JP2016018042A - Voice decryption device, voice decryption method, voice decryption program, and communication apparatus - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a voice decryption device configured to improve sound quality of decryption voice of the code method of a MEB (multi-band excitation) system, a voice decryption method, a voice decryption program, and a communication device.SOLUTION: A voice decryption device 1A decrypts digital code information coded according to a MBE system voice decryption method, comprising: MBE system decryption means 12 configured to decrypt digital code information to generate decryption voice; plosive detection means 13 configured to detect plosive of the decryption voice; and rupture processing means 14 configured to rupture the detected plosive.SELECTED DRAWING: Figure 1

Description

  The present invention relates to a speech decoding apparatus, a speech decoding method, a speech decoding program, and a communication device. For example, the present invention decodes an encoded speech signal using a MBE (Multi-Band Excitation) -based speech coding scheme. It is suitable for application in some cases.

  With the revision of the Radio Law due to concerns about increased demand for data transmission and frequency constraints, it has been decided that the simple wireless device will be completely transferred from the conventional analog system to the digital system. In response to this trend, a standard for a communication method of a digital simple wireless device (hereinafter referred to as a digital wireless device) was established by the Japan Radio Industry Association. In contrast to the modulation method 4-level FSK widely used in specific low-power radios, the broadcasting business uses 4FSK communication radio system (STD-B54), and the communication field uses narrowband digital communication system (SCPC / 4). Value FSK system) (STD-T102), and all voice coding systems are "Recommends Digital Voice System, Inc. (USA company) AMBE + 2 Enhanced Half-Rate" . Note that AMBE ++ (sometimes referred to as AMBE ++) is available from Digital Voice System, Inc. Trademark.

  AMBE + 2 has the advantage that the decoded speech is not likely to be unnatural even in a noisy environment, and the advantage that it can provide stable quality even at a low bit rate, but it has the disadvantage of altering the voice color. It has also been reported that it becomes “sound” (Non-patent Document 1).

  AMBE + 2 is a system to which MBE (Multi-Band Excitation), which is one of speech coding systems, is applied, and AMBE is an abbreviation for Advanced MBE. In addition to AMBE, there is a speech encoding method called IMBE (Improved MBE). All of AMBE and IMBE including AMBE + 2 are based on MBE. In the present specification, MBE, AMBE, and IMBE are referred to as “MBE-based speech encoding methods”. It should be noted that simply describing the MBE speech encoding method indicates that the speech encoding method is MBE.

  FIG. 9 shows the configuration of a speech encoding apparatus described in Non-Patent Document 2 that conforms to the MBE encoding scheme.

  In FIG. 9, the speech coding apparatus 100 includes a frequency conversion unit 101, an initial pitch selection unit 102, a pitch improvement unit 103, a voiced envelope estimation unit 104, a voiceless envelope estimation unit 105, a voiced / unvoiced determination unit 106, and a voiced / unvoiced selection. Means 107, multiplexing means 108 and quantization means 109 are provided.

  A speech signal obtained by digitizing a speech signal captured by a microphone or the like by a D / A converter (not shown) (hereinafter referred to as input speech) is input to speech encoding apparatus 100. The frequency conversion means 101 converts the input sound into a frequency spectrum by using a FFT (Fast Fourier Transform) while overlapping the input sound. The initial pitch selection means 102 is based on the criterion of minimizing the harmonic model error when the input speech is assumed to be a complete voiced sound, while using dynamic programming together with the pitch period (integer sample value). And the obtained initial pitch is given to the pitch improving means 103. The pitch improving unit 103 represents the initial pitch represented by the integer sample value based on the input spectrum from the frequency converting unit 101 so that the harmonic model error is further reduced. Update to a highly accurate pitch period.

  The voiced envelope estimation unit 104 calculates envelope information for the voiced sound that minimizes the harmonic model error based on the input spectrum from the frequency conversion unit 101 and the real number pitch from the pitch improvement unit 103. Envelope information for voiced sound is composed of power and phase for each harmonic component. The unvoiced envelope estimation unit 105 calculates the power for each harmonic band based on the input spectrum and the real number pitch and calculates the power for each harmonic band as unvoiced envelope information. The harmonic band is a band occupied by each harmonic component in the voiced sound, and is defined by a real pitch. Adjacent harmonic bands do not overlap or separate from each other. Voiced / unvoiced determining means 106, for each harmonic band defined by the real number pitch, based on the harmonic model error and unvoiced envelope information of the harmonic band calculated from the input spectrum and voiced envelope information. It is determined whether the band is voiced sound or unvoiced sound. Voiced / unvoiced selection means 107 alternatively selects voiced or unvoiced envelope information for each harmonic band based on voiced / unvoiced information.

  Multiplexing means 108 combines pitch information, voiced / unvoiced information for each harmonic band, and envelope information for each harmonic band into one series. The quantizing unit 109 quantizes the encoded information (for example, quantizes so as to have a predetermined number of bits for each element), and outputs the obtained digital speech encoded information.

  FIG. 10 shows the configuration of a speech decoding apparatus described in Non-Patent Document 2 that conforms to the MBE encoding method. The speech decoding apparatus 200 shown in FIG. 10 is opposite to the speech encoding apparatus 100 described above, and is given the digital speech encoding information output by the speech encoding apparatus 100.

  In FIG. 10, speech decoding apparatus 200 includes inverse quantization means 201, demultiplexing means 202, voiced / unvoiced envelope separation means 203, harmonic oscillation means 204, interpolation means 205, noise generation means 206, frequency conversion means 207, An envelope information replacing unit 208, a waveform restoring unit 209, and an adding unit 210 are included.

  In FIG. 10, inverse quantization means 201 estimates encoded information before quantization by inverse quantization from the arrived digital speech encoded information. The demultiplexing means 202 demultiplexes the dequantized speech coding information into pitch information, voiced / unvoiced information, and envelope information.

  The voiced / unvoiced envelope separation means 203 separates the envelope information into voiced envelope information and unvoiced envelope information based on the demultiplexed voiced / unvoiced information. The voiced envelope information has zero power and phase in the harmonic band that is unvoiced, and the voiced envelope information has zero power in the harmonic band that is voiced. Based on the pitch information and the voiced envelope information, the harmonic oscillation means 204 generates a sine wave signal having an amplitude and a phase corresponding to the voiced comprehensive information for each harmonic component, and adds the sine wave signals of all the harmonic components. In addition, voiced speech is synthesized. The generated sine wave signal is adjusted so that the amplitude and the phase are continuous in accordance with the voiced comprehensive information.

  The interpolation unit 205 interpolates (for example, linear interpolation) the unvoiced envelope information in accordance with the frequency resolution of the frequency conversion unit 207 to obtain a unvoiced amplitude spectrum. The noise generation unit 206 generates white noise by any known method, and the frequency conversion unit 207 performs frequency conversion of the white noise signal with the same parameters as the frequency conversion unit 101 described above to obtain a noise spectrum. The envelope information replacement unit 208 multiplies the noise spectrum from the frequency conversion unit 207 by the unvoiced amplitude spectrum from the interpolation unit 205 to calculate the unvoiced spectrum. The waveform restoration unit 209 performs an IFFT on the unvoiced spectrum with parameters corresponding to the frequency conversion unit 207 and generates an unvoiced voice by performing overlap addition.

  The adding means 210 adds the voiced voice from the harmonic oscillation means 204 and the unvoiced voice from the waveform restoration means 209 to obtain a decoded voice and outputs it.

  In the above, the configurations and operations of the speech encoding apparatus 100 and speech decoding apparatus 200 that comply with the MBE encoding scheme have been described. However, the AMBE encoding scheme and the IMBE encoding scheme also include estimation of speech parameters and accuracy of quantization. And in principle, they are very similar in principle. Any of the MBE speech coding systems is highly resistant to noise and can provide stable quality at a low bit rate.

"Survey report on frequency sharing with digital system in frequency for 150MHz analog simple radio station", Ministry of Internal Affairs and Communications, Hokuriku General Communications Bureau, Study Group Information, 2011. Daniel W. Griffin and Jae S .; Lim, “Multiband Excitation Vocoder,” IEEE Trans. on Acoustics, Speech and Signal Processing, Vol. ASSP-36, no. 8, pp. 1223-1235, 1988.

  However, as reported in Non-Patent Document 1, the MBE-based speech encoding method has a problem that the decoded speech becomes “sound that seems to be clogged with a nose” and the clarity is impaired.

  Therefore, in view of the above problems, a speech decoding apparatus, speech decoding method, speech decoding program, and communication device that improve the clarity of decoded speech are desired.

  A speech decoding apparatus according to a first aspect of the present invention is a speech decoding apparatus for decoding digitally encoded information encoded according to an MBE-based speech encoding method. (1) Decoding digital speech encoded information MBE decoding means for generating decoded speech; (2) a burst sound detecting means for detecting a burst sound of the decoded voice; and (3) a bursting processing means for bursting the detected burst sound. And

  A speech decoding apparatus according to a second aspect of the present invention is a speech decoding apparatus that decodes digitally encoded information encoded in accordance with an MBE speech encoding method. (1) Decodes digital speech encoded information MBE decoding means for generating decoded speech, (2) frequency domain burst detection means for detecting the burst sound of the decoded speech in the frequency domain, and (3) continuous burst sound in the frequency domain burst detection means. Centroid time calculation means for calculating the centroid time by calculating the centroid of the sum of the non-negative signals obtained by converting the decoded speech in the same number of detected frames into non-negative values for each sample; (4 A time domain burst sound detecting means for detecting a burst sound of the decoded speech in the time domain; and (5) a burst for selecting the centroid time and the burst information obtained from the time domain burst sound detecting means based on the burst information. Affection A selection means; (6) a burst verification means for re-determining whether or not the sound is a burst sound based on the determination result of the frequency domain burst sound detection means and the burst information; and (7) a burst sound in the burst verification means. Rupture processing means for redesigning a predetermined weight coefficient designed in advance based on the weight coefficient design information obtained from the burst information selection means based on the determined frame and multiplying the decoded speech by the weight coefficient. It is characterized by providing.

