JP5407831B2 - Speech coding apparatus, speech coding method, speech coding program, speech decoding apparatus, speech decoding method, and speech decoding program - Google Patents

Description

  The present invention relates to a speech encoding apparatus, a speech encoding method, and a speech encoding program that encode a speech signal. The present invention also relates to a speech decoding apparatus, a speech decoding method, and a speech decoding program for decoding a speech signal encoded by the speech encoding apparatus, speech encoding method, and speech encoding program.

  2. Description of the Related Art Conventionally, a technique for performing voice compression based on human auditory characteristics by encoding a voice signal using frequency conversion such as MDCT is well known (for example, see Patent Document 1).

JP-A-2005-128404

  Conventionally, a frequency conversion coefficient obtained by frequency-converting an audio signal is used to identify an unnecessary part in the audio signal on hearing, but is effectively used in terms of reducing redundancy. There is no actual situation.

  The present invention has been made in view of the above circumstances, and provides a speech encoding device, a speech encoding method, and a speech encoding program capable of improving encoding efficiency by predicting a frequency conversion coefficient. The purpose is to do.

  It is another object of the present invention to provide a speech decoding apparatus, speech decoding method, and speech decoding program that can restore a speech signal from the data encoded as described above.

The speech encoding apparatus according to the present invention is
Frequency conversion means for frequency-converting a digital audio signal and calculating a frequency conversion coefficient group;
A dividing unit that divides the calculated frequency conversion coefficient group into a frequency set for each predetermined number;
For each frequency set, a predicted frequency set having the highest correlation is searched from prediction source data in which a frequency set in the past of the frequency set is stored as a predicted frequency set, and each frequency transform coefficient of the frequency set is derived. Predicting means for calculating the predicted value based on the corresponding frequency transform coefficient of the predicted frequency set obtained as a result of the search;
Encoding means for encoding coefficient data including at least the prediction value calculated by the prediction means and position information of frequency conversion coefficients corresponding to the prediction frequency set, and generating encoded coefficient data;
With respect to all the frequency transform coefficients of the frequency transform coefficient group, when the coding coefficient data is generated, the encoded data generating means for generating encoded data that aggregates all the encoded coefficient data,
The frequency set is sequentially added to the prediction source data as the predicted frequency set when the processing by the prediction unit is completed.

In addition, a speech encoding apparatus according to another aspect of the present invention includes:
Frequency conversion means for frequency-converting a digital audio signal and calculating a frequency conversion coefficient group;
Logarithmic conversion means for performing logarithmic conversion on the calculated frequency conversion coefficient group and calculating a logarithmic conversion coefficient group;
A dividing unit that divides the calculated logarithmic transformation coefficient group as a frequency set for each predetermined number;
For each frequency set, a predicted frequency set having the highest correlation is searched from prediction source data in which a frequency set in the past of the frequency set is stored as a predicted frequency set, and each logarithmic transformation coefficient of the frequency set is derived. Predicting means for calculating the predicted value based on the corresponding logarithmic transformation coefficient of the predicted frequency set obtained as a result of the search;
Coefficient data including the prediction value calculated by the prediction unit, position information of a logarithmic transformation coefficient corresponding to the prediction frequency set, and a sign of the frequency transformation coefficient corresponding to the logarithmic transformation coefficient of the frequency set. Encoding means for encoding and generating encoded coefficient data;
With respect to all logarithmic transform coefficients of the logarithmic transform coefficient group, when the coding coefficient data is generated, the encoded data generation means for generating encoded data that aggregates all the encoded coefficient data, and
The frequency set is sequentially added to the prediction source data as the predicted frequency set when the processing by the prediction unit is completed.

A speech encoding method according to the present invention includes:
A frequency conversion step of frequency-converting the digital audio signal and calculating a frequency conversion coefficient group;
A dividing step of dividing the calculated frequency transform coefficient group into a frequency set for each predetermined number;
For each frequency set, a predicted frequency set having the highest correlation is searched from prediction source data in which a frequency set in the past of the frequency set is stored as a predicted frequency set, and each frequency transform coefficient of the frequency set is derived. A predicting step of calculating a predicted value of a based on a corresponding frequency transform coefficient of a predicted frequency set obtained as a result of the search;
An additional step of adding the frequency set for which the processing in the prediction step has been completed to the prediction source data as the predicted frequency set;
An encoding step of encoding coefficient data including at least the calculated predicted value and position information of a corresponding frequency transform coefficient of the predicted frequency set, and generating encoded coefficient data;
An encoded data generation step for generating encoded data that aggregates all the encoded coefficient data when the encoded coefficient data is generated for all frequency converted coefficients of the frequency converted coefficient group; It is characterized by.

The speech encoding program according to the present invention is:
On the computer,
A frequency conversion step of frequency-converting the digital audio signal and calculating a frequency conversion coefficient group;
A dividing step of dividing the calculated frequency transform coefficient group into a frequency set for each predetermined number;
For each frequency set, a predicted frequency set having the highest correlation is searched from prediction source data in which a frequency set in the past of the frequency set is stored as a predicted frequency set, and each frequency transform coefficient of the frequency set is derived. A predicting step of calculating a predicted value of a based on a corresponding frequency transform coefficient of a predicted frequency set obtained as a result of the search;
An additional step of adding the frequency set for which the processing in the prediction step has been completed to the prediction source data as the predicted frequency set;
An encoding step of encoding coefficient data including at least the calculated predicted value and position information of a corresponding frequency transform coefficient of the predicted frequency set, and generating encoded coefficient data;
For all frequency transform coefficients of the frequency transform coefficient group, when the coding coefficient data is generated, an encoded data generation step of generating encoded data that aggregates all the encoded coefficient data is executed. It is characterized by that.

