JP2008032823A - Voice encoding apparatus - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a voice encoding apparatus, capable of performing sectioning suitable for a voice signal with a small processing amount. <P>SOLUTION: A coefficient which is a maximum value is extracted from among Modified Discrete Cosine Transform (MDCT) coefficients included in each scale factor band. When a ratio of the maximum values of the MDCT coefficients of adjoining scale factor bands is within a predetermined range, those scale factor bands are set in the same section. <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention relates to a speech encoding apparatus, and more particularly to sectioning processing.

  When Huffman coding of an audio signal compressed by an AAC (Advanced Audio Coding) method or the like, the frequency spectrum obtained by frequency conversion of the audio signal, for example, MDCT (Modified Discrete Cosine Transform) is used. A certain MDCT coefficient is quantized, and the obtained quantized spectrum is encoded. In this encoding, a Huffman table with a small generated code amount is selected for each frequency band called a scale factor band (hereinafter referred to as sfb), and encoding is performed with reference to the selected table.

  Then, in addition to the encoded quantized spectrum, information indicating the selected table is included in the bit stream that is the encoded audio signal. A plurality of Huffman tables are prepared according to the maximum value of the quantized spectrum. If the maximum value of the quantized spectrum is smaller, efficient coding is performed by selecting a smaller table.

  In addition, a sectioning process for selecting the same Huffman table with a plurality of adjacent sfb called sections is known. According to the sectioning process, information indicating the selected table can be shared by adjacent sfb, and information indicating the table included in the bit stream can be reduced. By reducing this information, more bits can be allocated to the quantized spectrum, and the sound quality of the encoded speech can be improved.

  It is known that sectioning is performed before spectral quantization or after quantization. In order to perform before quantization, for example, the maximum value of the spectrum obtained by frequency conversion for each sfb is obtained, a predetermined number of sfb having the largest obtained maximum value is selected, and for each selected sfb Choose an appropriate Huffman table. A process is known in which the sfb adjacent to the selected sfb is set to the same section as the selected sfb (see, for example, Patent Document 1).

In order to perform sectioning after spectrum quantization, for example, a Huffman table is selected with reference to a predetermined table for each sfb, and after the selection, the above-described bit stream is used by using a Huffman table having the same sfb. If it is possible to reduce the information indicating the table included in the table, a process is known in which those adjacent sfb are set to the same section (see, for example, Patent Document 2).
JP 2002-91498 A (first page, FIG. 1, FIG. 3) Japanese Patent Laying-Open No. 2003-233397 (first page, FIG. 2)

  However, in the method disclosed in Patent Document 1 described above, since the number of sections is determined without depending on the audio signal, there is a problem that it is difficult to perform sectioning suitable for the audio signal. This problem is significant when the spectrum of the audio signal is concentrated and distributed over sfb in the near frequency band.

  Further, the method disclosed in Patent Document 2 described above has a problem in that a Huffman table suitable for an audio signal may not be selected because a Huffman table is selected with reference to a predetermined table. In addition, when a bit stream that is an encoded audio signal is created and the bit length is not a predetermined value, it is necessary to return to the spectrum quantization stage and repeat the process. Therefore, there is a problem that the processing amount may become excessive.

  The present invention has been made to solve the above problems, and an object of the present invention is to provide a speech coding apparatus that performs sectioning suitable for speech signals with a small amount of processing.

  To achieve the above object, the speech coding apparatus of the present invention converts a speech signal into a frequency spectrum classified into a predetermined scale factor band, and the frequency classified into the scale factor band for each scale factor band. Depending on the representative value of the spectrum, one of the plurality of Huffman tables is selected, the frequency spectrum classified into the scale factor band is encoded with reference to the selected Huffman table, and the code Encoding means for creating an encoded speech signal including a frequency spectrum that has been converted and a code that identifies the referenced Huffman table, wherein the encoding means is classified into the adjacent scale factor bands When the ratio of the representative values of the frequency spectrum is within the specified ratio And selects the same the Huffman table to their scale factor band.