  A speech decoding method according to a third aspect of the present invention is a speech decoding method for decoding digitally encoded information encoded according to an MBE speech encoding method, wherein (1) the MBE decoding means is a digital speech Decoding the encoded information to generate decoded speech, (2) the burst sound detection means detects the burst sound of the decoded voice, and (3) the burst processing means bursts the detected burst sound. It is characterized by.

  A speech decoding method according to a fourth aspect of the present invention is a speech decoding method for decoding digitally encoded information encoded according to an MBE speech encoding method, wherein (1) the MBE decoder is a digital speech Decoding encoded information to generate decoded speech, (2) frequency domain burst sound detection means detects the burst sound of the decoded speech in the frequency domain, and (3) centroid time calculation means detects frequency domain burst sound The centroid time is calculated by calculating the centroid of the sum of the non-negative signal obtained by converting the decoded speech in the same number of frames as the number of times the plosive is continuously detected by the means into a non-negative value for each sample. (4) The time domain burst sound detection means detects the burst sound of the decoded speech in the time domain, and (5) the burst information selection means determines the centroid time and the burst information obtained from the time domain burst sound detection means, Based on the burst information (6) The burst verification means re-determines whether or not the sound is a burst sound based on the determination result of the frequency domain burst sound detection means and the burst information, and (9) the burst processing means determines the burst test. Based on the frame determined to be a plosive sound in the means, the predetermined weight coefficient designed in advance based on the weight coefficient design information obtained from the burst information selection means is redesigned, and the weight coefficient is added to the decoded speech. It is characterized by multiplication.

  A speech decoding program according to a fifth aspect of the present invention is a speech decoding program for decoding digitally encoded information encoded according to an MBE speech encoding method. MBE-based decoding means for generating decoded speech, (2) burst sound detecting means for detecting the burst sound of the decoded voice, and (3) bursting processing means for bursting the detected burst sound It is characterized by making it.

  A speech decoding program according to a sixth aspect of the present invention is a speech decoding program for decoding digitally encoded information encoded according to an MBE speech encoding method, wherein (1) digital speech encoded information MBE decoding means for generating decoded speech by decoding the signal, (2) frequency domain burst sound detecting means for detecting a burst sound of the decoded voice in the frequency domain, and (3) frequency domain burst sound detecting means. Centroid time calculation means for calculating the centroid time by calculating the centroid of the sum of the non-negative signal obtained by converting the decoded speech in the same number of frames as the number of times the plosive is detected into a non-negative value for each sample; (4) Time domain burst sound detection means for detecting a burst sound of decoded speech in the time domain, and (5) the centroid time and the burst information obtained from the time domain burst sound detection means. Burst information selection means for selection based on information, (6) burst verification means for re-determining whether or not a burst sound is based on the determination result and burst information of the frequency domain burst sound detection means, and (7) burst Based on the frame determined to be a plosive sound by the verification means, the predetermined weight coefficient designed in advance based on the weight coefficient design information obtained from the burst information selection means is redesigned, and the weight coefficient is applied to the decoded speech. It is made to function as a bursting process means to multiply.

  A communication device according to a seventh aspect of the present invention includes the speech decoding apparatus according to the first and second aspects of the present invention.

  ADVANTAGE OF THE INVENTION According to this invention, the unvoiced plosive of the decoded audio | voice lost by encoding can be burst, and the audio | voice which improved the clarity of the said decoded audio | voice can be provided to a user.

It is a functional block diagram which shows the structure of the speech decoding apparatus which concerns on 1st Embodiment. It is a figure which shows the example of the design method of a weighting coefficient. It is a functional block diagram which shows the audio | voice decoding apparatus structure which concerns on 2nd Embodiment. It is a functional block diagram which shows the structure of the speech decoding apparatus which concerns on 3rd Embodiment. It is a functional block diagram which shows the structure of the speech decoding apparatus which concerns on 4th Embodiment. It is a functional block diagram which shows the structure of the speech decoding apparatus which concerns on 5th Embodiment. It is a functional block country which shows the structure of the speech decoding apparatus which concerns on 6th Embodiment. It is a functional block diagram which shows the structure of the speech decoding apparatus which concerns on 7th Embodiment. It is a functional block diagram which shows the structure of the encoding apparatus of the MBE type audio | voice encoding system. It is a functional block diagram which shows the structure of the decoding apparatus of the MBE type audio | voice coding system. It is a figure which shows the example of a waveform of an unvoiced plosive and explains the acoustic phenomenon of an unvoiced plosive. It is a figure which shows the example of a waveform of unvoiced plosive / k / before encoding and after decoding. It is a figure which shows the example of a waveform of unvoiced plosive / t / before encoding and after decoding.

(A) Reason why the clarity of the decoded speech can be improved by each embodiment Prior to the description of the speech encoding device of each embodiment, the speech encoding device of each embodiment allows the speech encoding device of each embodiment to decode the decoded speech of the MBE speech encoding scheme. Explain why the clarity can be improved.

  First, the reason why the clarity of decoded speech is impaired will be considered. As a result of careful listening and comparison between the input speech and the decoded speech, it was found that unvoiced plosives (for example, / k /, / t /, and / p / in Japanese) are unclear. Here, the unvoiced plosive is a phoneme having a characteristic acoustic phenomenon as shown in FIG. FIG. 11 is a diagram showing a Japanese “ka (/ ka /)” waveform and explanation of its acoustic phenomenon (quoted by Shuichi Itabashi et al., “Speech Engineering” Morikita Publishing Co., Ltd., 1st edition, No. 1) Chapter 2, P.27, Figure 2.13). As shown in FIG. 11, the silent plosive sound has three states of a silent part, a rupture part, and an aerial part. However, depending on the utterance, there is a case where the sound part does not exist, and such a phenomenon is particularly large in / p /.

  However, when the waveform of the decoded speech is investigated, it can be confirmed that this characteristic acoustic phenomenon is impaired. FIGS. 12 and 13 are waveforms relating to unvoiced plosives / k / and / t /, respectively, and the upper (FIGS. 12A and 13A) and the lower (FIGS. 12B and 13), respectively. (B) shows the waveforms of speech before encoding and after decoding, respectively. In both cases, the solid line is a waveform, and the range enclosed by a broken-line rectangle represents the range of unvoiced plosives (ruptured portion and aerial portion). In FIGS. 12 and 13, the horizontal axis represents time (unit: millisecond), and the vertical axis represents amplitude value without unit.

  In any of the waveforms of / k / and / t /, the amplitude suddenly increases when the transition is made from the silent part to the rupture part before encoding (this is called “rupture” and is called the rupture sound). After decoding, the amplitude gradually increases in the two states of the rupture part and the sound part. That is, since the unvoiced plosive sound of the decoded speech is not ruptured, the decoded speech becomes unclear.

  Next, the mechanism by which the characteristic acoustic phenomenon of unvoiced plosives is impaired will be considered. The length of the rupture part and the sound part of the unvoiced plosive is about 5 to 30 ms. On the other hand, in many cases, the MBE speech coding method performs analysis with a half overlap (that is, a frame period of 10 ms) using, for example, a frame length of 20 ms, and does not store the phase information of unvoiced sound.

  In such frame processing, the characteristic acoustic phenomenon of the unvoiced plosive sound that occurs in a short time cannot be retained in the encoded information, and therefore cannot be correctly reproduced in the decoded speech. In the decoding of unvoiced sound, noise is generated and added in a half-overlapping manner as in the case of encoding. Therefore, the unvoiced plosive sound is transformed into an unvoiced sound whose amplitude gradually changes.

  Next, how the present invention improves the clarity of decoded speech will be described. In the unvoiced plosive of the decoded speech, a sudden change in amplitude is lost due to encoding, but the envelope of the power spectrum is not lost. In other words, if a sudden change in amplitude is reproduced, the unvoiced plosive of the decoded speech will be perceived as a correct unvoiced plosive. Therefore, by detecting the unvoiced plosive sound by an appropriate method and applying amplitude modulation to the detected unvoiced plosive sound to cause a characteristic acoustic phenomenon that seems to be a silent plosive sound, the unvoiced plosive sound can be clearly heard. The present inventor thought that the clarity of the decoded speech was improved.

  As described above, the present invention improves the intelligibility of decoded speech by detecting the burst sound of the decoded speech and reproducing the characteristic acoustic phenomenon of the unvoiced burst sound by amplitude modulation.

(B) First Embodiment Next, a first embodiment of the speech decoding apparatus, speech decoding method, speech decoding program, and communication device according to the present invention will be described in detail with reference to the drawings.

  The speech decoding apparatus according to the first embodiment performs decoding in accordance with the MBE-based speech encoding method, and the same applies to the second and later embodiments described later. In the first to seventh embodiments, since voiced plosives are not handled, for the sake of convenience, “unvoiced plosives” are hereinafter simply referred to as “plosives”.

(B-1) Configuration of the First Embodiment FIG. 1 is a functional block diagram showing the configuration of the speech decoding apparatus according to the first embodiment.

  In FIG. 1, the speech decoding apparatus 1 </ b> A according to the first embodiment includes a receiving unit 11, an MBE decoding unit 12, a burst sound detecting unit 13, and a bursting processing unit 14.