The speech decoding apparatus according to the present invention includes:
For each frequency conversion coefficient of the frequency conversion coefficient group corresponding to the digital audio signal, at least a predicted value for deriving the frequency conversion coefficient from the past frequency conversion coefficient and position information of the past frequency conversion coefficient are included. A speech decoding apparatus for restoring a digital speech signal from encoded data in which encoded coefficient data obtained by encoding coefficient data is stored,
Decoding means for reading out and decoding the encoded data from a storage device and restoring the coefficient data;
For each restored coefficient data, a frequency conversion coefficient corresponding to the position information is searched from prediction source data in which a frequency conversion coefficient in the past of the frequency conversion coefficient related to the coefficient data is stored, and the search Based on the frequency conversion coefficient obtained as a result and the predicted value, coefficient restoration means for restoring the frequency conversion coefficient related to the coefficient data;
Frequency inverse transform means for restoring a digital audio signal by performing frequency inverse transform on the frequency transform coefficient group obtained by restoring the frequency transform coefficients related to all coefficient data, and
The restored frequency transform coefficients are sequentially added to the prediction source data.

The speech decoding method according to the present invention includes:
For each frequency conversion coefficient of the frequency conversion coefficient group corresponding to the digital audio signal, at least a predicted value for deriving the frequency conversion coefficient from the past frequency conversion coefficient and position information of the past frequency conversion coefficient are included. An audio decoding method for restoring a digital audio signal from encoded data in which encoded coefficient data obtained by encoding coefficient data is stored,
A decoding step of reading out and decoding the encoded data from a storage device, and restoring each coefficient data;
For each restored coefficient data, a frequency conversion coefficient corresponding to the position information is searched from prediction source data in which a frequency conversion coefficient in the past of the frequency conversion coefficient related to the coefficient data is stored, and the search A coefficient restoration step for restoring the frequency conversion coefficient related to the coefficient data based on the frequency conversion coefficient obtained as a result and the predicted value;
An adding step of adding the restored frequency transform coefficient to the prediction source data;
A frequency inverse transform step for restoring a digital audio signal by performing frequency inverse transform on the frequency transform coefficient group obtained by restoring the frequency transform coefficients related to all coefficient data, To do.

A speech decoding program according to the present invention includes:
On the computer,
For each frequency conversion coefficient of the frequency conversion coefficient group corresponding to the digital audio signal, at least a predicted value for deriving the frequency conversion coefficient from the past frequency conversion coefficient and position information of the past frequency conversion coefficient are included. An audio decoding program for restoring a digital audio signal from encoded data in which encoded coefficient data obtained by encoding coefficient data is stored,
A decoding step of reading out and decoding the encoded data from a storage device, and restoring each coefficient data;
For each restored coefficient data, a frequency conversion coefficient corresponding to the position information is searched from prediction source data in which a frequency conversion coefficient in the past of the frequency conversion coefficient related to the coefficient data is stored, and the search A coefficient restoration step for restoring the frequency conversion coefficient related to the coefficient data based on the frequency conversion coefficient obtained as a result and the predicted value;
An adding step of adding the restored frequency transform coefficient to the prediction source data;
A frequency inverse transform step for restoring a digital audio signal is performed by performing frequency inverse transform on the frequency transform coefficient group obtained by restoring frequency transform coefficients related to all coefficient data. And

  According to the present invention, encoding efficiency can be improved in audio compression.

It is a block diagram which shows the structure of the audio processing apparatus which concerns on Embodiment 1 of this invention. It is a block diagram which shows the structure of the encoding part shown in FIG. It is a figure for demonstrating the search of a frequency set. 3 is a flowchart illustrating a procedure of encoding processing according to the first embodiment. It is a block diagram which shows the structure of the decoding part shown in FIG. 6 is a flowchart illustrating a procedure of decoding processing according to the first embodiment. It is a block diagram which shows the structure of the encoding part which concerns on Embodiment 2 of this invention. 10 is a flowchart illustrating a procedure of encoding processing according to the second embodiment. It is a block diagram which shows the structure of the decoding part which concerns on Embodiment 2 of this invention. 12 is a flowchart illustrating a procedure of decoding processing according to the second embodiment.

  A speech encoding apparatus and speech decoding apparatus according to embodiments of the present invention will be described below with reference to the drawings. In each of the following embodiments, a case where a speech encoding device and a speech decoding device are applied to a speech processing device will be described.

(Embodiment 1)
As shown in FIG. 1, the voice processing device 1 according to the present embodiment includes a voice input / output device 11, a storage device 12, a ROM 13, a RAM 14, and a CPU 15. These units are connected to each other by a system bus 18. The system bus 18 is a transmission path for transferring commands and data.

  The voice input / output device 11 converts the input voice into a digital signal. The audio input / output device 11 generates a digital audio signal by, for example, sampling the input audio at a sampling frequency of 16 kHz and quantizing the input audio by 16 bits. Further, when a digital audio signal is supplied, the audio input / output device 11 outputs audio corresponding to the digital audio signal.

  The storage device 12 includes, for example, a readable / writable nonvolatile semiconductor memory, a hard disk drive, and the like. The storage device 12 stores a digital audio signal generated by the audio input / output device 11, encoded data (encoded data), and the like, as well as various data necessary for audio processing.

  A ROM (Read Only Memory) 13 stores programs and data necessary for audio processing. A RAM (Random Access Memory) 14 is temporarily used as a work area when the CPU 15 executes sound processing. For example, the RAM 14 stores prediction source data (details will be described later) during audio processing.

  The CPU 15 controls each part of the sound processing apparatus 1 and executes each process described later based on a predetermined program stored in the ROM 13. Functionally, the CPU 15 includes an encoding unit 16 and a decoding unit 17. These are functions realized by the CPU 15 executing a program.

