JP4470122B2 - Speech coding apparatus, speech decoding apparatus, speech coding program, and speech decoding program - Google Patents

Description

  The present invention relates to coding and decoding of speech waveforms using a predictive coding scheme.

  As a method for coding / decoding a speech waveform, a predictive coding method represented by ADPCM (Adaptive Differential Pulse Code Modulation) is known. Predictive coding encodes a difference (predicted value) from a value converted into data immediately before based on the feature that a speech waveform changes continuously. PCM (Pulse-Code Modulation) : Pulse code modulation) has an advantage that the compression rate is high. However, since this encoding method involves a prediction error, the prediction error is superimposed as the decoding process progresses, and a deviation occurs from the original waveform. As a result, when a plurality of voices are continuously played back, the PCM value at the end point of the previous voice waveform is matched with the PCM value at the start point of the subsequent voice waveform. There is a discrepancy (gap) due to the prediction error between the points. As a result, there arises a problem that noise occurs at the joint between the front and rear voices.

  Patent Document 1 discloses a playback device that prevents a sound waveform from becoming discontinuous without muting. Specifically, two uncompressed speech waveforms are used, and crossfade processing is performed at the connection points of these speech waveforms, thereby preventing sudden fluctuations in values at the connection points. However, when the crossfade process is performed, there is a problem that the reproduction time is shortened by the amount of the crossfade process.

JP 2001-128119 A

  The present invention has been made in view of such circumstances, and an object of the present invention is to effectively suppress the generation of noise at a joint between voices when a plurality of voices are continuously played back.

  In order to solve such a problem, the first invention provides a speech encoding apparatus that includes a dividing unit and an encoder and encodes a series of speech waveforms. The dividing unit divides the speech waveform into a plurality of groups in time series. The encoder performs an audio waveform encoding process using each of the plurality of groups as a processing unit. At this time, predictive coding is performed in the normal order in time series on the speech waveforms in the start group including the start point of the speech waveform. Also, predictive coding is performed in reverse order in time series on the speech waveforms in the end group including the end point of the speech waveform.

  Here, in the first invention, the encoder preferably includes an inverter and a prediction code encoder. The inverter inverts the speech waveform in the end group so that the speech waveform in the end group input in the normal order in time series is in reverse order in the time series. The predictive code encoder inputs to itself the time-series forward start group input without going through the inverter and the time-series reverse end group inputted through the inverter. Predictive encoding is performed according to the order. The encoder may further include a transform code encoder that performs transform coding on the speech waveform in the intermediate group excluding the start group and the end group.

  In the first invention, the predictive code encoder preferably adds identification information indicating a time-series coding order within a group to a codeword of a speech waveform. For example, the prediction code encoder may add identification information for specifying a group corresponding to the start group and identification information for specifying a group corresponding to the end group to the codeword of the speech waveform. Further, for example, identification information for specifying a group corresponding to the intermediate group may be added to the codeword of the speech waveform.

  In the first invention, the dividing unit preferably divides the speech waveform so that the speech waveforms partially overlap between groups adjacent to each other in time series.

  2nd invention has a determination part, a decoder, and a synthetic | combination part, The audio | voice decoding apparatus which decodes the codeword produced | generated by encoding a series of audio | voice waveform by a time-sequential group unit I will provide a. Based on the identification information added to the codeword, the determination unit determines whether or not the group of codewords to be decoded is a start group including the start point of the speech waveform, and the end of the speech waveform It is determined whether or not the end group includes a point. The decoder performs codeword decoding processing in units of groups, and for the codewords in the group determined to be the start group, performs decoding by the predictive coding method in the normal order in time series, For the code words in the group determined to be the end group, decoding is performed by the predictive coding method in reverse order in time series. The synthesis unit restores a series of speech waveforms by synthesizing a plurality of decoded waveforms generated by the decoder in time series.

  Here, in the second invention, the decoder generates a decoded waveform in units of groups by performing decoding using a predictive coding method for the codewords in each group according to the input order to itself. It is preferable to include a predictive code decoder that performs this operation and an inverter that inverts the decoded waveforms in the group so that the decoded waveforms in the group determined to be the end group are reversed in time series.