  The audio decoding device 1A decodes digitally encoded information encoded by the MBE audio encoding method, and improves the clarity of the obtained decoded audio. Here, the speech decoding apparatus 1A can be configured by hardware, and can be realized by the CPU and software executed by the CPU (speech decoding program). 1 is functionally represented in FIG.

  The opposing speech coding apparatus has a configuration illustrated in FIG. 9, for example, and transmits digital speech coding information (hereinafter referred to as digital coding information) coded according to the MBE speech coding method. Is sent to the wireless line. In this embodiment, the digital encoded information is exemplified by a case where it is transmitted by wireless communication through a wireless line, but may be transmitted through a wired line.

  The receiving unit 11 receives digital encoded information transmitted by wireless communication, and gives the obtained digital encoded information to the MBE decoding unit 12.

  In the first embodiment, it is assumed that the receiving unit 11 is a digital wireless device, and the processing of the receiving unit 11 is further simplified. As long as the digital encoded information can be acquired, the digital encoded information may be acquired by, for example, not wireless communication but wired communication. Since any communication means may cause a packet loss, the receiving means 11 performs a process for compensating for the packet loss, and the information subjected to the compensation process is sent to the MBE decoding means 12 as digitally encoded information. It may be given. Depending on the type of the MBE decoding unit 12, the MBE decoding unit 12 may include a packet loss compensation process, and in this case, a prior compensation process by the receiving unit 11 is unnecessary.

  The MBE decoding means 12 decodes the digital encoded information using a decoding method corresponding to the encoding method used to generate the digital encoded information. The decoded speech obtained by the decoding by the MBE decoding unit 12 is given to the burst sound detecting unit 13 and the bursting processing unit 14. Here, the MBE decoding means 12 can be widely applied as long as it is a decoding means using an MBE speech coding method. For example, the decoding device (decoding method) shown in FIG. 10 may be used, or AMBE (including AMBE + 2) or IMBE described above may be used.

  The plosive sound detection means 13 acquires the decoded voice decoded by the MBE decoding means 12, analyzes the decoded voice, and determines whether or not the frame has a burst sound burst start point. The plosive sound detecting means 13 gives the obtained determination result (referred to as a rupture truth value) to the rupturing processing means 14.

  As shown in FIG. 1, the plosive sound detection means 13 includes a short cycle power calculation unit 15, a power ratio calculation unit 16, and a burst detection unit 17.

  The short cycle power calculation unit 15 calculates the power of decoded speech in a cycle shorter than that of the MBE decoding means 12, and the obtained short cycle power is given to the power ratio calculation unit 16.

  The power ratio calculation unit 16 calculates the power ratio by dividing the short cycle power given from the short cycle power calculation unit 15 by the reference power determined by a predetermined rule, and the obtained power ratio is This is given to the burst detection unit 17. The predetermined rule for determining the reference power in the power ratio calculation unit 16 will be described in detail in the operation section.

  The rupture detection unit 17 generates a rupture truth value by determining whether or not a given short period power ratio has a value equal to or greater than a predetermined threshold, and the obtained rupture truth value is a rupture value. The output from the sound detection means 13 is given to the bursting processing means 14. That is, if the power ratio of the short cycle exceeds the threshold value, the burst true value is set to true (TRUE), and if it does not exceed the threshold value, it is set to false (FALSE).

  The bursting processing means 14 performs amplitude modulation by multiplying the decoded speech given from the MBE decoding means 12 by a predetermined weighting factor if the burst truth value given from the burst sound detection means 13 is true, If the burst truth value is false, the decoded speech is passed as it is to obtain improved speech, and the obtained improved speech is output. The amplitude modulation is a process for artificially generating a silent portion and a rupture portion, and a predetermined weight coefficient used for the amplitude modulation is calculated in advance by a design method described later.

(B-2) Operation of the First Embodiment Next, the operation of the speech decoding apparatus 1A of the first embodiment will be described.

  The opposing speech encoding apparatus transmits digitally encoded information encoded according to the MBE speech encoding method through a wireless line.

  The digitally encoded information communicated wirelessly is received by the receiving means 11 of the speech decoding apparatus 1A. The received digital encoded information is decoded by the MBE decoding means 12 by a decoding method corresponding to the MBE voice encoding method, and the obtained decoded speech is converted into the burst sound detecting means 13 and the bursting process. Provided to means 14.

  In the plosive sound detection means 13, the short cycle power calculation unit 15 calculates the power of the decoded speech at a shorter cycle than the MBE decoding means 12. For example, if the short cycle power calculation cycle is 2 ms, five short cycle powers are calculated from the decoded speech for a given 10 ms frame. The calculation cycle of the short cycle power may be 2.5 ms. This short period power is given to the power ratio calculation unit 16. As the short cycle power calculation cycle, an arbitrary time length of, for example, 1 ms to 5 ms can be set. However, for the convenience of calculation, a time length of (1 / integer of the frame cycle) is preferably used.

  The power ratio calculation unit 16 calculates the power ratio by dividing the given short period power by the reference power determined by a predetermined rule. The obtained power ratio is given to the burst detection unit 17.

  The predetermined rule for determining the reference power in the power ratio calculation unit 16 differs depending on whether or not the information of the past frame is used.

  For example, when calculating the power ratio using only the current frame without using the past frame, the minimum value of the short period power from the time immediately before the time at which the power ratio is desired to the first time in the frame is used as the reference power. The method is preferably used. However, since this method cannot calculate the first power ratio in the frame, the power ratio is set to 1.

  Also, for example, when calculating the power ratio using past frames, a predetermined reference time width is set, and the minimum value of the short cycle power from immediately before the time at which the power ratio is to be calculated to the previous reference time width is set. A method of using the reference power is preferably used.

  The rupture detection unit 17 determines whether or not a given short period power ratio has a value equal to or greater than a predetermined threshold value, and generates a rupture truth value. The obtained burst truth value is given to the bursting processing means 14 as an output of the burst sound detection means 13. That is, the burst detection unit 17 sets the burst true / false value to true (TRUE) if the short cycle power ratio exceeds the threshold, and false (FALSE) if it does not exceed the threshold. The threshold value is for detecting a transition from the silent part to the rupture part, and the threshold value is not particularly limited, but a value of about 100 to 1000 is preferably used.

  In the bursting processing means 14, if the given burst true / false value is true, the given decoded voice is multiplied by a predetermined weight coefficient to perform amplitude modulation, and if the burst true / false value is false, the decoded voice is decoded. To get the improved voice. The obtained improved speech is output to the subsequent stage.

  The amplitude modulation is a process for artificially generating a silent portion and a rupture portion, and a predetermined weight coefficient used for the amplitude modulation is calculated in advance by a predetermined design method.

  FIG. 2 is a diagram illustrating the concept of weighting factor design. The amplitude is designed on a logarithmic scale (decibels) so that it is natural to the sense of hearing.

  First, the silent portion before the rupture portion is suppressed in order to make the rupture phenomenon at the rupture portion start end more clear. Here, the final gain of the silent part is set to −100 dB in order to achieve almost complete silence and to make the burst phenomenon more abrupt.

  Next, it amplifies to 9 dB so that the rupture portion start end has sufficiently larger power than the original waveform. By increasing it rapidly in a short time rather than discretely, the waveform becomes discontinuous so that no extra noise is generated. Finally, since the second half of the rupture portion is a section in which the acoustic phenomenon shifts from the rupture portion to the air sound portion or vowel, it gradually approaches 0 dB.

  In FIG. 2, the silent part is 10 ms, the rupture part start end is 5 ms, and the rupture part latter half is 15 ms. In consideration of the arrangement of the burst start point in the frame where the burst sound is detected, in order to realize amplitude modulation by this weighting coefficient, when the frame period is 10 ms, at least one frame of decoded speech is stored. As a result, the output of the improved voice is delayed by one frame.

  Note that the lengths of the silent part, the start of the rupture part, and the latter half of the rupture part do not have to be as shown in FIG. 2, for example, the silence part may be 5 ms. In this case, amplitude modulation can be performed without delaying the improved sound. Even if there is no silent part, that is, amplifying from 0 dB to 9 dB at the start of the rupture part without creating a silent part, a constant effect can be obtained, and such a design can also be applied.

  In the manner shown in FIG. 2, the beginning of the silent part is represented as 0 ms, but this does not suggest that the beginning of the silent part coincides with the beginning of the frame, and the burst is detected in the frame in which the burst sound is detected. As long as the condition that the part must be formed is satisfied, the design of the weighting factor can be freely translated in the time direction.

(B-3) Effect of First Embodiment According to the first embodiment, in the decoded speech of the MBE speech coding scheme, the plosive sound in which the characteristic acoustic phenomenon is impaired is disrupted. It is possible to provide a user with improved voice clarity by improving the clarity of the decoded voice.

(C) Second Embodiment Next, a second embodiment of the speech decoding apparatus, speech decoding method, speech decoding program, and communication device according to the present invention will be described with reference to the drawings.

  In the first embodiment, the case where the plosive is detected based on the temporal change in the amplitude of the waveform of the decoded speech has been shown. However, as described with reference to FIGS. 12 and 13, since it is highly possible that the information of the temporal change in amplitude that seems to be a plosive sound is lost in the decoded speech, such a plosive sound cannot be detected. On the other hand, since the MBE-based speech encoding method encodes the power of the frequency spectrum as envelope information, the reproducibility of the power spectrum is high.

  Therefore, in the second embodiment, a plosive sound is detected using the power spectrum.

(C-1) Configuration of Second Embodiment FIG. 3 is a functional block diagram showing the configuration of the decoded speech sound quality improvement apparatus 1B of the second embodiment, and is related to FIG. 1 according to the first embodiment. The same and corresponding components are indicated by the same reference numerals.