  The encoding unit 16 encodes the digital audio signal converted by the audio input / output device 11. As shown in FIG. 2, the encoding unit 16 includes a DC removing unit 161, a frequency converting unit 162, a dividing unit 163, a coefficient predicting unit 164, an entropy encoding unit 165, and an encoded data generating unit 166. .

  The DC removal unit 161 removes a direct current component from the digital audio signal generated by the audio input / output device 11. The reason why the direct current component is removed is that the direct current component is independent of the sound quality. The DC removal unit 161 can be realized by a high-pass filter, for example. The following equation (1) is an example of the transfer function H (z) of the high-pass filter.

  The DC removal unit 161 supplies the digital audio signal from which the direct current component has been removed through the high-pass filter to the frequency conversion unit 162. The frequency conversion unit 162 converts the frequency of the supplied digital audio signal by a predetermined frequency conversion method, and calculates a frequency conversion coefficient group. In the present embodiment, the frequency conversion unit 162 performs frequency conversion by using MDCT (Modified Discrete Cosine Transform). The frequency conversion unit 162 supplies the calculated frequency conversion coefficient group (here, MDCT coefficient group) to the division unit 163.

  The dividing unit 163 divides the frequency conversion coefficient group supplied from the frequency conversion unit 162 for each predetermined number. Hereinafter, this group of divided frequency conversion coefficients is referred to as a frequency set. Each frequency set obtained by the dividing unit 163 is supplied to the coefficient predicting unit 164. It should be noted that the number of frequency conversion coefficients included in the frequency set is smaller as the frequency range is lower (lower frequency band) than the predetermined number of divisions, for example, in accordance with human auditory characteristics. The frequency band may be divided so as to increase.

  The coefficient prediction unit 164 searches the prediction source data 141 stored in the RAM 14 for each frequency set, and finds the frequency set having the highest correlation (in other words, the most similar) with the frequency set to be processed. (See FIG. 3). The prediction source data 141 is data composed of processed frequency sets, that is, past frequency sets (predicted frequency sets) that have been processed by the coefficient prediction unit 164.

The number of frequency conversion coefficients constituting the frequency set is n, the frequency set to be processed and the predicted frequency set of the prediction source data 141 are F ₁ = {f _1i | i = 0 ... n−1} and F _{2, respectively.} = | case of the _{{f 2i i = 0 ... n} -1}, the coefficient prediction unit _164, the F 1, the distance _{F 2 (d} (F _{1, F} 2)) the following equation (2) Ask. The frequency set to be processed is sequentially extracted from the low frequency side of the frequency conversion coefficient group.

The coefficient predicting unit 164 acquires F ₂ having the smallest d (F ₁ , F ₂ ) from the prediction source data 141 as the most similar predicted frequency set. The coefficient prediction unit 164 obtains a predicted value for deriving a frequency conversion coefficient corresponding to the frequency set to be processed from each frequency conversion coefficient of the acquired predicted frequency set. Specifically, the predicted value p when the frequency conversion coefficient of the frequency set to be processed is C _c and the corresponding frequency conversion coefficient of the acquired predicted frequency set is C _p is calculated by the following equation (3).

  In the above formula (3), the smaller absolute value is used as the denominator. This is to prevent noise generation at the time when the audio content is switched.

  The coefficient prediction unit 164 obtains the prediction value obtained as described above, the position information in the prediction source data 141 of the frequency conversion coefficient corresponding to the prediction frequency set, and the absolute value of the frequency conversion coefficient of the frequency set to be processed. The entropy encoding unit 165 is supplied with a flag (large or small flag) indicating which of the absolute values of the frequency conversion coefficient of the predicted frequency set is larger.

  Also, the coefficient prediction unit 164 adds the processed frequency set to the prediction source data 141 stored in the RAM 14 as a predicted frequency set. In general, digital audio signals before and after the time when the audio content is switched have a low correlation. In such a case, it can be said that the frequency conversion coefficients related to the digital audio signal to be processed are more highly correlated. Therefore, the accuracy of the predicted value (in other words, the search accuracy of the closer frequency set) can be improved by sequentially adding the frequency sets for which processing has been completed to the prediction source data 141.

  The entropy encoding unit 165 entropy-encodes the coefficient data including the prediction value, the position information, and the magnitude flag supplied from the coefficient prediction unit 164 to generate encoded coefficient data. Here, since the predicted value is expected to be biased near 0, an improvement in coding efficiency can be expected.

  When the encoded coefficient data is generated for all the frequency conversion coefficients of the frequency conversion coefficient group, the encoded data generation unit 166 generates encoded data in which all the encoded coefficient data are aggregated. The encoded data generation unit 166 stores the generated encoded data in the storage device 12 as part of the audio compression data.

  FIG. 4 is a flowchart showing the procedure of the encoding process executed by the encoding unit 16 described above. This encoding process is started when an operation for starting the audio compression process is performed by the user via an operation input unit (not shown). The operation input unit includes input devices such as a keyboard, a keypad, a touch pad, and a mouse, receives an operation input from the user, and sends the received signal to the CPU 15.

  The DC removal unit 161 acquires the digital audio signal generated by the audio input / output device 11 from the storage device 12 (step S101). The DC removal unit 161 removes a direct current component from the acquired digital audio signal (step S <b> 102) and supplies it to the frequency conversion unit 162. The frequency conversion unit 162 performs frequency conversion (here, MDCT conversion) on the digital audio signal from which the DC component has been removed (step S103), and calculates a frequency conversion coefficient group.

  The dividing unit 163 divides the frequency conversion coefficient group calculated by the frequency conversion unit 162 in units of frequency sets (step S104). The coefficient prediction unit 164 searches the prediction source data 141 for a prediction frequency set that is most similar to the frequency set to be processed that is extracted in order from the low frequency side of the frequency transform coefficient group (step S105). The coefficient predicting unit 164 calculates a predicted value for each frequency conversion coefficient of the frequency set to be processed based on the corresponding predicted frequency set (step S106), and calculates the predicted value, the position information, the magnitude flag, Are supplied to the entropy encoding unit 165.