  In the second invention, the determination unit may determine whether or not a group of codewords to be subjected to the decoding process is an intermediate group excluding a start group and an end group. In this case, it is desirable that the decoder further includes a transform code decoder that performs decoding by a transform coding scheme for codewords in the group determined to be an intermediate group.

  In the second invention, the determination unit determines whether the decoded waveform obtained by decoding the code word overlaps with another decoded waveform in time series based on the identification information added to the code word. Also good. In this case, when it is determined that the decoded waveform is in a range where the decoded waveform overlaps with another decoded waveform adjacent to each other in time series, the synthesis unit performs a fade process on the decoded waveform. It is preferable to carry out.

  The third invention provides a speech encoding program that causes a computer to execute a speech encoding method that encodes a series of speech waveforms using a group as a processing unit. This program performs a first step of dividing a speech waveform into a plurality of groups in time series, a speech waveform encoding process in which each of the plurality of groups is a processing unit, and a start including a start point of the speech waveform Predictive coding is applied to the speech waveforms in the group in chronological order, and predictive codes are applied to the speech waveforms in the end group including the end point of the speech waveform in reverse order in chronological order. And causing the computer to execute a speech encoding method including a second step of performing the conversion.

The fourth invention provides a speech decoding program for causing a computer to execute a speech decoding method for decoding a codeword generated by encoding a series of speech waveforms in time-series group units. Based on the identification information added to the codeword, this program determines whether or not the group of codewords to be decoded is a start group including the start point of the speech waveform, and the end of the speech waveform The first step of determining whether or not the end group includes a point, and the codeword decoding process is performed in units of groups. For codewords in the group determined to be the start group, First, decoding using the predictive coding method is performed in the forward sequence, and the codeword in the group determined to be the end group is decoded using the predictive coding method in the reverse sequence in the time series. The computer is caused to execute a speech decoding method including the second step and a third step of restoring a series of speech waveforms by synthesizing the decoded waveforms decoded for each group in time series.

  According to the present invention, the start group is encoded by the predictive encoding method in the time-series order from the start point of the speech waveform, and thus the start point of the speech waveform obtained by decoding the start group is the same as that before encoding. This coincides with the end point of the voice waveform. On the other hand, for the end group, encoding by predictive encoding is performed in an order reverse to the time-series order from the end point of the speech waveform. This coincides with the end point of the voice waveform. That is, since the start point and end point of the restored speech waveform match the speech waveform before encoding, noise due to the gap between the start point and the end point between speech waveforms can be suppressed.

(First embodiment)
<Encoding>
FIG. 1 is a configuration diagram of a speech encoding apparatus according to this embodiment. This speech encoding apparatus includes a storage unit 1, a dividing unit 2, and an encoder 3, and encodes a series of speech waveforms (original waveforms). The storage unit 1 stores a speech waveform to be encoded. The speech waveform is composed of a plurality of sample points, and position information (address) is associated with each of the plurality of sample points in time series order.

  The dividing unit 2 reads the speech waveform stored in the storage unit 1 and divides the speech waveform into a plurality of groups in time series. In this embodiment, the number of divisions is two, and the speech waveform is divided into two groups: a start group including the start point and an end group including the end point. Further, the dividing unit 2 divides the speech waveform so that the speech waveforms partially overlap between groups adjacent to each other in time series. Specifically, the start group is set so that the end thereof includes a predetermined time (margin) after the dividing point. Alternatively, or in combination with this, the end group may be set so that the tip thereof includes a predetermined time (margin) before the dividing point. As a result, the start group and the end group overlap by several samples corresponding to the margin.

  The encoder 3 uses each group defined by the dividing unit 2 as a processing unit, and performs encoding processing on the speech waveform in the group. At this time, predictive coding is performed in the normal order in time series on the start group including the start point of the speech waveform. Also, for the end group including the end point of the speech waveform, predictive coding is performed in reverse order in time series. As an example of the predictive encoding method, ADPCM (Adaptive Differential Pulse Code Modulation) is used.