  In FIG. 3, the decoded speech sound quality improvement apparatus 1 </ b> B according to the second embodiment includes a receiving unit 11, an MBE decoding unit 12, a burst sound detecting unit 21, and a bursting processing unit 14.

  The second embodiment is different from the first embodiment in that a plosive detection means 21 is provided in place of the plosive detection means 13 as compared with the first embodiment.

  The plosive sound detecting means 21 analyzes the decoded speech given from the MBE decoding means 12 to determine whether or not the frame has a rupture start point of the plosive sound, and the obtained determination result (rupture authenticity). Is referred to as a value). That is, the plosive detection means 21 is a method of detecting whether or not the frame is the start of a rupture part of a plosive by performing pattern matching of characteristic information of the frequency characteristics of the frame.

  As shown in FIG. 3, the plosive sound detection means 21 includes a frequency analysis unit 22 and a pattern identification unit 23.

  The frequency analysis unit 22 acquires the decoded speech from the MBE decoding unit 12, calculates a frequency spectrum for each frame, and calculates a power spectrum for each frame. The obtained power spectrum is given to the pattern recognition unit 23. Any method can be used as the method for calculating the power spectrum, and for example, FFT (First Fourier Transform), wavelet transform, filter bank, or the like can be applied. Further, if wavelet transform or a filter bank is used, it is possible to divide the band evenly. However, here, equal division of the band using the filter bank is preferably used. The band to be analyzed is a meaningless DC component, a band having a pitch frequency, and 3400 Hz or more that is often suppressed by speech coding is unnecessary. For example, the bandwidth is set to 400 Hz and the center frequency is set to 400 Hz, 800 Hz,. An 8-band filter bank at 3200 Hz is recommended.

  The pattern identification unit 23 recognizes the pattern of the power spectrum given from the frequency analysis unit 22 to determine whether or not the frame is a plosive sound, sets the determination result as a burst truth value, and obtains the obtained burst The true / false value is given to the rupturing processing means 14 as an output of the plosive sound detecting means 21. Here, various methods can be applied to the pattern recognition. For example, simple pattern matching or a neural network can be selected, and a support vector machine is preferably used.

  In addition, although it wrote as if only a power spectrum was used above, the pattern recognition part 23 may add the power of the decoded speech itself which is not band-divided, and may recognize a pattern. Although it has been written as if only the power spectrum obtained from a single frame has been used so far, the pattern identification unit 23 may use a past frame, or a future frame by delaying the output. May be used. Further, the power value may be normalized using an arbitrary value, for example, it may be normalized by dividing another power value by the power of decoded speech of the current frame.

(C-2) Operation of the Second Embodiment Next, the operation of the speech decoding apparatus 1B according to the second embodiment will be described. Since the overall operation of the speech decoding apparatus 1B is the same as that of the first embodiment, the description thereof is omitted, and the operation of the plosive detection means 21 will be described below.

  The decoded speech output from the MBE decoding means is given to the plosive sound detection means 21. In the plosive detection means 21, the decoded speech is subjected to frequency analysis by the frequency analysis unit 22 to calculate a power spectrum for each frame, and the obtained power spectrum for each frame is given to the pattern recognition unit 23.

  In the pattern recognition unit 23 of the plosive sound detection means 21, the power spectrum for each frame is subjected to pattern recognition by a predetermined pattern recognition method to determine whether or not the frame is a plosive sound. It is given to the rupturing processing means 14 as a rupture truth value.

  Then, in the bursting processing means 14, if the given burst truth value is true, the given decoded speech is multiplied by a predetermined weight coefficient to perform amplitude modulation, and if the burst truth value is false, The decoded speech is passed as it is to obtain improved speech. The obtained improved speech is output to the subsequent stage.

(C-3) Effects of the Second Embodiment According to the second embodiment, the burst sound in which the characteristic acoustic phenomenon is impaired is more reliably burst in the decoded speech of the MBE speech coding scheme. Therefore, it is possible to provide the user with the voice that improves the clarity of the decoded voice and improves the listening comfort.

(D) Third Embodiment Next, a third embodiment of the speech decoding apparatus, speech decoding method, speech decoding program, and communication device according to the present invention will be described with reference to the drawings.

  In the second embodiment, the configuration in which all frames in which a plosive sound is detected is ruptured has been described. However, in the case of a plosive sound in which the rupture portion exceeds 10 ms, for example, there is a possibility that the plosive sound is detected in two consecutive frames. In the first place, a plosive sound may be detected continuously for two or more frames due to a pattern recognition error. In such a case, in the configuration of the second embodiment, there is a possibility that one plosive sound is ruptured twice.

  Therefore, in the third embodiment, even when a plurality of frames are continuously detected as a plosive sound, the bursting is performed only once.

(D-1) Configuration of Third Embodiment FIG. 4 is a functional block diagram showing a configuration of a speech decoding apparatus 1C according to the third embodiment. In FIG. 4, the same and corresponding components as those in FIG. 1 according to the first embodiment and FIG. 3 according to the second embodiment are denoted by the same reference numerals.

  In FIG. 4, the speech decoding apparatus 1 </ b> C according to the third embodiment includes a receiving unit 11, an MBE decoding unit 12, a burst sound detecting unit 21, and a bursting processing unit 31.

  The third embodiment is different from the second embodiment in that a rupture treatment means 31 is provided instead of the rupture treatment means 14.

  The bursting processing means 31 bursts the decoded speech based on the burst truth value from the bursting sound detection means 21 and outputs the obtained improved speech.

  As shown in FIG. 4, the bursting processing unit 31 includes a burst verification unit 32 and an amplitude modulation unit 33.

  The burst verification unit 32 compares the burst true value given from the burst sound detection means 21 with the past burst true value, and generates a corrected burst true value whose true value is not continuous. The corrected burst truth value is given to the amplitude modulation unit 33.

  Here, a method of generating a corrected burst truth value in the burst verification unit 32 will be described.

  First, the burst verification unit 32 stores the burst true / false value of one frame in the past, and when the burst true / false value of the current frame is newly input, the burst verification unit 32 stores the past burst true / false value stored therein. Compare the value with the burst truth value of the current frame. Then, only when the previous burst true value is “false” and the current burst true value is “true”, the burst test unit 32 generates the corrected burst true value as “true”.

  In other cases, that is, (a) the past burst true value is “false” and the current burst true value is “false”, (b) the past burst true value is “true”. When the current burst truth value is “false”, (c) when the previous burst truth value is “true” and the current burst truth value is “true”, the burst test unit 32 corrects the burst. A Boolean value is generated as “false”.

  After generating the corrected burst true / false value, the burst verification unit 32 stores the current burst true / false value overwriting the previous burst true / false value.

  If the given corrected burst true / false value is “true”, the amplitude modulating unit 33 performs amplitude modulation by multiplying the given decoded speech by a predetermined weight coefficient. On the other hand, if the corrected burst true / false value is “false”, the amplitude modulation unit 33 passes the decoded speech as it is to obtain improved speech. The obtained improved speech is output as the output of the bursting processing means 31. The operation of the amplitude modulation unit 33 is the same as that of the bursting processing unit 14 of FIG. 1 according to the first embodiment except that the second input is a bursting truth value or a corrected bursting truth value. It is.

  As described above, by performing amplitude modulation on the basis of the corrected burst true / false value, an error that erroneously bursts continuously for two or more frames does not occur.

(D-2) Operation of the Third Embodiment Next, the operation of the speech decoding apparatus 1C according to the third embodiment will be described. Since the overall operation of the speech decoding apparatus 1C is the same as that in the first and second embodiments, the description thereof will be omitted, and the operation of the bursting processing means 31 will be described below.

  The rupture truth value output from the rupture sound detection means 21 is given to the rupture verification unit 32 of the rupture processing means 31. In the burst verification unit 32 of the burst processing means 31, the given burst true / false value of the current frame is compared with the burst true / false values of the past frame. At this time, if the burst true / false value of the past frame is false and the burst true / false value of the current frame is true, the burst verification unit 32 generates and outputs the corrected burst true / false value as true. On the other hand, in other cases, the corrected burst true / false value is generated and output as false. The obtained corrected burst true / false value is given to the amplitude modulator 33. In the burst verification unit 32, the generated corrected burst true / false value is overwritten and stored on the burst true / false value of the past frame.

  The amplitude modulation unit 33 of the bursting processing unit 31 performs amplitude modulation by multiplying the duplicated voice by a predetermined weighting factor if the given corrected burst true / false value is true. On the other hand, if the corrected burst true / false value is false, the amplitude modulation unit 33 passes the decoded speech as it is to obtain improved speech. The obtained improved speech is output as the output of the bursting processing means 31.

(D-3) Effect of the Third Embodiment According to the third embodiment, the burst sound in which the characteristic acoustic phenomenon is impaired is burst in the decoded speech of the MBE speech coding method, In addition, since it is not accidentally continuously burst, it is possible to provide the user with a voice that improves the clarity of the decoded voice and improves the listening comfort.

(E) Fourth Embodiment Next, a fourth embodiment of the speech decoding apparatus, speech decoding method, speech decoding program, and communication device according to the present invention will be described with reference to the drawings.

  In the first to third embodiments, a fixed weight coefficient is used. However, the burst portion of the input voice does not always occur at a specific position in the frame. For this reason, there is a possibility that a sufficient effect of rupturing may not be obtained in an attempt to rupture the originally silent portion or the latter half of the rupture portion.

  Therefore, in the fourth embodiment, the weight coefficient is translated in the time direction based on the centroid time of the decoded speech, so that the part that was originally the start of the rupture part is correctly ruptured.