  Also, the coefficient prediction unit 164 adds the processed frequency set to the prediction source data 141 as a predicted frequency set (step S107). The entropy encoding unit 165 entropy encodes the coefficient data including the prediction value, the position information, and the magnitude flag supplied from the coefficient prediction unit 164 (step S108), and generates encoded coefficient data. In this way, the process for one frequency set is completed. If the processing has not been completed for all frequency sets (NO in step S109), the coefficient predicting unit 164 extracts a frequency set to be processed next from the frequency conversion coefficient group, and performs the process of step S105. Subsequent processing is as described above.

  On the other hand, when the processing is completed for all the frequency sets (YES in step S109), the encoded data generation unit 166 generates encoded data in which all the encoded coefficient data is aggregated (step S110), and the generated code The stored data is stored in the storage device 12 as a part of the audio compression data related to the audio compression processing.

  The CPU 15 determines whether or not an operation (end operation) for ending the audio compression process has been performed by the user (step S111). When the end operation has not been performed (NO in step S111), the DC removal unit 161 acquires the next digital audio signal generated by the audio input / output device 11 from the storage device 12 (step S101). Subsequent processing is as described above. On the other hand, if an end operation has been performed (YES in step S111), the process (encoding process) and the audio compression process are ended. At that time, the prediction source data 141 stored in the RAM 14 is deleted.

  Next, the decoding unit 17 will be described. The decoding unit 17 restores the digital audio signal from the encoded data generated by the encoding unit 16. As illustrated in FIG. 5, the decoding unit 17 includes an entropy decoding unit 171, a coefficient restoration unit 172, and a frequency inverse transform unit 173.

  The entropy decoding unit 171 reads the encoded data from the audio compression data stored in the storage device 12 and performs entropy decoding to restore each coefficient data (including the prediction value, the position information, and the size flag).

  The coefficient restoration unit 172 extracts the corresponding frequency conversion coefficient from the prediction source data 142 stored in the RAM 14 based on the position information for each restored coefficient data. The prediction source data 142 is data composed of frequency conversion coefficients that have already been restored by the coefficient restoration unit 172.

  The coefficient restoration unit 172 restores the frequency conversion coefficient corresponding to the coefficient data based on the predicted value, the magnitude flag, and the frequency conversion coefficient extracted from the prediction source data 142. The coefficient restoration unit 172 supplies the restored frequency transform coefficient to the frequency inverse transform unit 173 and adds it to the prediction source data 142 stored in the RAM 14.

  When the restoration of all the frequency transform coefficients corresponding to the encoded data is completed, the frequency inverse transform unit 173 performs frequency inverse transform (here, IMDCT (Inverse Modified Describe Cosine Transform). : Inversely modified discrete cosine transform))) to restore the digital audio signal. The frequency inverse transform unit 173 stores the restored digital audio signal in the storage device 12.

  FIG. 6 is a flowchart showing the procedure of the decoding process executed by the decoding unit 17 described above. This decoding process is started when the user performs an operation to start the process of expanding the audio compression data stored in the storage device 12 (audio expansion process) via an operation input unit (not shown). The The entropy decoding unit 171 acquires encoded data from the compressed audio data stored in the storage device 12 (step S201), and performs entropy decoding (step S202).

  The coefficient restoration unit 172 performs a process of restoring the corresponding frequency conversion coefficient for each restored coefficient data (predicted value, position information, and magnitude flag). First, the coefficient restoration unit 172 extracts a corresponding frequency conversion coefficient from the prediction source data 142 based on the restored position information (step S203).

  Then, the coefficient restoration unit 172 restores the frequency conversion coefficient corresponding to the coefficient data based on the restored predicted value and magnitude flag and the frequency conversion coefficient extracted from the prediction source data 142 (step S204). . The coefficient restoration unit 172 supplies the restored frequency transform coefficient to the frequency inverse transform unit 173 and adds it to the prediction source data 142 (step S205).

  In this way, when the process for one coefficient data is completed, the coefficient restoration unit 172 completes the process for all restored coefficient data (that is, all the frequency transform coefficients corresponding to the encoded data). It is determined whether or not the restoration has been completed (step S206). When the restoration of all the frequency conversion coefficients has not been completed (NO in step S206), the coefficient restoration unit 172 performs a process of restoring the corresponding frequency conversion coefficient for the next coefficient data as described above (steps S203 to S203). S205).

  On the other hand, when the restoration of all the frequency transform coefficients is completed (YES in step S206), the frequency inverse transform unit 173 performs frequency inverse transform (here, IMDCT) on the obtained frequency transform coefficient group. The digital audio signal is restored (step S207). The frequency inverse transform unit 173 stores the restored digital audio signal in the storage device 12.

  The CPU 15 determines whether or not the user has performed an operation (end operation) for ending the audio decompression process (step S208). When the end operation has not been performed (NO in step S208), the entropy decoding unit 171 acquires the next encoded data from the audio compression data stored in the storage device 12 (step S201). Subsequent processing is as described above. On the other hand, if an end operation has been performed (YES in step S208), the present process (decoding process) and the audio decompression process are ended. At that time, the prediction source data 142 stored in the RAM 14 is deleted.

  As described above, according to the speech processing apparatus 1 according to this embodiment of the present invention, it is possible to improve the coding efficiency by predicting the frequency conversion coefficient during speech compression, and to reduce the bit rate. Audio compression at a rate can be realized.

(Embodiment 2)
Next, Embodiment 2 of the present invention will be described. In the present embodiment, the configurations of the encoding unit 16 and the decoding unit 17 are different from the configurations of the encoding unit 16 and the decoding unit 17 according to the first embodiment.