  The speech coding apparatus according to the present invention converts a speech signal into a frequency spectrum classified into a predetermined scale factor band, and depends on a representative value of the frequency spectrum classified into the scale factor band for each scale factor band. And selecting one of the plurality of Huffman tables, encoding the frequency spectrum classified into the scale factor band with reference to the selected Huffman table, and the encoded frequency spectrum Encoding means including a code identifying the referenced Huffman table, the encoding means comprising: a first frequency spectrum classified into the first scale factor band; Low representative value, low frequency of its first scale factor band And a representative value of the frequency spectrum classified into the second scale factor band adjacent to the second scale factor band and a representative of the frequency spectrum classified into the third scale factor band adjacent to the high frequency side of the first scale factor band When the value is larger than the representative value of the first scale factor band, the same Huffman table is selected for the first, second, and third scale factor bands.

  According to the present invention, it is possible to provide a speech coding apparatus that performs sectioning suitable for speech signals with a small amount of processing.

  Embodiments of a speech encoding apparatus according to the present invention will be described below with reference to the drawings.

(First embodiment)
FIG. 1 is a block diagram showing a configuration of a speech encoding apparatus according to the first embodiment of the present invention. The speech coding apparatus includes a control unit 11 that controls the entire apparatus, a speech signal storage unit 21, a time / frequency conversion unit 31, a psychoacoustic analysis unit 32, a scale factor multiplication unit 33, and a sectioning unit 34. And a quantization loop processing unit 41, a formatter unit 51, and an encoded audio signal storage unit 52.

  The quantization loop processing unit 41 includes a quantization unit 42, a Huffman coding unit 43, and a generated code amount counting unit 44.

  The operation of each unit of the speech coding apparatus according to the first embodiment of the present invention configured as described above will be described with reference to FIG.

  The audio signal storage unit 21 stores an audio signal converted into a digital signal by a PCM (Pulse Code Modulation) method.

  The time / frequency conversion unit 31 reads the audio signal stored in the audio signal storage unit 21, performs time / frequency conversion, creates a frequency spectrum, and transmits the frequency spectrum. The MDCT method is used for time / frequency conversion. Then, an MDCT coefficient is created as a frequency spectrum. The time / frequency conversion is not limited to the MDCT method.

  The psychoacoustic analysis unit 32 stores a psychoacoustic model including a masking effect characteristic. Then, the psychoacoustic analysis unit 32 receives the MDCT coefficient created by the time / frequency conversion unit 31, and calculates the allowable quantization distortion amount to which the masking effect is applied by the stored psychoacoustic model of the received MDCT coefficient. Calculate and transmit for each sfb. Also, rate control information is calculated and transmitted.

  Note that the psychoacoustic analysis unit 32 reads the audio signal stored in the audio signal storage unit 21 and performs time / frequency conversion using a method different from the time / frequency conversion unit 31, for example, FFT (Fast). (Fourier Transform) conversion may be performed, and a frequency spectrum obtained by the conversion may be used. According to the method using different conversions, an increase in processing amount occurs due to two time / frequency conversions performed by the time / frequency conversion unit 31 and the psychoacoustic analysis unit 32, which are appropriate for these processing units. Transformations can be selected independently.

  The scale factor multiplication unit 33 receives the allowable quantization distortion amount for each sfb calculated by the psychoacoustic analysis unit 32, and calculates the scale factor for each sfb from the distortion amount. Then, the MDCT coefficient created by the time / frequency conversion unit 31 is received, the MDCT coefficient is multiplied by the scale factor calculated for each sfb, and the product MDCT coefficient is transmitted.

  The sectioning unit 34 receives the MDCT coefficient created by the scale factor multiplication unit 33, and sets sectioning, that is, two or more adjacent sfb as the same section. Then, the received MDCT coefficient and information indicating the determined sectioning are transmitted. When a certain sfb is silent, that is, when there is no MDCT coefficient in the sfb, the sectioning unit 34 transmits that fact. The sfb is not subject to sectioning. This is because the sfb is not an object to be encoded by the Huffman table.

  The quantization loop processing unit 41 repeats the operation of the quantization unit 42, the operation of the Huffman encoding unit 43, and the operation of the generated code amount counting unit 44. Then, the information of the Huffman table selected by the Huffman encoding unit 43, the code encoded using the Huffman table, and the rate control information transmitted from the psychoacoustic analysis unit 32 are stored.

  The quantization unit 42 non-uniformly quantizes the MDCT coefficients transmitted by the sectioning unit 34 in units of sfb, and transmits the quantized MDCT coefficients. That is, the quantization step, which is a divisor when dividing the MDCT coefficient, depends on sfb. If a certain sfb is silent, the quantization unit 42 does not set the sfb as a quantization target.