  The feature of the fourth embodiment is a rupture processing means, and any of the plosive detection means of the first and second embodiments can be used as the plosive detection means. Then, the case where the plosive detection means 21 of 2nd Embodiment is used is illustrated.

(E-1) Configuration of the Fourth Embodiment FIG. 5 is a functional block diagram showing the configuration of the speech decoding apparatus of the fourth embodiment. In FIG. 5, the same and corresponding components as those in FIG. 1 according to the first embodiment and FIG. 3 according to the second embodiment are denoted by the same reference numerals.

  In FIG. 5, the speech decoding apparatus 1D according to the fourth embodiment includes a receiving unit 11, an MBE decoding unit 12, a burst sound detecting unit 21, and a bursting processing unit 41.

  The fourth embodiment differs from the second and third embodiments in that a rupture treatment means 41 is provided in place of the rupture treatment means 14 as compared to the second and third embodiments. Is different.

  The bursting processing means 41 bursts the given decoded voice based on the given burst truth value, and outputs the obtained improved voice.

  The bursting processing means 41 includes a non-negative value unit 42, a centroid time calculation unit 43, and an amplitude modulation unit 44.

  The decoded speech given to the bursting processing means 41 is given to the non-negative value converting unit 42 and the amplitude modulating unit 44.

  The non-negative value converting unit 42 converts each sample of the given decoded speech into a non-negative value, and the obtained non-negative value signal is supplied to the centroid time calculating unit. As a method of converting to a non-negative value, any method can be applied as long as the output is a non-negative value, but an absolute value is preferably used.

The centroid time calculation unit 43 calculates the centroid time of energy within the frame of the given non-negative signal, and the obtained centroid time is given to the amplitude modulation unit 44. The centroid time is a unique feature amount defined by equation (1). In equation (1), t is the time, T is the frame length, X (t) is the non-negative signal, C is the centroid time, and the unit of t and T is a sample. For convenience, t represents a relative time within the frame.

  The amplitude modulation unit 44 translates a predetermined weight coefficient in the time direction based on the centroid time given from the centroid time calculation unit 43, and the burst truth value given from the plosive sound detection means 21 is true. If there is, the given decoded speech is multiplied by the weighting factor to perform amplitude modulation. If the burst truth value is false, the decoded speech is passed as it is to obtain improved speech, and the obtained improved speech is ruptured. Is output as the output of the conversion processing means 41.

The design method of the weighting factor is the same as that of the bursting processing unit 14 of the first embodiment, but differs in that the weighting factor calculated in advance based on the centroid time is translated in the time direction. The parallel movement of the weighting factor is performed by determining the peak position of the weighting factor based on the centroid time. The method for determining the peak position is preferably used because the simplest method is to match the centroid time to the peak position as it is, but the centroid time is closer to the inside than the original peak position (in the equation (1), C is T / 2). In consideration of the tendency to approach (close to) and the tendency to move backward (C approaches T in equation (1)), the center-of-gravity time C ′ corrected by equation (2) may coincide with the peak position. .

  As described above, by changing the peak position of the weighting factor based on the centroid time, it is possible to avoid the problem that the effect of rupture is weakened by rupturing the silent part and the latter half of the rupture part.

(E-2) Operation of the Fourth Embodiment Next, the operation of the speech decoding apparatus 1D according to the fourth embodiment will be described. Since the overall operation of the speech decoding apparatus 1D is the same as that in the first and second embodiments, the description thereof will be omitted, and the operation of the bursting processing means 41 will be described below.

  The decoded speech output from the MBE decoding unit 12 is given to the non-negative value converting unit 42 and the amplitude modulating unit 44 of the bursting processing unit 41. In the non-negative value converting unit 42, each sample of the given decoded speech is converted into a non-negative value, and the obtained non-negative value signal is supplied to the centroid time calculating unit 43.

  The centroid time calculation unit 43 calculates the centroid time of energy within the frame of the given non-negative signal according to the equation (1), and the obtained centroid time is provided to the amplitude modulation unit 44.

  The amplitude modulation unit 44 translates a predetermined weighting factor in the time direction based on the given centroid time, and if the given burst truth value is true, the weighting factor is added to the given decoded speech. If the burst true / false value is false, the decoded speech is passed as it is, the improved speech is obtained, and the obtained improved speech is output as the output of the burst processing means 41.

(E-3) Effect of the Fourth Embodiment According to the fourth embodiment, the burst sound in which the characteristic acoustic phenomenon is impaired is more reliably burst in the decoded speech of the MBE speech coding scheme. Therefore, it is possible to provide the user with the voice that improves the clarity of the decoded voice and improves the listening comfort.

(F) Fifth Embodiment Next, a fifth embodiment of the speech decoding apparatus, speech decoding method, speech decoding program, and communication device according to the present invention will be described with reference to the drawings.

  In the fourth embodiment, the weighting factor is translated in the time direction based on the centroid time in the frame of the decoded speech. This method is effective, but if a burst sound is detected continuously in multiple frames, most of the latter half of the rupture is applied to the next frame, and the relationship between the centroid time and the peak position of the weighting coefficient. May be weak.

  Therefore, in the fifth embodiment, the peak position of the weighting coefficient is determined based on the number of frames in which a plosive is detected continuously.

(F-1) Configuration of Fifth Embodiment FIG. 6 is a functional block diagram showing the configuration of the speech decoding apparatus 1E of the fifth embodiment. In FIG. 6, the same reference numerals are given to the same and corresponding components as those in FIG. 1 according to the first embodiment, FIG. 3 according to the second embodiment, and FIG. 5 according to the third embodiment. Show.

  In FIG. 6, the speech decoding apparatus 1 </ b> E according to the fifth embodiment includes a reception unit 11, an MBE decoding unit 12, a burst sound detection unit 21, and a bursting processing unit 51. The fifth embodiment is different from the second to fourth embodiments in that a rupture treatment means 51 is provided instead of the rupture treatment means 14 and 51. Is different.

  The feature of the fifth embodiment is the rupturing processing means 51, and any of the plosive detection means of the first and second embodiments can be used as the plosive detection means. In the form, the case where the plosive detection means 21 of the second embodiment is used is illustrated.

  The bursting processing means 51 bursts the given decoded voice based on the given burst truth value, and outputs the obtained improved voice.

  The bursting processing means 51 includes a burst verification unit 32, a detection frequency calculation unit 53, a burst time estimation unit 54, and an amplitude modulation unit 55.

  The decoded speech given to the rupturing processing means 51 is given to the amplitude modulation section 55, and the rupture truth value from the rupture sound detection means 21 is given to the rupture verification section 32 and the detection frequency calculation section 53.

  Since the operation of the burst verification unit 32 is the same as that of the third embodiment, description thereof is omitted.

  The number-of-detections calculation unit 53 counts the number of times that the burst sound has been detected continuously, and the obtained number of detections is given to the burst time estimation unit 54. The specific operation of the detection frequency calculation unit 53 has a detection frequency counter inside, and if the rupture truth value given from the plosive sound detection means 21 is true, the operation frequency counter is set to “1”. If the burst true / false value is false, the detection counter is reset to “0”, and the current detection counter value is output as the detection count.

  The rupture time estimation unit 54 estimates, based on the given number of detections, at which time of the first frame among the frames in which the rupture sound is continuously detected, the peak position of the rupture part power, The obtained burst time is given to the amplitude modulator 55.

  The rupture time estimation unit 54 has a note on the algorithm. In other words, if the burst sound is detected continuously for a long period of time (as a problem that is not realistic), the burst truth value becomes false in order to burst the frame that first detected the burst sound. The output must continue to be delayed until In order to prevent such a problem, an operation of ignoring the number of detections greater than a predetermined number of times is necessary.

  Hereinafter, the number of frames from the frame including the rupture portion start end to the last frame where the latter half of the rupture portion is considered to be present is referred to as a continuous rupture frame number.

  Since the rupture part is about 30 ms at the longest, it is reasonable to consider that the number of continuous rupture frames is about 3 frames at the maximum when the frame period is 10 ms. Therefore, it is assumed that the burst time is calculated based on the number of consecutive burst frames, the past detection count is stored for one frame, the current detection count is “0”, and the previous detection count is If it is “1” or “2”, the current number of detections is the number of continuous burst frames. If the current number of detections is “3”, the current number of detections is the number of continuous burst frames. In this case, the continuous burst frame number is set to “0”.

The calculation of the rupture time based on the number of times of detection is performed by equation (3). In equation (3), N is the number of continuous burst frames, P is the burst time, and P0 is the minimum burst time. P0 is an arbitrary value not less than 0 and less than (T-1), but the length of the rupture start point in the design of the weighting factor is preferably used. Note that, when N = 0, amplitude modulation by the weighting coefficient is not performed, so P does not need to be calculated. For example, the previous value is held as it is.

  The amplitude modulation unit 55 translates a predetermined weighting factor in the time direction based on the burst time given from the burst time estimation unit 54, and the corrected burst true / false value given from the burst verification unit 32 is true. If there is, the given decoded speech is multiplied by the weighting factor to perform amplitude modulation, and if the corrected burst truth value is false, the decoded speech is passed as it is to obtain improved speech, and the obtained improved speech is Output as the output of the bursting processing means 51.

  The design method of the weighting factor is the same as that of the rupturing processing means 14 of the first embodiment, except that the weighting factor calculated in advance based on the rupture time is translated in the time direction. The parallel movement of the weighting factor is performed by matching the peak position of the weighting factor with the burst time.

  As described above, by changing the peak position of the weighting coefficient based on the number of times that the burst sound has been detected continuously, the burst sound may be continuously burst when detected. The problem that the effect of the bursting is weakened by bursting the silent part can be avoided.