  As illustrated in FIG. 7, the encoding unit 16 of this embodiment includes a DC removal unit 1601, a frequency conversion unit 1602, a logarithmic conversion unit 1603, a division unit 1604, a coefficient prediction unit 1605, and a quantization unit 1606. An entropy encoding unit 1607, a frequency set restoration unit 1608, and an encoded data generation unit 1609.

  The DC removal unit 1601 removes a direct current component from the digital audio signal generated by the audio input / output device 11 by performing the calculation according to the above equation (1), as with the DC removal unit 161 of the first embodiment. Similar to the frequency conversion unit 162 of the first embodiment, the frequency conversion unit 1602 performs frequency conversion on the digital audio signal from which the DC component has been removed using MDCT, and calculates a frequency conversion coefficient group (here, MDCT coefficient group). .

The logarithmic conversion unit 1603 performs logarithmic conversion on the calculated frequency conversion coefficient group to calculate a logarithmic conversion coefficient group. Specifically, the logarithmic conversion unit 1603 calculates the logarithmic conversion coefficient group (C) from the frequency conversion coefficient group (C = {c _i | i = 0... M−1}) based on the following equation (4). ^L = {c ^L _i | i = 0... M−1}).

但し、ｉ＝０...Ｍ−１

However, i = 0 ... M-1

In addition, the logarithmic conversion unit 1603 acquires the code of C (S = {s _i | i = 0... M−1}) (see Expression (5)).

但し、ｉ＝０...Ｍ−１

However, i = 0 ... M-1

  The dividing unit 1604 divides the logarithmic conversion coefficient group into a predetermined number. Hereinafter, this collection of logarithmic transformation coefficients is called a frequency set.

  The coefficient prediction unit 1605 searches the prediction source data 143 stored in the RAM 14 for each frequency set to be processed, and finds a frequency set that is most similar to the frequency set to be processed. The prediction source data 143 is data composed of processed past frequency sets (predicted frequency sets).

The number of logarithmic transformation coefficients constituting the frequency set is n, the frequency set to be processed and the predicted frequency set of the prediction source data 143 are F ^L ₁ = {f ^L _1i | i = 0 ... n−1} and When F ^L ₂ = {f ^L _2i | i = 0... N−1}, the coefficient predicting unit 1605 calculates the distance between F ^L ₁ and F ^L ₂ (d (F ^L ₁ , F ^L ₂ )). Is obtained by the following equation (6). The frequency set to be processed is sequentially extracted from the low frequency side.

The coefficient predicting unit 1605 obtains, from the prediction source data 143, the prediction frequency set that is the most similar to F ^L ₂ that minimizes d (F ^L ₁ , F ^L ₂ ). The coefficient prediction unit 1605 obtains a prediction value for deriving a logarithmic transformation coefficient corresponding to the frequency set to be processed from each logarithmic transformation coefficient of the prediction frequency set of the acquired prediction source data 143. Specifically, the predicted value p when the log transformation coefficient of the frequency set to be processed is c ^L _c and the corresponding log transformation coefficient of the acquired predicted frequency set is c ^L _p is calculated by the following equation (7). To do.

  The quantization unit 1606 quantizes the prediction value calculated by the coefficient prediction unit 1605. The entropy encoding unit 1607, the quantized predicted value, the code of the frequency transform coefficient corresponding to the logarithmic transform coefficient of the frequency set to be processed, the position information in the prediction source data 143 of the logarithmic transform coefficient corresponding to the predicted frequency set Are entropy encoded to generate encoded coefficient data.

  The frequency set restoration unit 1608 inversely quantizes the quantized predicted value to return it to the predicted value, and further restores the logarithmic transformation coefficient from the predicted value using the position information. The frequency set restoration unit 1608 adds the restored logarithmic transformation coefficient in units of frequency sets as a prediction frequency set to the prediction source data 143 stored in the RAM 14. By sequentially adding the frequency sets that have been processed in this way to the prediction source data 143, the accuracy of the prediction value can be improved as in the encoding unit 16 of the first embodiment.

  When the encoded coefficient data is generated for all logarithmic transform coefficients of the logarithmic transform coefficient group, the encoded data generation unit 1609 generates encoded data in which all the encoded coefficient data are aggregated. The encoded data generation unit 1609 stores the generated encoded data in the storage device 12 as part of the audio compression data.

  FIG. 8 is a flowchart showing the procedure of the encoding process executed by the encoder 16 of the present embodiment described above. This encoding process is started when an operation for starting the audio compression process is performed by the user via an operation input unit (not shown). The operation input unit includes input devices such as a keyboard, a keypad, a touch pad, and a mouse, receives an operation input from the user, and sends the received signal to the CPU 15.

  The DC removal unit 1601 acquires the digital audio signal generated by the audio input / output device 11 from the storage device 12 (step S301). The DC removal unit 1601 removes a direct current component from the acquired digital audio signal (step S302), and supplies it to the frequency conversion unit 1602. The frequency conversion unit 1602 performs frequency conversion (here, MDCT conversion) on the digital audio signal from which the DC component has been removed to calculate a frequency conversion coefficient group (step S303).

  The logarithmic conversion unit 1603 performs logarithmic conversion on the calculated frequency conversion coefficient group (step S304), and calculates the logarithmic conversion coefficient group. In addition, the logarithmic conversion unit 1603 acquires the sign of each frequency conversion coefficient.

  The dividing unit 1604 divides the logarithmic transformation coefficient group calculated by the logarithmic transformation unit 1603 in units of frequency sets (step S305). The coefficient prediction unit 1605 searches for the most similar predicted frequency set from the prediction source data 143 stored in the RAM 14 for each frequency set (step S306). The coefficient predicting unit 1605 calculates a predicted value for each logarithmic transformation coefficient of the frequency set to be processed based on the corresponding predicted frequency set (step S307).