  The encoder 3 includes an inverter 3a and a prediction code encoder 3b. The inverter 3a outputs the end groups in the reverse order in time series by inverting the speech waveforms of the end groups input in the normal order in time series. As a result, the end of the end group corresponding to the end point of the series of speech waveforms is shifted to the end of the end group. Specifically, this inversion process is performed by reversing the address corresponding to each sample point of the speech waveform in the end group on the time axis. The predictive code encoder 3b applies the time series forward start group input without going through the inverter 3a and the time series reverse end group inputted through the inverter 3a to these Are subjected to predictive coding in accordance with the input order (i.e., normal order in time series), and a speech waveform code word is output.

  Note that, for the later-described decoding, the predictive code encoder 3b adds identification information indicating the type of group, specifically, time-series coding order within the group, to the code word of the speech waveform as a group flag. To do. By this group flag, a group corresponding to the start group, a group corresponding to the end group, and a group corresponding to the intermediate group are identified and specified.

  Furthermore, the predictive code encoder 3b adds, as a fade flag, identification information indicating whether or not the encoding target group corresponds to a portion overlapping with another group adjacent in time series. In the present embodiment, when a sample point belonging to a range overlapping with the end group in the start group is encoded, this is added to the corresponding code word. On the contrary, when the sample point in the range which overlaps with the start group among the end groups is encoded, the fact is added to the corresponding code word. When a non-overlapping sample point is encoded, this is added to the corresponding code word.

<Decryption>
FIG. 2 is a configuration diagram of the speech decoding apparatus according to the present embodiment. This speech decoding apparatus decodes the codeword encoded by the speech encoding apparatus shown in FIG. 1 into a speech waveform. The speech decoding apparatus includes a determination unit 4, a decoder 5, a synthesis unit 6, and a storage unit 7.

  The determination unit 4 determines, based on the identification information attached to the code word, to which group of the original waveform the decoded waveform (more precisely, the sample value) obtained by decoding the code word corresponds. To do. Specifically, the group type (start group, end group, intermediate group) is determined based on the group flag added to the codeword. Further, the above-described overlap portion is also determined based on the fade flag attached to the code word. The determination result by the determination unit 4 is output to a decoder 5 (inverter 5b) and a synthesis unit 6 (multiplier 6a) described later.

  The decoder 5 includes a prediction code decoder 5a and an inverter 5b. The predictive code decoder 5a decodes the input codeword in the time series in the normal order according to the predictive encoding method, and generates a decoded waveform. At this time, in the encoding shown in FIG. 1, since all groups are encoded by the predictive encoding method in the normal order in time series, the predictive code decoder 5a is in the normal order in time series. Decoding can be performed regardless of whether or not there is.

  The inverter 5b inverts the decoded waveform generated from the predictive code decoder 5 in time series according to the determination of the determination unit 4. When it is determined that the code word is obtained by encoding the start group, the decoded waveform generated by the predictive code decoder 5 for this code word is output as it is (forward encoding). On the other hand, when it is determined that the code word is obtained by encoding the end group, the decoded waveform generated by the predictive code decoder 5a of the code word is inverted and output by the inverter 5b in time series ( Reverse encoding).

  The synthesizer 6 restores a series of speech waveforms by synthesizing the decoded waveforms for each group generated by the decoder 5 in time series. At the time of synthesis, according to the determination of the determination unit 4, the decoded waveform output from the inverter 5b is subjected to crossfade processing. FIG. 3 is an explanatory diagram of the crossfade processing according to the present embodiment. Crossfade processing is a method of mixing two audio waveforms. For example, within a range where adjacent groups overlap in time series, the previous group fades out and the subsequent group fades in and plays. To do. The crossfade process of the present embodiment is performed in a range where the decoding groups corresponding to the start group and the end group overlap each other.

  In the encoding shown in FIG. 1, the start group and the end group are divided into a time series with some margin. Therefore, by connecting the groups using the cross-fade process, the reproduction time of the decoded speech waveform can be matched with that of the speech waveform before encoding. Further, noise generated when two groups continuously reproduce can be suppressed by the crossfade process. The synthesizer 6 includes a multiplier 6a and an adder 6b.