(F-2) Operation of Fifth Embodiment Next, the operation of the speech decoding apparatus 1E according to the fifth embodiment will be described. Since the overall operation of the speech decoding apparatus 1E is the same as that in the first to fourth embodiments, the description thereof will be omitted, and the operation of the bursting processing means 51 will be described below.

  The rupture truth value output from the plosive sound detection means 21 is given to the rupture verification section 32 and the detection frequency calculation section 53 of the rupture processing means 51.

  In the burst verification unit 32 of the bursting processing means 51, the burst true value of the given current frame is compared with the burst true value of the past frame in the same manner as in the third embodiment. A boolean value is generated. The obtained corrected burst true / false value is given to the amplitude modulation section 55.

  The number-of-detections calculation unit 53 of the bursting processing means 51 counts the number of times that the bursting sound is continuously detected based on the given burst true / false value of each frame, and the obtained number of detections is the burst time estimation unit. 54.

  The rupture time estimation unit 54 estimates, based on the given number of detections, at which time of the first frame among the frames in which the rupture sound is continuously detected, the peak position of the rupture part power, The obtained burst time is given to the amplitude modulator 55.

  The amplitude modulation unit 55 translates a predetermined weight coefficient in the time direction based on the burst time given from the burst time estimation unit 54, and the corrected burst true / false value given from the burst verification unit 32 is true. If there is, the given decoded speech is multiplied by the weighting factor to perform amplitude modulation, and if the corrected burst truth value is false, the decoded speech is passed as it is to obtain improved speech, and the obtained improved speech is Output as the output of the bursting processing means 51.

(F-3) Effect of Fifth Embodiment According to the fifth embodiment, in the decoded speech of the MBE speech coding method, the plosive sound in which the characteristic acoustic phenomenon is impaired is more reliably ruptured. Therefore, it is possible to provide the user with a voice that improves the clarity of the decoded voice and improves the listening comfort.

(G) Sixth Embodiment Next, a sixth embodiment of the speech decoding apparatus, speech decoding method, speech decoding program, and communication device according to the present invention will be described with reference to the drawings.

  In the fifth embodiment, even when a burst sound is continuously detected, only the frame in which the burst sound is first detected is burst, and the number of times the burst sound is continuously detected at the position to be further burst Can be set dynamically according to the burst time estimated based on the above, but the number of times that the burst sound is detected depends strongly on the accuracy of pattern recognition, so the result of burst time estimation becomes unstable. There is.

  Therefore, in the sixth embodiment, the rupture time is estimated based on both the center-of-gravity time of the fourth embodiment and the number of detections of the fifth embodiment.

(G-1) Configuration of Sixth Embodiment FIG. 7 is a functional block diagram showing a configuration of a speech decoding apparatus 1F according to the sixth embodiment. In FIG. 7, the same and corresponding components as those in FIGS. 1 and 3 to 6 according to the first to fifth embodiments are denoted by the same reference numerals.

  In FIG. 7, the speech decoding apparatus 1 </ b> F according to the sixth embodiment includes a receiving unit 11, an MBE decoding unit 12, a burst sound detecting unit 21, and a bursting processing unit 61. The sixth embodiment is different from the fourth and fifth embodiments in that a rupture treatment means 61 is provided instead of the rupture treatment means 41 and 51. Is different.

  The bursting processing means 61 bursts the given decoded voice based on the given burst truth value, and outputs the obtained improved voice.

  The burst processing means 61 includes a burst verification unit 32, a non-negative value unit 42, a detection frequency calculation unit 53, a centroid time calculation unit 62, and an amplitude modulation unit 63.

  The decoded speech given to the bursting processing means 61 is given to the non-negative value converting section 42 and the amplitude modulating section 63, and the burst true / false value that is also given is given to the burst verification section 32 and the detection frequency calculation section 53. .

  Since the operation of the non-negative value unit 42 is the same as that of the fourth embodiment, description thereof is omitted.

  Since the operation of the burst verification unit 32 is the same as that of the third embodiment, description thereof is omitted.

  Since the operation of the detection frequency calculation unit 53 is the same as that of the fifth embodiment, description thereof is omitted.

  The center-of-gravity time calculation unit 62 calculates the center-of-gravity time in a frame in which a plosive is detected continuously based on the given non-negative signal and the number of detections, and the obtained center-of-gravity time is sent to the amplitude modulation unit 63. Given.

The center-of-gravity time calculation unit 43 of the fourth embodiment calculates the center-of-gravity time for a single frame, but the center-of-gravity time calculation unit 62 of the sixth embodiment uses the number of frames for the number of consecutive burst frames. Calculate the time. Therefore, the centroid time C is calculated by the equation (4). Note that t represents a relative time from the first frame when the plosive sound is detected.

  Note that the center-of-gravity time calculation unit 62 has the same algorithm cautions as the burst time estimation unit 54 of the fifth embodiment. Therefore, the number of continuous burst frames is set in the same manner as the burst time estimation unit 54.

  The amplitude modulation unit 63 translates a predetermined weighting factor in the time direction based on the centroid time given from the centroid time calculation unit 62, and the corrected burst true / false value given from the burst test unit 32 is true. If there is, the given decoded speech is multiplied by the weighting factor to perform amplitude modulation, and if the corrected burst truth value is false, the decoded speech is passed as it is to obtain improved speech, and the obtained improved speech is Output as the output of the bursting processing means 61.

The design method of the weighting coefficient is the same as that of the amplitude modulation unit 44 of the fourth embodiment, except that the centroid time does not stay within the frame in which the plosive is first detected. The method for determining the peak position of the weighting factor is preferably used since the simplest method is to match the centroid time to the peak position as is, as in the amplitude modulation unit 44 of the fourth embodiment, but the centroid time is the original peak position. In consideration of the tendency toward the inside and the tendency toward the rear, the center-of-gravity time C ′ corrected by the equation (5) may be matched with the peak position.

  As described above, when the burst sound is detected continuously, the center of gravity time is calculated over a plurality of frames, and the peak position of the weighting coefficient is changed based on the obtained center of gravity time. The problem of continuously bursting when the noise is continuously detected and the problem that the effect of bursting is weakened by bursting the silent part can be avoided.

(G-2) Operation of Sixth Embodiment Next, the operation of the speech decoding apparatus 1F according to the sixth embodiment will be described. Since the overall operation of the speech decoding apparatus 1F is the same as that in the first to fifth embodiments, the description thereof will be omitted, and the operation of the bursting processing means 61 will be described below.

  The rupture truth value output from the plosive sound detection means 21 is given to the rupture verification section 32 and the detection frequency calculation section 53 of the rupture processing means 61.

  In the burst verification unit 32 of the bursting processing means 61, the burst true / false value of the given current frame is compared with the burst true / false value of the past frame to generate a corrected burst true / false value. The obtained corrected burst truth value is given to the amplitude modulation unit 63.

  The number-of-detections calculation unit 53 of the bursting processing means 61 counts the number of times that the bursting sound is continuously detected based on the given burst true / false value of each frame, and the obtained number of detections is the centroid time calculation unit. 62.

  The decoded speech output from the MBE decoding unit 12 is given to the non-negative value converting unit 42 and the amplitude modulating unit 63 of the bursting processing unit 61.

  Each sample of the decoded speech is converted to a non-negative value by the non-negative value converting unit 42 of the bursting processing means 61, and the obtained non-negative value signal is provided to the centroid time calculating unit 62.

  The center-of-gravity time calculation unit 62 calculates the center-of-gravity time in the frame in which the burst sound has been continuously detected based on the given non-negative signal and the number of detections in which the burst sound has been continuously detected. The obtained barycentric time is given to the amplitude modulation unit 63.

  The amplitude modulation unit 63 translates a predetermined weighting factor in the time direction based on the centroid time given from the centroid time calculation unit 62, and the corrected burst true / false value given from the burst test unit 32 is true. If there is, the given decoded speech is multiplied by the weighting factor to perform amplitude modulation, and if the corrected burst truth value is false, the decoded speech is passed as it is to obtain improved speech, and the obtained improved speech is Output as the output of the bursting processing means 61.

(G-3) Effect of Sixth Embodiment As described above, according to the sixth embodiment, a plosive sound in which a characteristic acoustic phenomenon is impaired in the decoded speech of the MBE speech coding scheme. Since it is surely ruptured and is not accidentally continuously ruptured, it is possible to provide the user with a voice that improves the clarity of the decoded voice and improves the listening comfort.

(H) Seventh Embodiment Next, a seventh embodiment of the speech decoding apparatus, speech decoding method, speech decoding program, and communication device according to the present invention will be described with reference to the drawings.

  In the second to sixth embodiments, the burst sound of the decoded speech is detected using only the power spectrum on the assumption that the characteristic acoustic phenomenon is impaired. However, not all plosives are damaged, and fortunately, the plosives of the input voice may be reproduced with high accuracy. In such a case, when the bursting process is performed according to the second to sixth embodiments, the burst sound of the improved voice becomes excessively strong and may be heard unnaturally. In addition, in such a case, the time waveform is observed more accurately than the power spectrum, even if the power peak position and the peak size of the plosive sound are detected.

  Therefore, in the seventh embodiment, time domain plosive detection by observation of a time waveform and plosive detection by power spectrum pattern recognition are simultaneously performed, and the result is appropriately selected and used.

(H-1) Configuration of Seventh Embodiment FIG. 8 is a functional block diagram showing a configuration of a speech decoding apparatus according to the seventh embodiment. In FIG. 8, the same and corresponding components as those in FIGS. 1 and 3 to 7 according to the first to sixth embodiments are denoted by the same reference numerals.