  The quantization unit 1606 quantizes the prediction value calculated by the coefficient prediction unit 1605 (step S308). The entropy encoding unit 1607 performs entropy encoding on the quantized predicted value, the code of the corresponding frequency transform coefficient, and position information (step S309), and generates encoded coefficient data. The frequency set restoration unit 1608 inversely quantizes the quantized predicted value (step S310), returns it to the predicted value, and further restores the logarithmic transformation coefficient from the predicted value using the position information. Then, the restored logarithmic transformation coefficient is added as a predicted frequency set to the prediction source data 143 in units of frequency sets (step S311).

  In this way, the process for one frequency set is completed. If the processing has not been completed for all frequency sets (NO in step S312), the coefficient predicting unit 1605 extracts the next frequency set to be processed from the logarithmic transform coefficient group, and performs the process of step S306. Subsequent processing is as described above.

  On the other hand, when the processing is completed for all frequency sets (YES in step S312), the encoded data generation unit 1609 generates encoded data in which all the encoded coefficient data is aggregated (step S313), and the generated code The stored data is stored in the storage device 12 as a part of the audio compression data related to the audio compression processing.

  The CPU 15 determines whether or not an operation (end operation) for ending the audio compression process has been performed by the user (step S314). If the end operation has not been performed (NO in step S314), the DC removal unit 1601 acquires the next digital audio signal generated by the audio input / output device 11 from the storage device 12 (step S301). Subsequent processing is as described above. On the other hand, if an end operation has been performed (YES in step S314), the process (encoding process) and the audio compression process are ended. At this time, the prediction source data 143 stored in the RAM 14 is deleted.

  Next, the decoding unit 17 of this embodiment will be described. As illustrated in FIG. 9, the decoding unit 17 includes an entropy decoding unit 1701, an inverse quantization unit 1702, a coefficient restoration unit 1703, a linear conversion unit 1704, and a frequency inverse conversion unit 1705.

  The entropy decoding unit 1701 reads the encoded data from the compressed audio data stored in the storage device 12 and performs entropy decoding, and each coefficient data (quantized prediction value, code of the corresponding frequency transform coefficient and position information) Is included). The inverse quantization unit 1702 restores the predicted value by inversely quantizing the quantized predicted value for each restored coefficient data.

  The coefficient restoration unit 1703 extracts the corresponding logarithmic transformation coefficient from the prediction source data 144 stored in the RAM 14 based on the position information for each restored coefficient data. The prediction source data 144 is data composed of logarithmic transformation coefficients that have already been restored by the coefficient restoration unit 1703.

  The coefficient restoration unit 1703 restores a logarithmic transformation coefficient corresponding to the coefficient data based on the predicted value and the logarithmic transformation coefficient extracted from the prediction source data 144. The coefficient restoration unit 1703 supplies the restored logarithmic transformation coefficient and the code of the corresponding frequency transformation coefficient to the linear transformation unit 1704 and adds the restored logarithmic transformation coefficient to the prediction source data 144 stored in the RAM 14.

  The linear transformation unit 1704 restores the frequency transformation coefficient by performing transformation from the logarithmic axis to the linear axis with respect to the logarithmic transformation coefficient supplied from the coefficient restoration unit 1703.

  When the restoration of all the frequency transform coefficients corresponding to the encoded data is completed, the frequency inverse transform unit 1705 performs frequency inverse transform (here, IMDCT) on the obtained frequency transform coefficient group, and performs digital conversion. Restore the audio signal. The frequency inverse transform unit 1705 stores the restored digital audio signal in the storage device 12.

  FIG. 10 is a flowchart showing the procedure of the decoding process executed by the decoding unit 17 of this embodiment described above. This decoding process is started when the user performs an operation to start the process of expanding the audio compression data stored in the storage device 12 (audio expansion process) via an operation input unit (not shown). The The entropy decoding unit 1701 acquires encoded data from the audio compression data stored in the storage device 12 (step S401), and performs entropy decoding (step S402). As a result, each coefficient data (including the quantized predicted value, the code of the corresponding frequency transform coefficient, and position information) is restored. The inverse quantization unit 1702, the coefficient restoration unit 1703, and the linear conversion unit 1704 perform the following processing for each restored coefficient data.

  The inverse quantization unit 1702 performs inverse quantization on the quantized predicted value (step S403) to restore the predicted value. The coefficient restoration unit 1703 extracts a corresponding logarithmic transformation coefficient from the prediction source data 144 based on the restored position information (step S404). Then, the coefficient restoration unit 1703 restores a logarithmic transformation coefficient corresponding to the coefficient data based on the predicted value and the logarithmic transformation coefficient extracted from the prediction source data 144 (step S405). The coefficient restoration unit 1703 supplies the restored logarithmic transformation coefficient and the code of the corresponding frequency transformation coefficient to the linear transformation unit 1704, and adds the restored logarithmic transformation coefficient to the prediction source data 144 (step S406).

  The linear transformation unit 1704 restores the frequency transformation coefficient by performing transformation from the logarithmic axis to the linear axis with respect to the logarithmic transformation coefficient supplied from the coefficient restoration unit 1703 (step S407).

  When the process for one coefficient data is completed in this way, the inverse quantization unit 1702 has completed the process for all the restored coefficient data (that is, all the frequency transform coefficients corresponding to the encoded data). It is determined whether or not the restoration has been completed (step S408). When the restoration of all the frequency transform coefficients has not been completed (NO in step S408), the inverse quantization unit 1702, the coefficient restoration unit 1703, and the linear transformation unit 1704 perform the same process as described above for the next coefficient data ( Steps S403 to S407).

  On the other hand, when the restoration of all the frequency transform coefficients is completed (YES in step S408), the frequency inverse transform unit 1705 performs frequency inverse transform (here, IMDCT) on the obtained frequency transform coefficient group. Then, the digital audio signal is restored (step S409). The frequency inverse transform unit 1705 stores the restored digital audio signal in the storage device 12.