  The multiplier 6a multiplies a part of the decoded waveform generated from the inverter 5b by a predetermined value as necessary according to the instruction of the determination unit 4. For example, when the determination unit 4 determines that a part of the decoded waveform in the previous stage overlaps with the decoded waveform in the subsequent stage that is adjacent in time series, the multiplier 6a determines the sampling points belonging to the overlapping range. The value is multiplied by a predetermined value and output. On the other hand, when the determination unit 4 determines that a part of the decoded waveform in the subsequent stage overlaps with the decoded waveform in the preceding stage that is adjacent in time series, the multiplier 6a determines the decoded waveform belonging to the overlapping range. The sample point value is multiplied by another predetermined value and output. When it is determined by the determination unit 4 that the waveform does not overlap any decoded waveform, it is output as it is without being multiplied by a predetermined value.

  The adder 6b adds the decoded waveform output from the multiplier 6a and the decoded waveform stored in the storage unit 7 while making them correspond in time series. The storage unit 7 stores the decoded waveform output in the previous process through a series of decoding processes (decoder 5 and synthesis unit 6). The adder 6b adds the decoded waveform corresponding to the decoded waveform output from the multiplier 6a in time series and the decoded waveform output from the multiplier 7a, and stores the result in the storage unit 7 again.

  According to this embodiment, when a plurality of voices are continuously played back, it is possible to effectively suppress the occurrence of noise due to the prediction error at the voice joint. Since predictive coding is performed on the speech waveforms in the start group in chronological order, no prediction error occurs at the start point of the speech waveform. At the same time, since the predictive coding is performed on the speech waveforms in the end group in reverse order in time series, no prediction error occurs at the end point of the speech waveform. Thereby, when a plurality of voices are continuously played back, it is possible to eliminate a mismatch (gap) between the end point of the previous voice waveform and the start point of the subsequent voice waveform.

  For example, even if the above-described encoding / decoding is performed using two consecutive speech waveforms (previous waveform and subsequent waveform) as original waveforms, a gap between the two waveforms does not occur. It is possible to play audio without any problem. Even if the subsequent waveform is selected in real time and randomly during playback of the previous waveform, if the end point of the previous waveform matches the start point of the selected subsequent waveform, crossfade processing is performed. Since it becomes unnecessary, it becomes unnecessary to prepare (decode) the entire subsequent waveform in advance.

  In the present embodiment, ADPCM is employed as predictive coding. However, the present invention only needs to be able to make the value of the start point (end point) in the original waveform constant in coding / decoding. Therefore, for example, DPCM (Differential Pulse Code Modulation) may be used.

  In this embodiment, whether or not the decoder 5 (inverter 5b) and the synthesizer 6 (multiplier 6a) can perform processing is determined based on identification information (group flag and fade flag) added to the codeword separately. However, the present invention only needs to be able to determine at the time of decoding processing whether the decoded waveform is a group in the original waveform and whether it overlaps with a group adjacent in time series. Therefore, for example, the position information (address) corresponding to each group may be changed by the inverter 3a, and the determination unit 4 may determine the type of group and the presence or absence of overlap based on this position information.

  In this embodiment, there are two groups (start group and end group) to be encoded, but the present invention is not limited to this. According to the present invention, even when a speech waveform is divided into three or more groups, encoding / decoding by a predictive encoding method can be performed on a start group including a start point and an end group including an end point. It's enough. The case of dividing into three or more groups will be described in the next embodiment.

(Second Embodiment)
<Encoding>
FIG. 4 is a configuration diagram of the speech encoding apparatus according to the present embodiment. This speech encoding apparatus encodes a series of speech waveforms as in the first embodiment, but differs in that in addition to a start group and an end group, an intermediate group is also encoded. The intermediate group is a group excluding a start group and an end group among a plurality of groups, that is, a group that includes neither the start point nor the end point of the speech waveform to be encoded. Structurally, this embodiment differs from the first embodiment in that it further includes a transform code encoder 3c.