  In FIG. 8, the speech decoding apparatus 1G according to the seventh embodiment includes a receiving means 11, an MBE decoding means 12, a frequency domain burst sound detecting means 71, a time domain burst sound detecting means 72, and a burst processing means 73. Have.

  Since the receiving means 11 and the MBE decoding means 12 are the same as those in the first embodiment, description thereof is omitted.

  Since the frequency domain plosive detecting means 71 is the same as the plosive detecting means 21 of the second embodiment, the description thereof is omitted. However, it should be noted that the frequency domain plosive detection means 71 detects a plosive by performing pattern recognition of the power spectrum with respect to the time domain plosive detection means 72 described below.

  The time domain plosive sound detection means 72 analyzes the time waveform of the decoded speech and extracts the burst information that summarizes the power peak position and peak value information of the rupture part related to the plosive sound. Is provided to the processing unit 73.

  The time domain plosive sound detection means 72 includes a short cycle power calculation unit 15, a power ratio calculation unit 16, and a rupture information extraction unit 77.

  The time domain plosive detection means 72 is substantially the same as the plosive detection means 13 of the first embodiment, but is provided with a rupture information extraction unit 77 instead of the rupture detection unit 17. The time domain plosive detection means 72 is different from the plosive detection means 13 of the first embodiment in that the time domain plosive detection means 72 outputs a rupture information while outputting a truth value.

  The rupture information extraction unit 77 searches for the time when the given power ratio is maximum to determine the rupture time, sets the power ratio at the rupture time as the rupture power ratio, and summarizes the rupture time and the rupture power ratio as rupture information. The burst information is given to the bursting processing means 73 as an output of the time domain burst sound detection means 72.

  The bursting processing means 73 is based on the burst truth value given by the frequency domain burst sound detecting means 71 and the burst information given by the time domain burst sound detecting means 72, and is decoded by the MBE decoding means 12. The voice is ruptured and the obtained improved voice is output.

  The rupture processing means 73 includes a non-negative value unit 42, a detection frequency calculation unit 53, a centroid time calculation unit 62, a rupture verification unit 74, a rupture information selection unit 75, and an amplitude modulation unit 76.

  Since the non-negative value unit 42 is the same as that of the fourth embodiment, the description thereof is omitted.

  Since the detection frequency calculation unit 53 is the same as that of the fifth embodiment, the description thereof is omitted.

  The center-of-gravity time calculation unit 62 is the same as that of the sixth embodiment, and a description thereof will be omitted.

  The rupture test unit 74 corrects the rupture truth value based on the given rupture truth value and the rupture information to generate a corrected rupture truth value. The obtained corrected rupture truth value is the amplitude modulation unit 76. Given to. The corrected burst truth value is true if the burst power ratio of the burst information is equal to or greater than a predetermined threshold, and otherwise, based on the burst true value, the same as the burst verification unit 32 of the third embodiment. Is set.

  The rupture information selection unit 75 selects the centroid time or the rupture information based on the rupture information, and the selected information is given to the amplitude modulation unit 76 as weight coefficient design information. The information is selected based on the burst power ratio of the burst information. That is, when a predetermined threshold is set, if the burst power ratio is larger than the threshold, the burst information is selected as weighting factor design information, and if the burst power ratio is smaller than the threshold, the centroid time is selected and the weighting factor is selected. Design information. The threshold value may be the same value as or different from the threshold value used in the burst verification unit 74, but the operation of the burst verification unit 74 and the burst information selection unit 75 can be performed by setting the two threshold values to the same value. A configuration for synchronization is preferably used.

  The amplitude modulation unit 76 designs the weighting factor based on the weighting factor design information given from the bursting information selection unit 75, and gives the corrected burst true / false value given from the burst testing unit 74 if it is true. When the decoded speech is multiplied by the weighting factor to perform amplitude modulation and the corrected burst truth value is false, the decoded speech is passed as it is to obtain improved speech, and the obtained improved speech is ruptured. 73 is output.

  When the weighting factor design information is the centroid time, the operation of the amplitude modulation unit 76 is the same as that of the amplitude modulation unit 63 of the sixth embodiment.

  When the weighting factor design information is burst information, after correcting the peak value of the weighting factor according to the bursting power ratio, the weighting factor is determined in the same manner as the bursting processing unit 14 of the first embodiment based on the bursting time. Is translated in the time direction. The correction of the peak value of the weighting factor is performed by combining a predesigned weighting factor peak gain (9 dB in the example of FIG. 2) and a burst peak ratio gain (sum for a logarithmic scale, product for a linear scale), Do not exceed the peak gain of the original weighting factor. That is, if the burst peak ratio is 9 dB or more, the weighting factor is always 0 dB (that is, no bursting is performed), and if the burst peak ratio is less than 9 dB, the combined gain of the peak gain factor and the burst peak ratio is 9 dB. The peak gain of the weighting coefficient is corrected so that For example, if the burst peak ratio is 4 dB, the weight gain peak gain is corrected to 5 dB.

  As described above, by selecting and using the result of the time domain plosive sound detection means and the result of the frequency domain plosive sound detection means, it is excessive when the characteristic acoustic phenomenon of the plosive sound is reproduced in the decoded speech. Rupture can be prevented.

(H-2) Operation of Seventh Embodiment Next, the operation of the speech decoding apparatus 1G according to the seventh embodiment will be described.

  The decoded speech output from the MBE decoding means 12 is given to the frequency domain burst sound detection means 71, the time domain burst sound detection means 72, and the burst processing means 73.

  In the frequency domain plosive detection means 71, the frequency analysis of the decoded speech is performed and the pattern of the power spectrum of each frame is recognized in the same manner as the plosive detection means 21 according to the second embodiment. Is determined whether or not it has a rupture start point of the plosive sound, and the obtained determination result is given to the detection number calculation unit 53 and the rupture test unit 74 of the rupture processing means 73 as a rupture truth value.

  The time domain plosive sound detection means 72 analyzes the time waveform of the given decoded speech, extracts the rupture information that summarizes the power peak position and peak value information of the rupture part related to the plosive sound, and obtained. The rupture information is given to the rupture verification unit 74 and the rupture information selection unit 75 of the rupture processing means 73.

  Here, the rupture information extraction unit 77 of the time-domain rupture sound detection means 72 searches for a time at which the power ratio calculated by the power ratio calculation unit 16 is maximum, and sets this as the rupture time. Then, the power ratio at the rupture time is defined as the rupture power ratio, the rupture time and the rupture power ratio are summarized as rupture information, and the obtained rupture information is output as the output of the time domain rupture sound detecting means 72 to the rupture test unit 73. 74 and the burst information selection unit 75.

  The number-of-detections calculation unit 53 of the bursting processing means 73 counts the number of times that the burst sound has been continuously detected based on the given burst true / false value of each frame, and the obtained number of detections is the centroid time calculation unit. 62.

  In the non-negative section 42 of the bursting processing means 73, each sample of the given decoded speech is converted into a non-negative value, and the obtained non-negative value signal is supplied to the centroid time calculation section 62.

  In the center-of-gravity time calculation unit 62 of the bursting processing means 73, based on the given non-negative value signal and the number of detections in which the bursting sound is continuously detected, the frames within the frames in which the bursting sound is continuously detected are detected. Centroid time is calculated, and the obtained centroid time is given to the burst information selection unit 75.

  The rupture test unit 74 corrects the rupture truth value based on the given rupture truth value and the rupture information to generate a corrected rupture truth value. The obtained corrected rupture truth value is the amplitude modulation unit 76. Given to. The corrected burst truth value is true if the burst power ratio of the burst information is equal to or greater than a predetermined threshold, and otherwise, based on the burst true value, the same as the burst verification unit 32 of the third embodiment. Is set.

  The rupture information selection unit 75 selects either the centroid time from the centroid time calculation unit 62 or the rupture information based on the rupture information from the time domain plosive sound detection means 72. At this time, if the burst power ratio of the burst information is greater than a predetermined threshold, the burst information is selected, and this burst information is output to the amplitude modulator 76 as weight coefficient design information. If the burst power ratio is smaller than the threshold, the centroid time Is selected, and this barycentric time is output to the amplitude modulation section 76 as weight coefficient design information.

  The amplitude modulation unit 76 designs the weighting factor based on the given weighting factor design information, and if the corrected burst true / false value given from the burst validation unit 74 is true, If the correction burst true / false value is false, the decoded speech is passed through as it is when the amplitude modulation is performed by multiplying the weight coefficient, and the improved speech is obtained, and the obtained improved speech is output as the output of the burst processing means 73. .

(H-3) Effect of Seventh Embodiment As described above, according to the seventh embodiment, a plosive sound in which a characteristic acoustic phenomenon is impaired in the decoded speech of the MBE speech coding scheme. Since it can be more appropriately ruptured, it is possible to provide the user with a voice that improves the clarity of the decoded voice and improves the listening comfort.

(I) Other Embodiments In the above-described embodiments, various modified embodiments have been referred to, but further modified embodiments as exemplified below can be given.

  In each of the above embodiments, one method for improving the quality of decoded speech from the MBE decoding means has been described. However, a configuration that can handle a plurality of improvement methods is provided so that the user can select an improvement method. May be.

  Further, instead of selecting from a plurality of improvement methods, the user may be able to select whether or not to apply the improvement method. This selection may be made automatically instead of by the user. For example, a characteristic value such as power and an average value of LPC coefficients for each hour is calculated for decoded speech, and the improvement method for the decoded speech described in each of the above embodiments is applied by comparing the calculated characteristic value with a threshold value. Whether or not to do so may be determined.

  In each of the above embodiments, the case where speech is decoded has been described. However, in the case of MBE encoding to which sound can be applied, the technical idea of the present invention can be applied to sound decoding. It is assumed that the term “sound” described in the claims includes “sound” in such a case.