  The CPU 15 determines whether or not the user has performed an operation (end operation) for ending the audio decompression process (step S410). If the end operation has not been performed (NO in step S410), the entropy decoding unit 1701 acquires the next encoded data from the audio compression data stored in the storage device 12 (step S401). Subsequent processing is as described above. On the other hand, if an end operation has been performed (YES in step S410), the present process (decoding process) and the audio decompression process are ended. At that time, the prediction source data 144 stored in the RAM 14 is deleted.

  As described above, according to the speech processing apparatus 1 according to the present embodiment of the present invention, the frequency conversion coefficient is predicted at the time of speech compression in the same manner as the speech processing apparatus 1 of the first embodiment. Therefore, it is possible to realize audio compression at a lower bit rate than in the prior art.

  The present invention is not limited to the above-described embodiments, and various modifications can be made without departing from the scope of the present invention.

  For example, in each of the above embodiments, MDCT is used as the frequency conversion. However, the frequency conversion is not limited to MDCT, and DCT (Descrete Cosine Transform) or DFT (Discrete Fourier Transform). It may be.

  In each of the above embodiments, the functions of the encoding unit 16 and the decoding unit 17 described above are realized by logical processing by the CPU 15 executing a program. For example, ASIC (Application Specific Integrated) The above-described functions may be realized by hardware such as Circuit).

  Furthermore, an existing personal computer (PC) or the like can be caused to function as a speech encoding device or speech decoding device according to the present invention. That is, the program as executed by the CPU 15 described above is applied to an existing PC or the like, and the CPU or the like of the PC or the like executes the program so that the PC or the like can be used as the speech encoding apparatus or the speech according to the present invention. It becomes possible to function as a decoding device.

  Such a program distribution method is arbitrary, for example, a computer-readable recording medium such as a CD-ROM (Compact Disk Read-Only Memory), a DVD (Digital Versatile Disk), an MO (Magneto Optical Disk), or a memory card. It may be stored and distributed in a network, or distributed via a communication network such as the Internet.

  In this case, when the functions according to the present invention described above are realized by sharing an OS (operating system) and an application program, or by cooperation between the OS and the application program, only the application program portion is stored in a recording medium or the like. May be.

  DESCRIPTION OF SYMBOLS 1 ... Voice processing apparatus, 11 ... Voice input / output device, 12 ... Memory | storage device, 13 ... ROM, 14 ... RAM, 15 ... CPU, 16 ... Encoding part, 17 ... Decoding part, 18 ... System bus, 141-144 ... Prediction source data, 161, 1601 ... DC removing unit, 162, 1602 ... frequency converting unit, 163,1604 ... dividing unit, 164,1605 ... coefficient predicting unit, 165,1607 ... entropy encoding unit, 166,1609 ... encoding Data generation unit, 171, 1701, entropy decoding unit, 172, coefficient restoration unit, 173, 1705, frequency inverse transformation unit, 1603, logarithmic transformation unit, 1606, quantization unit, 1608, frequency set restoration unit, 1702, inverse quantum Conversion unit, 1703 ... coefficient restoration unit, 1704 ... linear conversion unit

Claims (9)