  The Huffman encoder 43 stores a Huffman table. Then, the Huffman encoder 43 receives the quantized MDCT coefficient transmitted by the quantizer 42 and selects an appropriate Huffman table for each section or sfb created by the sectioning unit 34. And the code is stored in the quantization loop processing unit 41 and transmitted. Further, for sfb that is silent, encoding is performed without using a Huffman table. The Huffman encoding unit 43 stores the used Huffman table in the quantization loop processing unit 41.

  The generated code amount counting unit 44 measures the code amount of the code encoded by the Huffman encoding unit 43, that is, the number of bits, and stores it in the quantization loop processing unit 41 while accumulating the number of bits for each sfb. Let At this time, the number of bits of the code for identifying the Huffman table selected by the Huffman encoder 43 stored in the quantization loop processor 41 by the Huffman encoder 43 is also measured and stored.

  Furthermore, the generated code amount counting unit 44 stores the rate control information transmitted from the psychoacoustic analysis unit 32 in the quantization loop processing unit 41, and the scale factor used for each sfb identified by the information. The number of bits of the indicated code is also measured and stored. This is because the bits for identifying these are included in a bit stream that is an encoded audio signal.

  The formatter unit 51 stores information for identifying the Huffman table selected by the Huffman encoding unit 43, the code encoded using the Huffman table, and the psychoacoustic analysis stored in the quantization loop processing unit 41 The information for identifying the rate control information transmitted from the unit 32 is read, a bit stream that is an encoded audio signal is created in a predetermined format, and stored in the encoded audio signal storage unit 52.

  Details of the sectioning operation performed by the sectioning unit 34 of the speech encoding apparatus according to this embodiment will be described below.

  The sectioning unit 34 receives the MDCT coefficient transmitted by the scale factor multiplication unit 33. An example of the received MDCT coefficient is shown in FIG. Here, the horizontal axis is the frequency, and the vertical axis is the amplitude value, that is, the magnitude of the MDCT coefficient. FIG. 2 also shows that the MDCT coefficients are divided into sfb. The amplitude of the MDCT coefficient, which is the maximum amplitude for each sfb, is indicated by a bold line. In FIG. 2, the frequency band width of each sfb is constant, and a fixed number of MDCT coefficients are included for each sfb. However, these are examples, and this embodiment It does not add any limitation to.

  The sectioning unit 34 performs sectioning by the first method and the second method. First, the first method will be described. The sectioning unit 34 extracts the coefficient that is the maximum value from the MDCT coefficients included in each sfb. FIG. 3 shows a state in which the maximum MDCT coefficient is extracted from each sfb. Here, the horizontal axis is sfb, and the vertical axis is the maximum amplitude value of the MDCT coefficient for each sfb.

  Here, the maximum value of the MDCT coefficient extracted by each sfb is assumed to be max_quant [sfb]. When the ratio of the maximum values of the MDCT coefficients of adjacent sfb is within a predetermined range, the sectioning unit 34 determines that the sfb is the same section.

  Whether or not the ratio of the maximum values is within a predetermined range depends on whether or not the following inequality 1 holds, for example.

SEC_TH1 <(max_quant [sfb + 1] / max_quant [sfb]) <SEC_TH2 (inequality 1)
Note that 0 <SEC_TH1 <1 and 1 <SEC_TH2.

  Here, the product obtained by multiplying SEC_TH1 and SEC_TH2 may be 1. For example, SEC_TH1 is 1/2 and SEC_TH2 is 2. Based on these values, it is possible to determine whether or not the ratio of the maximum values of the MDCT coefficients of the adjacent sfb is within twice.

  Thus, by performing sectioning according to whether the ratio of the maximum values of the MDCT coefficients of sfb adjacent to each other is within a predetermined range, the sectioning unit 34 has sfb0 to sfb2 as illustrated in FIG. It is determined that sfb3 and sfb4 are one section, and sfb5 to sfb7 are one section.

  Further, the inequality 1 becomes difficult to hold even if the maximum values of the MDCT coefficients of adjacent sfb are the same as SEC_TH1 is increased and SEC_TH2 is decreased. That is, it is difficult to determine the same section, in other words, it is easy to determine that the Huffman table is selected independently for each sfb.