  The transform code encoder 3c performs transform coding on the intermediate group extracted by the dividing unit 2 and outputs a code word. Transform coding is a technique for performing compression coding after data is converted to an axis with less redundancy based on orthogonal transform. The transform coding in the present embodiment is, for example, MP3 (MPEG-1 Audio Layer-3) or AAC (Advanced Audio Coding) using MDCT (Modified Discrete Cosine Transform). For decoding described later, the transform code encoder 3c corresponds to the generated codeword corresponding to a group flag indicating an intermediate group and a portion overlapping with another group adjacent in time series. A fade flag indicating whether or not is added.

<Decryption>
FIG. 5 is a configuration diagram of the speech decoding apparatus according to the present embodiment. This speech decoding apparatus decodes the codeword encoded by the speech encoding apparatus shown in FIG. 4 into an original waveform. The speech decoding apparatus according to this embodiment is different from that according to the first embodiment in that it includes a selector 8 and a transform code decoder 5c. Accordingly, the functions of the determination unit 4 and the synthesis unit 6 are slightly changed. Other configurations and functions are the same as those in the first embodiment, and thus description thereof is omitted.

  The determination unit 4 determines the group flag and the fade flag attached to the code word as in the first embodiment. When the determination unit 4 determines that the encoding is performed by the prediction code encoder 3b, the determination unit 4 causes the selector 8 to input the code word to the prediction code decoder 5a. When the determination unit 4 determines that the encoding is performed by the conversion code encoder 3c, the determination unit 4 causes the selector 8 to input the code word to the conversion code decoder 5c. The transform code decoder 5c decodes the codeword input from the selector 8 using transform coding to generate a decoded waveform.

  The synthesizer 6 performs a crossfade process on the decoded waveform output from the decoder 5 in accordance with an instruction from the determination unit 4. FIG. 6 is an explanatory diagram of the crossfade processing according to the present embodiment. As in the first embodiment, the crossfade process in the present embodiment is performed in a range where the decoded waveform overlaps with other decoded waveforms in time.

  According to the present embodiment, as in the first embodiment, when a plurality of voices are continuously played back, it is possible to effectively suppress the occurrence of noise due to the prediction error at the voice joint. In the first embodiment, the entire original waveform is encoded / decoded by the predictive encoding method. On the other hand, in the present embodiment, only the start group and the end group are encoded / decoded in the same format, and the remaining part (intermediate group) is encoded / decoded by transform encoding. Decryption is in progress. Therefore, this embodiment can prevent deterioration of quality more effectively than the first embodiment.

  In the present embodiment, the transform coding scheme is used as the intermediate group coding format, but the present invention is not limited to this, and for example, a predictive coding scheme may be used. Further, even when the speech waveform is divided into four or more and there are two or more intermediate groups, the same effect as that of the first embodiment can be obtained by performing the encoding by the above encoding method on the intermediate group. Can be played.

  The first and second embodiments have been described on the hardware circuit, but the present invention is not limited to this. For example, the present invention can be realized by recording a program that realizes the above functions on a computer-readable storage medium and causing the computer system to execute the program recorded on the storage medium.

1 is a configuration diagram of a speech encoding apparatus according to a first embodiment. 1 is a configuration diagram of a speech decoding apparatus according to a first embodiment. Explanatory drawing of crossfade processing in a 1st embodiment The block diagram of the audio | voice coding apparatus concerning 2nd Embodiment. The block diagram of the audio | voice decoding apparatus concerning 2nd Embodiment. Explanatory drawing of the cross-fade process concerning 2nd Embodiment

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Memory | storage part 2 Dividing part 3 Encoder 3a Inverter 3b Prediction code encoder 3c Conversion code encoder 4 Judgment part 5 Decoder 5a Prediction code decoder 5b Inverter 5c Conversion code decoder 6 Synthesis | combination part 6a Multiplier 6b Adder 7 Storage part 8 Selector

Claims (13)