Frequency conversion means for frequency-converting a digital audio signal and calculating a frequency conversion coefficient group;
A dividing unit that divides the calculated frequency conversion coefficient group into a frequency set for each predetermined number;
For each frequency set, a predicted frequency set having the highest correlation is searched from prediction source data in which a frequency set in the past of the frequency set is stored as a predicted frequency set, and each frequency transform coefficient of the frequency set is derived. Predicting means for calculating the predicted value based on the corresponding frequency transform coefficient of the predicted frequency set obtained as a result of the search;
Encoding means for encoding coefficient data including at least the prediction value calculated by the prediction means and position information of frequency conversion coefficients corresponding to the prediction frequency set, and generating encoded coefficient data;
With respect to all the frequency transform coefficients of the frequency transform coefficient group, when the coding coefficient data is generated, the encoded data generating means for generating encoded data that aggregates all the encoded coefficient data,
When the processing by the prediction unit is completed, the frequency set is sequentially added to the prediction source data as the predicted frequency set.
A speech encoding apparatus characterized by that. The prediction means calculates the predicted value by division using either one of the frequency conversion coefficient of the frequency set and the corresponding frequency conversion coefficient of the predicted frequency set as a denominator and the other as a numerator.
The speech coding apparatus according to claim 1. The coefficient data encoded by the encoding means further includes a flag indicating which one of the absolute value of the frequency transform coefficient of the frequency set and the absolute value of the frequency transform coefficient of the predicted frequency set is larger. Included,
In the division, the prediction means uses the smaller one of the absolute value of the frequency conversion coefficient of the frequency set and the absolute value of the frequency conversion coefficient of the prediction frequency set as a denominator.
The speech encoding apparatus according to claim 2. Frequency conversion means for frequency-converting a digital audio signal and calculating a frequency conversion coefficient group;
Logarithmic conversion means for performing logarithmic conversion on the calculated frequency conversion coefficient group and calculating a logarithmic conversion coefficient group;
A dividing unit that divides the calculated logarithmic transformation coefficient group as a frequency set for each predetermined number;
For each frequency set, a predicted frequency set having the highest correlation is searched from prediction source data in which a frequency set in the past of the frequency set is stored as a predicted frequency set, and each logarithmic transformation coefficient of the frequency set is derived. Predicting means for calculating the predicted value based on the corresponding logarithmic transformation coefficient of the predicted frequency set obtained as a result of the search;
Coefficient data including the prediction value calculated by the prediction unit, position information of a logarithmic transformation coefficient corresponding to the prediction frequency set, and a sign of the frequency transformation coefficient corresponding to the logarithmic transformation coefficient of the frequency set. Encoding means for encoding and generating encoded coefficient data;
With respect to all logarithmic transform coefficients of the logarithmic transform coefficient group, when the coding coefficient data is generated, the encoded data generation means for generating encoded data that aggregates all the encoded coefficient data, and
When the processing by the prediction unit is completed, the frequency set is sequentially added to the prediction source data as the predicted frequency set.
A speech encoding apparatus characterized by that. A frequency conversion step of frequency-converting the digital audio signal and calculating a frequency conversion coefficient group;
A dividing step of dividing the calculated frequency transform coefficient group into a frequency set for each predetermined number;
For each frequency set, a predicted frequency set having the highest correlation is searched from prediction source data in which a frequency set in the past of the frequency set is stored as a predicted frequency set, and each frequency transform coefficient of the frequency set is derived. A predicting step of calculating a predicted value of a based on a corresponding frequency transform coefficient of a predicted frequency set obtained as a result of the search;
An additional step of adding the frequency set for which the processing in the prediction step has been completed to the prediction source data as the predicted frequency set;
An encoding step of encoding coefficient data including at least the calculated predicted value and position information of a corresponding frequency transform coefficient of the predicted frequency set, and generating encoded coefficient data;
For all frequency transform coefficients of the frequency transform coefficient group, when the coding coefficient data is generated, an encoded data generation step of generating encoded data that aggregates all the encoded coefficient data, and
A speech encoding method characterized by the above. On the computer,
A frequency conversion step of frequency-converting the digital audio signal and calculating a frequency conversion coefficient group;
A dividing step of dividing the calculated frequency transform coefficient group into a frequency set for each predetermined number;
For each frequency set, a predicted frequency set having the highest correlation is searched from prediction source data in which a frequency set in the past of the frequency set is stored as a predicted frequency set, and each frequency transform coefficient of the frequency set is derived. A predicting step of calculating a predicted value of a based on a corresponding frequency transform coefficient of a predicted frequency set obtained as a result of the search;
An additional step of adding the frequency set for which the processing in the prediction step has been completed to the prediction source data as the predicted frequency set;
An encoding step of encoding coefficient data including at least the calculated predicted value and position information of a corresponding frequency transform coefficient of the predicted frequency set, and generating encoded coefficient data;
For all frequency transform coefficients of the frequency transform coefficient group, when the coding coefficient data is generated, an encoded data generation step of generating encoded data that aggregates all the encoded coefficient data is executed.
A speech encoding program characterized by the above. For each frequency conversion coefficient of the frequency conversion coefficient group corresponding to the digital audio signal, at least a predicted value for deriving the frequency conversion coefficient from the past frequency conversion coefficient and position information of the past frequency conversion coefficient are included. A speech decoding apparatus for restoring a digital speech signal from encoded data in which encoded coefficient data obtained by encoding coefficient data is stored,
Decoding means for reading out and decoding the encoded data from a storage device and restoring the coefficient data;
For each restored coefficient data, a frequency conversion coefficient corresponding to the position information is searched from prediction source data in which a frequency conversion coefficient in the past of the frequency conversion coefficient related to the coefficient data is stored, and the search Based on the frequency conversion coefficient obtained as a result and the predicted value, coefficient restoration means for restoring the frequency conversion coefficient related to the coefficient data;
Frequency inverse transform means for restoring a digital audio signal by performing frequency inverse transform on the frequency transform coefficient group obtained by restoring the frequency transform coefficients related to all coefficient data, and
The restored frequency transform coefficient is sequentially added to the prediction source data.
A speech decoding apparatus characterized by that. For each frequency conversion coefficient of the frequency conversion coefficient group corresponding to the digital audio signal, at least a predicted value for deriving the frequency conversion coefficient from the past frequency conversion coefficient and position information of the past frequency conversion coefficient are included. An audio decoding method for restoring a digital audio signal from encoded data in which encoded coefficient data obtained by encoding coefficient data is stored,
A decoding step of reading out and decoding the encoded data from a storage device, and restoring each coefficient data;
For each restored coefficient data, a frequency conversion coefficient corresponding to the position information is searched from prediction source data in which a frequency conversion coefficient in the past of the frequency conversion coefficient related to the coefficient data is stored, and the search A coefficient restoration step for restoring the frequency conversion coefficient related to the coefficient data based on the frequency conversion coefficient obtained as a result and the predicted value;
An adding step of adding the restored frequency transform coefficient to the prediction source data;
A frequency inverse transform step for restoring the digital audio signal by performing frequency inverse transform on the frequency transform coefficient group obtained by restoring the frequency transform coefficients related to all coefficient data,
A speech decoding method characterized by the above. On the computer,
For each frequency conversion coefficient of the frequency conversion coefficient group corresponding to the digital audio signal, at least a predicted value for deriving the frequency conversion coefficient from the past frequency conversion coefficient and position information of the past frequency conversion coefficient are included. An audio decoding program for restoring a digital audio signal from encoded data in which encoded coefficient data obtained by encoding coefficient data is stored,
A decoding step of reading out and decoding the encoded data from a storage device, and restoring each coefficient data;
For each restored coefficient data, a frequency conversion coefficient corresponding to the position information is searched from prediction source data in which a frequency conversion coefficient in the past of the frequency conversion coefficient related to the coefficient data is stored, and the search A coefficient restoration step for restoring the frequency conversion coefficient related to the coefficient data based on the frequency conversion coefficient obtained as a result and the predicted value;
An adding step of adding the restored frequency transform coefficient to the prediction source data;
Performing a frequency inverse transform step for restoring a digital audio signal by performing frequency inverse transform on the frequency transform coefficient group obtained by restoring the frequency transform coefficients related to all coefficient data,
A speech decoding program characterized by the above.

Images