  In addition, when it is allowed that the number of bits of the bit stream that is the encoded audio signal generated by the apparatus is large, it is appropriate for any sfb by selecting the Huffman table almost independently for each sfb. A Huffman table may be selected. For this selection, it is effective to make inequality 1 difficult to hold as described above.

  When the inequality 1 holds, according to the processing of the same section, the number of sections changes depending on the audio signal stored in the audio signal storage unit 21 and is not constant. That is, excellent sectioning depending on the audio signal is performed by easy determination as to whether inequality 1 holds.

  In the case of the second method, the sectioning unit 34 sets the sfb having a relatively small maximum value of the MDCT coefficient as the same section as the sfb having a relatively large maximum value of the MDCT coefficient adjacent to the sfb. That is, sfb having a relatively small maximum value of the MDCT coefficient is added to a section using a relatively large Huffman table suitable for sfb having a relatively large maximum value of the MDCT coefficient.

  That is, in FIG. 5, since the maximum value of the MDCT coefficient of a certain sfb (sfb1 in FIG. 5) is relatively small, a relatively small Huffman table is selected, and the sfb on both sides of the sfb (sfb0 in FIG. 5). And sfb2)) because the maximum value of the MDCT coefficient is relatively large, an example of a situation where a relatively large Huffman table is selected is shown. Here, sfb2 to sfb4 are made into one section by the sectioning unit 34 according to the first method.

  In this situation, when the relatively large maximum value is equal to or greater than a predetermined threshold, the sectioning unit 34, as shown in FIG. 6, sets the sfb and sfb adjacent to the sfb (sfb0 to sfb4 in FIG. 6). Make one section. In this section, a Huffman table suitable for the sfb on both sides is selected.

  In this way, the MDCT coefficients included in the sfb (sfb1 in FIG. 6) are encoded by an excessive Huffman table as compared with the maximum values of these coefficients. However, it is possible to reduce two sets of information for identifying the Huffman table from the bit stream that is the number of sections reduced by two, that is, an encoded audio signal, and the bit length of the bit stream can be reduced. It becomes possible.

  Here, especially when the ratio of the maximum values of the MDCT coefficients of the adjacent sfb is substantially equal and the Huffman tables selected for the sfb are expected to be the same, the above effect is particularly remarkable. However, it is not limited to the case where the Huffman tables selected for them are expected to be the same. When these Huffman tables are different, the sectioning unit 34 increases the amount of encoded code generated by the newly selected Huffman table and the code amount obtained by reducing two sets of information for identifying the Huffman table. Whether or not to perform sectioning is determined based on the amount of reduction or the expected value of these amounts.

  The first method and / or the second method are not limited to being performed only once. That is, if the number of sections cannot be sufficiently reduced by performing sectioning once, the sectioning unit 34 can reduce the number of sections by repeatedly performing the sectioning operation. It goes without saying that this repetition may be performed any number of times.

  Here, in the first method, the sectioning unit 34 sets a value smaller than the value used for the i-th time in SEC_TH1, and sets a value larger than the value used for the i-th time in SEC_TH2. i + 1 and subsequent sectioning. Further, in the second method, the threshold is set to a value smaller than the value used for the i-th time, and the (i + 1) -th and subsequent sectioning are performed. Here, i is an integer of 1 or more.

  In the above description, the sectioning unit 34 extracts the coefficient having the maximum value from the MDCT coefficients included in each sfb. However, the present invention is not limited to this. Any value representative of the MDCT coefficient may be used. For example, an average value of the MDCT coefficients included in each sfb may be calculated, and the calculated average value may be used instead of the maximum value.

  When the average value is used, when many MDCT coefficients are included in sfb and there is no large difference in the values of these coefficients, the Huffman table is selected without strongly depending on the value maximized by a slight difference. Become. By this selection, an encoded audio signal with excellent sound quality is created, and the bit length of the encoded audio signal bit stream can be reduced.

  Next, the audio signal encoding control operation by the control unit 11 will be described. FIG. 7 is a flowchart showing the control operation of the audio signal encoding by the control unit 11.

  The control unit 11 starts the control operation of the audio signal encoding (step S11a), reads a predetermined amount of the audio signal stored in the audio signal storage unit 21, controls each unit of the audio encoding device, and performs the following Let the action take place.