  Although not mentioned in the description of each of the above embodiments, the method of mounting the elements constituting the speech decoding apparatus on the device or chip is arbitrary. For example, the MBE decoding means 12 is realized by an IC chip, and the plosive detection means (including the frequency domain plosive detection means and the time domain plosive detection means) and the bursting processing means of the above embodiments are executed by the CPU. It may be configured as software. Further, the explosive sound detecting means (including the frequency domain explosive sound detecting means and the time domain explosive sound detecting means) and the bursting processing means of each of the above embodiments may be integrated into an IC chip. The speech decoding apparatus according to each of the above embodiments may be mounted on a communication device connected to a digital wireless device or a wired line.

  1A, 1B, 1C, 1D, 1E, 1F, 1G ... speech decoding device, 11 ... receiving means, 12 ... MBE decoding means, 13, 21 ... plosive detecting means, 14, 31, 41, 51, 61, 77 ... bursting processing means, 71 ... frequency domain burst sound detecting means, 72 ... time domain burst sound detecting means.

Claims (22)

In a speech decoding apparatus for decoding digitally encoded information encoded according to an MBE speech encoding method,
MBE decoding means for decoding the digital audio encoding information to generate decoded audio;
A burst sound detecting means for detecting a burst sound of the decoded speech;
A speech decoding apparatus comprising: bursting processing means for bursting the detected burst sound. The rupture treatment means is:
When the plosive sound detection means determines that the processing frame is a plosive sound, the decoded sound is multiplied by a predetermined weighting factor and output,
The speech decoding apparatus according to claim 1, wherein the decoded speech is output as it is when the plosive detection means determines that the processing frame is not a plosive. The weighting factor is
Consists of three states: silence, rupture start, and rupture late
The silent part has a value smaller than 0 dB,
From the value of the silent part at the beginning of the rupture part, the value increases to a value larger than 0 dB,
The speech decoding apparatus according to claim 2, wherein in the latter half of the rupture part, the value decreases from a value larger than 0 dB at the rupture part start end to 0 dB. The plosive detection means is
Power is calculated at a cycle shorter than the processing cycle of the MBE decoding means,
Divide the obtained short cycle power by the reference power determined by the predetermined rule,
The speech decoding apparatus according to claim 1, wherein if the obtained power ratio is equal to or greater than a predetermined threshold, it is determined that the sound is a plosive sound. The predetermined rule is
For the target time for calculating the power ratio,
If the target time is the beginning of the processing frame, the short period power of the target time is the reference power,
The minimum value of short cycle power from the start end in the processing frame to immediately before the target time is used as the reference power if the target time is after the start end in the processing frame. The speech decoding apparatus described. The predetermined rule is
The voice according to claim 4, wherein the reference power is a minimum value of a short cycle power from a predetermined time before the target time to immediately before the target time with respect to the target time for calculating the power ratio. Decryption device. The plosive detection means is
Frequency analysis of the decoded speech from the MBE decoding means,
The speech decoding apparatus according to claim 1, wherein pattern recognition is performed on the obtained power spectrum to determine whether or not the sound is a plosive sound. The rupture treatment means is:
8. The speech decoding according to claim 7, wherein when the plosive sound is detected continuously by the plosive sound detection means, the weighting coefficient is multiplied only once based on only the first detected frame. Device. The rupture treatment means is:
The decoded speech is converted to a non-negative value for each sample,
Calculate the center of gravity in the frame related to the sum of the obtained non-negative signal,
The speech decoding apparatus according to claim 7, wherein the weighted coefficient is translated in the time direction based on the obtained barycentric time and then multiplied by the decoded speech. The rupture treatment means is:
9. The speech decoding apparatus according to claim 8, wherein the weighting coefficient is translated in the time direction and then multiplied by the decoded speech based on the number of times that the burst sound is continuously detected by the burst sound detecting means. . The rupture treatment means is:
The decoded speech is converted to a non-negative value for each sample,
Based on the number of times the plosive detection means has continuously detected a plosive and the centroid time obtained by calculating the centroid related to the sum of the non-negative signals in the same number of frames, the weighting factor is set in the time direction. The speech decoding apparatus according to claim 8, wherein the decoded speech is multiplied after being translated. In a speech decoding apparatus for decoding digitally encoded information encoded according to an MBE speech encoding method,
MBE decoding means for decoding the digital audio encoding information to generate decoded audio;
A frequency domain burst sound detecting means for detecting a burst sound of the decoded speech in the frequency domain;
Calculates the center of gravity of the sum of the non-negative signals obtained by converting the decoded speech in the same number of frames as the number of consecutive bursts detected by the frequency domain plosive detection means into a non-negative value for each sample. Centroid time calculating means for calculating the centroid time,
A time domain burst sound detecting means for detecting the burst sound of the decoded speech in the time domain;
Burst information selecting means for selecting the centroid time and the burst information obtained from the time domain burst sound detecting means based on the burst information;
Based on the determination result of the frequency domain plosive sound detection means and the rupture information, a burst test means for re-determining whether or not a plosive sound,
Based on the weight coefficient design information obtained from the burst information selection unit, the predetermined weighting factor is redesigned based on the frame determined to be a plosive sound by the burst verification unit, and the decoded speech And a bursting processing unit for multiplying the weight coefficient by the speech decoding device. The predetermined weight coefficient designed in advance is
Consists of three states: silence, rupture start, and rupture late
The silent part has a value smaller than 0 dB,
From the value of the silent part at the beginning of the rupture part, the value increases to a value larger than 0 dB,
The speech decoding apparatus according to claim 12, wherein in the latter half of the rupture part, the value decreases from a value larger than 0 dB of the rupture part start end to 0 dB. The frequency domain plosive detection means is
Frequency analysis of the MBE decoding means,
The speech decoding apparatus according to claim 12 or 13, wherein pattern recognition is performed on the obtained power spectrum to determine whether or not it is a plosive sound. The time domain plosive detection means is
Power is calculated at a cycle shorter than the processing cycle of the MBE decoding means,
Divide the obtained short cycle power by the reference power determined by the predetermined rule,
The speech decoding apparatus according to any one of claims 12 to 14, wherein if the obtained power ratio is equal to or greater than a predetermined closed value, it is determined that the sound is a plosive sound. The predetermined rule is
For the target time for calculating the power ratio,
If the target time is the beginning of the processing frame, the short period power of the target time is the reference power,
16. If the target time is after the base end in the processing frame, the minimum value of the short cycle power from the start end in the processing frame to immediately before the target time is set as the reference power. The speech decoding apparatus according to 1. The predetermined rule is
The voice according to claim 15, wherein the reference power is a minimum value of a short period power from a predetermined time before the target time to immediately before the target time with respect to the target time for calculating the power ratio. Decryption device. In a speech decoding method for decoding digitally encoded information encoded according to an MBE-based speech encoding scheme,
MBE decoding means decodes the digital audio encoding information to generate decoded audio,
The plosive detection means detects a plosive of the decoded speech,
A speech decoding method, wherein the bursting processing means bursts the detected burst sound. In a speech decoding method for decoding digitally encoded information encoded according to an MBE-based speech encoding scheme,
MBE decoding means decodes the digital audio encoding information to generate decoded audio,
The frequency domain burst sound detecting means detects the burst sound of the decoded speech in the frequency domain,
A non-negative signal obtained by converting the decoded speech in the same number of frames as the number of times that the burst sound is continuously detected by the frequency domain plosive detection means into a non-negative value for each sample by the centroid time calculation means. Calculate the center of gravity for the sum of
The time domain plosive detection means detects the plosive sound of the decoded speech in the time domain,
The rupture information selection means selects the centroid time and the rupture information obtained from the time domain rupture sound detection means based on the rupture information,
Based on the determination result of the frequency domain plosive detection means and the burst information, the burst test means re-determines whether or not it is a plosive sound,
Redesign a predetermined weighting factor designed in advance based on the weighting factor design information obtained from the bursting information selection unit with reference to the frame determined by the bursting processing unit as a burst sound in the bursting verification unit Then, the speech decoding method, wherein the decoded speech is multiplied by the weight coefficient. In a speech decoding program for decoding digitally encoded information encoded according to the MBE speech encoding method,
Computer
MBE decoding means for decoding the digital audio encoding information to generate decoded audio;
A burst sound detecting means for detecting a burst sound of the decoded speech;
A speech decoding program that functions as bursting processing means for bursting the detected burst sound. In a speech decoding program for decoding digitally encoded information encoded according to the MBE speech encoding method,
Computer
MBE decoding means for decoding the digital audio encoding information to generate decoded audio;
A frequency domain burst sound detecting means for detecting a burst sound of the decoded speech in the frequency domain;
Calculates the center of gravity of the sum of the non-negative signals obtained by converting the decoded speech in the same number of frames as the number of consecutive bursts detected by the frequency domain plosive detection means into a non-negative value for each sample. Centroid time calculating means for calculating the centroid time,
A time domain burst sound detecting means for detecting the burst sound of the decoded speech in the time domain;
Burst information selecting means for selecting the centroid time and the burst information obtained from the time domain burst sound detecting means based on the burst information;
Based on the determination result of the frequency domain plosive sound detection means and the rupture information, a burst test means for re-determining whether or not a plosive sound,
Based on the weight coefficient design information obtained from the burst information selection unit, the predetermined weighting factor is redesigned based on the frame determined to be a plosive sound by the burst verification unit, and the decoded speech A speech decoding program that functions as bursting processing means that multiplies the weight coefficient by the weight coefficient.   A communication device comprising the speech decoding device according to claim 1.

Images