In a speech encoding apparatus that encodes a series of speech waveforms,
A division unit for dividing the speech waveform into a plurality of groups in time series;
A speech waveform encoding process is performed with each of the plurality of groups as a processing unit, and the speech waveforms in the start group including the start point of the speech waveform are predictively encoded in a chronological order. And an encoder that performs predictive coding in reverse order in time series on the speech waveform in the end group including the end point of the speech waveform. The encoder is
An inverter that inverts the speech waveform in the end group so that the speech waveform in the end group input in time sequence in normal order is in reverse order in time series;
The input order to the self with respect to the start group in the forward order in time series input without passing through the inverter and the end group in the reverse order in time series input through the inverter The speech encoding apparatus according to claim 1, further comprising: a predictive code encoder that performs predictive encoding according to the above. The encoder is
The speech coding apparatus according to claim 1, further comprising a transform code encoder that performs transform coding on the speech waveform in the intermediate group excluding the start group and the end group. The prediction code encoder is
The speech coding apparatus according to claim 1, wherein identification information indicating a time-series coding order within a group is added to a codeword of the speech waveform. The prediction code encoder is
The identification information for specifying a group corresponding to the start group and the identification information for specifying a group corresponding to the end group are added to a codeword of the speech waveform. The speech encoding apparatus described in 1. The transform code encoder is
4. The speech encoding apparatus according to claim 3, wherein identification information for specifying a group corresponding to the intermediate group is added to a codeword of the speech waveform.   The said division part divides | segments the said audio | voice waveform so that the said audio | voice waveform may overlap partially between the groups which mutually adjoined in time series, The description in any one of Claim 1 to 6 characterized by the above-mentioned. Speech encoding device. In a speech decoding apparatus that decodes a codeword generated by encoding a series of speech waveforms in time-series group units,
Based on the identification information added to the codeword, whether or not a group of codewords to be decoded is a start group including a start point of the speech waveform, and an end point of the speech waveform A determination unit for determining whether or not the end group includes
The codeword decoding process is performed in units of groups, and the codewords in the group determined to be the start group are decoded by the predictive coding method in the normal order in time series, and the end For a codeword in a group determined to be a group, a decoder that performs decoding by a predictive coding method in reverse order in time series,
A speech decoding apparatus comprising: a synthesizing unit that restores a series of speech waveforms by synthesizing the decoded waveforms for each group generated by the decoder in time series. The decoder
A predictive code decoder that generates a decoded waveform in units of groups by performing decoding by a predictive coding method for the codewords in each group according to an input order to the self;
9. The inverter according to claim 8, further comprising: an inverter that inverts the decoded waveform in the group so that the decoded waveform in the group determined to be the end group is in reverse order in time series. Voice decoding device. The determination unit determines whether or not a group of codewords subject to decoding processing is an intermediate group excluding the start group and the end group,
The decoder
The speech decoding according to claim 8 or 9, further comprising a transform code decoder that performs decoding by a transform coding scheme for a codeword in the group determined to be the intermediate group. Device. The determination unit
Based on the identification information added to the codeword, determine whether the decoded waveform obtained by decoding the codeword overlaps with other decoded waveforms in time series,
The synthesis unit is
When it is determined that the decoded decoded waveform is in a range that overlaps with other decoded waveforms that are adjacent to each other in time series, a fade process is performed on the decoded waveform. The speech decoding apparatus according to any one of claims 8 to 10. In a speech encoding program for causing a computer to execute a speech encoding method for encoding a series of speech waveforms using a group as a processing unit,
A first step of dividing the speech waveform into a plurality of groups in time series;
A speech waveform encoding process is performed with each of the plurality of groups as a processing unit, and the speech waveforms in the start group including the start point of the speech waveform are predictively encoded in a chronological order. And the computer executes a speech encoding method including a second step of predictively encoding the speech waveforms in the end group including the end point of the speech waveform in a time-series reverse order. A speech encoding program characterized by being caused to execute. In a speech decoding program for causing a computer to execute a speech decoding method for decoding a codeword generated by encoding a series of speech waveforms in time-series group units,
Based on the identification information added to the codeword, whether or not a group of codewords to be decoded is a start group including a start point of the speech waveform, and an end point of the speech waveform A first step of determining whether or not the end group includes:
The codeword decoding process is performed in units of groups, and the codewords in the group determined to be the start group are decoded by the predictive coding method in the normal order in time series, and the end A second step of performing decoding by a predictive coding method in reverse order in time series for codewords in the group determined to be a group;
A speech decoding program that causes a computer to execute a speech decoding method having a third step of restoring a series of speech waveforms by synthesizing the decoded waveforms decoded for each group in time series .

Images