  That is, the control unit 11 controls the time / frequency conversion unit 31 to calculate the MDCT coefficient from the frame (step S11b). Then, the psychoacoustic analysis unit 32 is controlled to calculate the allowable quantization distortion amount and rate control information from the MDCT coefficient, and the scale factor multiplication unit 33 is controlled to calculate the scale factor from the allowable quantization distortion amount. Is calculated and multiplied by the MDCT coefficient (step S11c). Then, the sectioning unit 34 is controlled to perform sectioning from the multiplied MDCT coefficient (step S11d).

  Next, the control unit 11 controls the quantization loop processing unit 41 to cause a quantization loop. That is, the quantization unit 42 is controlled to quantize the MDCT coefficient multiplied by the scale factor (step S11e). Then, the Huffman encoding unit 43 is controlled to select a Huffman table according to the sectioning result, and the quantized MDCT coefficients are Huffman encoded. Then, the code, the code amount, and the selected Huffman table are stored (step S11f).

  Then, the control unit 11 controls the generated code amount counting unit 44 to measure all generated code amounts, and determines whether or not the code amount is within a predetermined bit length (step S11g). If it is within the predetermined bit length, the formatter unit 51 is controlled to arrange the encoded audio signal into a predetermined bit stream format and stored in the encoded audio signal storage unit 52 (step S11h). ), The operation is terminated (step S11i).

  On the other hand, when it is not within the predetermined bit length, the control unit 11 controls the quantization unit 42 in step S11e to return to the step of performing the quantization to operate. At this time, quantization is performed by a larger quantization step.

(Second Embodiment)
FIG. 8 is a block diagram showing a configuration of a speech encoding apparatus according to the second embodiment of the present invention. In the speech encoding apparatus according to the second embodiment, the same parts as those in the speech encoding apparatus according to the first embodiment are denoted by the same reference numerals and description thereof is omitted. The speech coding apparatus according to the second embodiment includes a control unit 11 that controls the entire apparatus, a speech signal storage unit 21, a time / frequency conversion unit 31, a psychoacoustic analysis unit 32, and a scale factor multiplication. Unit 33, quantization loop processing unit 41, formatter unit 51, and encoded speech signal storage unit 52.

  The quantization loop processing unit 41 includes a quantization unit 42, a sectioning unit 34, a Huffman coding unit 43, and a generated code amount counting unit 44.

  The differences between the speech encoding apparatus according to the second embodiment and the speech encoding apparatus according to the first embodiment are as follows. That is, the sectioning unit 34 performs sectioning based on the MDCT coefficient created by the scale factor multiplication unit 33 in the first embodiment, whereas the quantization loop processing unit in the second embodiment. 41, and performing sectioning based on the MDCT coefficients quantized by the quantization unit 42.

  Accordingly, in the second embodiment, the MDCT coefficient created by the scale factor multiplication unit 33 is sent to the quantization unit 42, and the MDCT coefficient transmitted by the sectioning unit 34 and the information indicating sectioning are: The Huffman encoding unit 43 receives it. Further, when a certain sfb is silent, the scale factor multiplication unit 33 transmits that fact to the sectioning unit 34.

  Next, the control operation of speech signal encoding by the control unit 11 of the speech encoding apparatus according to the second embodiment will be described. FIG. 9 is a flowchart showing the control operation of the audio signal encoding by the control unit 11. In the control operation according to the second embodiment, the same operation steps as those of the control operation according to the first embodiment are denoted by the same reference numerals and description thereof is omitted.

  The differences between the control operation according to the second embodiment and the control operation according to the first embodiment are as follows. That is, in the first embodiment, the control operation for controlling the sectioning unit 34 in step S11d to perform sectioning includes the determination of the scale factor in step S11c, the multiplication operation thereof, and the quantization operation in step S11e. On the other hand, in the second embodiment, it is placed between the quantization operation in step S11e and the Huffman coding and generation code storage operations in step S11f.

  Accordingly, the operation for performing the sectioning in step S11d is determined not to be within the predetermined bit length by the operation for determining whether or not all the generated code amounts in step S11g are within the predetermined bit length. The case will be repeated. Thus, when repeatedly performed, the control unit 11 controls the quantization unit 42 to perform quantization by a larger quantization step, or alternatively, as described above, the sectioning unit 34. May be controlled to further reduce the number of sections.

