plurality of frequency bands, and each group includes a certain number of frequency data in time series for each frequency band. In an acoustic decoding method for decoding an encoded stream obtained by dividing into a plurality of groups and encoding,
A decoding step of decoding the encoded stream and recovering the frequency data divided into the plurality of groups;
The absolute value of the difference between the two frequency data restored by the restoration process is compared with a threshold value, and when the absolute value of the difference is larger than the threshold value, An error detection step of determining one as an error;
A correction process for correcting the frequency data determined to be the error;
A synthesis step of synthesizing audio data from the frequency data divided into the plurality of groups.   13. The acoustic decoding method according to claim 12, wherein the absolute value of the difference is calculated from two frequency data having continuous frequency bands at the same time point in time series.   13. The acoustic decoding method according to claim 12, wherein the absolute value of the difference is calculated from two frequency data in which groups are continuous in the same frequency band.   The acoustic decoding method according to claim 12, wherein the absolute value of the difference is calculated from two frequency data in which a frequency band and a group are continuous.   In the correction step, the frequency data value not determined to be an error in the error detection step is compared with a preset setting value, and a correction method for the frequency data determined to be the error is selected. The acoustic decoding method according to claim 12.   In the correction step, when the value of the frequency data that is not determined to be an error in the error detection step is equal to or greater than the set value, the absolute value of the difference from the frequency data that is not determined to be the error becomes the threshold value. The acoustic decoding method according to claim 16, wherein the frequency data determined to be the error is corrected.   In the correction step, when the value of the frequency data that is not determined to be an error in the error detection step is smaller than the set value, the value of the frequency data that is determined to be the error is the value of the frequency data that is not determined to be the error The acoustic decoding method according to claim 16, wherein the acoustic decoding method is modified by being replaced with the above.   13. The sound according to claim 12, wherein the correction step selects a correction method for correcting the frequency data determined to be the error according to a frequency band of the frequency data determined to be an error in the error detection step. Decryption method.   In the correction step, when the frequency band of the frequency data determined as an error in the error detection step is smaller than a predetermined boundary value, the absolute value of the difference from the frequency data not determined as the error is the threshold value. 20. The acoustic decoding method according to claim 19, wherein the frequency data determined as the error is corrected.   In the correction step, when the frequency band of the frequency data determined to be an error in the error detection step is equal to or higher than the first boundary value and less than the second boundary value, the value of the frequency data determined to be the error is The acoustic decoding method according to claim 19, wherein the acoustic decoding method is corrected by replacing with a value of frequency data not determined to be an error. In the correction step, when the frequency band of the frequency data determined to be an error in the error detection step is equal to or greater than a predetermined boundary value, the value of the frequency data determined to be the error is replaced with 0 and corrected. The acoustic decoding method according to claim 19.

Images