(Other embodiments)
In the above embodiment, the input audio signal is stored in the audio signal storage unit 21. This may store all audio signals to be encoded. For example, an analog signal input by a microphone (not shown) may be converted into a digital signal by the PCM method and stored in the audio signal storage unit 21 and an encoding operation may be performed in parallel. good.

  Also, the encoded audio signal is stored in the encoded audio signal storage unit 52. In this case, all encoded audio signals may be stored. In addition, an encoding operation and, for example, an operation in which an encoded signal is transmitted via a communication line may be performed in parallel. Needless to say, when this operation is performed in parallel, the permissible size of the bit stream that is the encoded audio signal corresponds to the speed of the bit rate of the transmission.

  The speech encoding apparatus according to the embodiment of the present invention may be a computer that operates using a program. In addition, the present invention can naturally be applied to any device that encodes an audio signal. Moreover, you may combine suitably the element demonstrated in said embodiment. The present invention is not limited to the above configuration, and various modifications are possible.

1 is a block diagram showing a configuration of a speech encoding apparatus according to a first embodiment of the present invention. The figure which shows an example of the MDCT coefficient which multiplied the scale factor which concerns on embodiment of this invention. The figure which shows an example of the maximum value of the MDCT coefficient which multiplied the scale factor for every sfb which concerns on embodiment of this invention. The figure which shows an example of the sectioning result which concerns on embodiment of this invention (the 1). FIG. 6 is a diagram (part 2) illustrating an example of a sectioning result according to the embodiment of the present invention. The figure which shows an example of the sectioning result which concerns on embodiment of this invention (the 3). The flowchart which shows the operation | movement which controls the audio | voice coding of the control part which concerns on the 1st Embodiment of this invention. The block diagram which shows the structure of the audio | voice coding apparatus which concerns on the 2nd Embodiment of this invention. The flowchart which shows the operation | movement which controls the audio | voice coding of the control part which concerns on the 2nd Embodiment of this invention.

Explanation of symbols

DESCRIPTION OF SYMBOLS 11 Control part 31 Time / frequency conversion part 32 Psychological auditory analysis part 33 Scale factor multiplication part 34 Sectioning part 41 Quantization loop process part 42 Quantization part 43 Huffman encoding part 44 Generated code amount count part 51 Formatter part

Claims (5)

The audio signal is converted into a frequency spectrum classified into a predetermined scale factor band, and one of a plurality of Huffman tables depending on the representative value of the frequency spectrum classified into the scale factor band for each scale factor band. A Huffman table is selected, the frequency spectrum classified into the scale factor band is encoded with reference to the selected Huffman table, and the encoded frequency spectrum and a code for identifying the referenced Huffman table are encoded. Encoding means for creating an encoded audio signal including:
When the ratio of the representative values of the frequency spectrum classified into the adjacent scale factor bands is within a predetermined ratio, the encoding means selects the same Huffman table for those scale factor bands. A speech encoding device. The audio signal is converted into a frequency spectrum classified into a predetermined scale factor band, and one of a plurality of Huffman tables depending on the representative value of the frequency spectrum classified into the scale factor band for each scale factor band. A Huffman table is selected, the frequency spectrum classified into the scale factor band is encoded with reference to the selected Huffman table, and the encoded frequency spectrum and a code for identifying the referenced Huffman table are encoded. Encoding means for creating an encoded audio signal including:
The encoding means has a small representative value of the frequency spectrum classified into the first scale factor band, and is classified into the second scale factor band adjacent to the low frequency side of the first scale factor band. The representative value of the frequency spectrum and the representative value of the frequency spectrum classified into the third scale factor band adjacent to the high frequency side of the first scale factor band are the representative values of the first scale factor band. If larger, the same Huffman table is selected for the first, second, and third scale factor bands. The encoding means converts the speech signal into a frequency spectrum quantized by a quantization step depending on the scale factor band when the speech signal is converted into a frequency spectrum classified into the predetermined scale factor band. The speech encoding apparatus according to claim 1 or 2. The encoding means converts the audio signal into a frequency spectrum multiplied by a scale factor depending on the scale factor band when the audio signal is converted into a frequency spectrum classified into the predetermined scale factor band. The speech encoding apparatus according to claim 1 or 2. The speech code according to claim 1 or 2, wherein a representative value of the frequency spectrum classified into the scale factor band is a maximum value or an average value of the frequency spectrum classified into the scale factor band. Device.

Images