JP3660676B2 - Speech coding apparatus and method - Google Patents

Description

  The present invention relates to a technique for improving coding quality and improving transmission quality of noise-superimposed speech when background noise such as outdoors is superimposed in a digital mobile radio communication system such as an automobile phone or a cellular phone.

  In recent years, digital mobile radio communication systems such as car phones and mobile phones have become widespread due to improvements in communication technology. Accordingly, an audio signal processing apparatus that compresses an audio signal with high efficiency has been demanded.

  In a digital mobile radio communication system, in order to effectively use a radio frequency, it is desirable to encode an audio signal in a 4 kHz band at a bit rate of about 4 to 8 kbps. A CELP system is known as a speech encoding system corresponding to this.

  In the CELP method, a speech signal is analyzed based on a linear prediction theory, and parameters representing frequency characteristics are extracted. At the same time, the driving sound source signal is encoded in a waveform by vector quantization. On the receiving side, the encoded speech transmitted on the transmission path is decoded by a procedure reverse to that on the transmitting side.

  By the way, in the CELP system, since the audio signal is compressed to a low bit rate and at the same time the reproduction audio quality is maintained, encoding (band compression) based on the audio generation model is performed, so that background noise is superimposed. When an audio signal is encoded, an unnatural reproduction sound may be output. In other words, in the conventional system, encoding processing is performed on the assumption that noise signals having properties different from those of speech have properties similar to those of speech. For this reason, a signal with only background noise is subjected to an encoding process even though there is no frequency correlation, and is reproduced as unnatural sound.

  Conventionally, when encoding speech, an adaptive codebook is referred to based on the speech waveform to detect index information of a similar waveform pattern. However, when noise is superimposed on speech, there is no waveform pattern similar to the adaptive codebook, and a waveform pattern that is not very similar has to be selected, which is unnatural when this is decoded. There is a problem of being output as audio.

  Here, taking the air-conditioning sound as an example of the background noise, the spectrum of the original sound of the air-conditioning sound shows a substantially flat characteristic as shown in FIG. On the other hand, as shown in FIG. 14, the reproduced sound of the air-conditioning sound has a spectrum envelope peak that varies from frame to frame. The inventor of the present invention has focused on the fact that fluctuations in the spectral envelope cause auditory unnaturalness, and have investigated the cause of the spectral fluctuations. In other words, in the conventional decoder, the excitation signal is generated from the adaptive codebook and the noise codebook, and the excitation signal is decoded through the synthesis filter. It was analyzed whether it was due to signal generation processing or a synthesis filter. As a result, no temporal variation was observed in the spectrum of the excitation signal. On the other hand, in the case of the synthesis filter, fluctuations as shown in FIG. 15 appeared.

Therefore, the present invention has been made in view of the above problems, and distinguishes between a signal containing only noise and a signal containing speech, and differentiates the encoding process or the decoding process of a signal containing only noise. Therefore, it is a first object to provide an apparatus that outputs an acoustically natural reproduced sound by suppressing the characteristics of a synthesis filter when only noise is decoded.

  Another object of the present invention is to provide a technology capable of performing high-quality encoding processing by preventing the influence of noise when encoding a signal in which noise is superimposed on speech.

The present invention employs the following means in order to solve the above problems. This will be described with reference to the drawings.
(Means for solving the first problem)
First, means for solving the first problem will be described with reference to the principle diagram of FIG.

  The speech decoding system of the present invention includes a noise weight section detecting means 1, speech decoding means 2, noise decoding means 3, and noise control means 4. The noise superimposition section detection means 1 has a function of monitoring a signal encoded on the transmission side and identifying whether the section is a voice section including speech or a noise section including only noise. For example, the noise superimposition section detecting means 1 may detect the power of the signal from the encoded signal and determine whether or not the power is equal to or higher than a preset threshold value. That is, if the power of the encoded signal is equal to or greater than the threshold, it may be determined as a speech interval, and if the power of the encoded signal is less than the threshold, it may be determined as a noise interval. Further, the gain of the encoded signal may be used instead of the power.

  The speech decoding means 2 has a function of decoding the encoded signal into a waveform signal when the noise superimposing section detecting means 1 discriminates the encoded signal in the speech section. Specifically, the code book 3a for registering the waveform pattern for each index information, the driving sound source 3b for exciting the waveform pattern read from the code book 3a, and the excitation signal output from the driving sound source 3b are filtered. And a synthesis filter 3c that performs processing.

  The noise decoding unit 3 has a function of decoding the encoded signal into a waveform signal when the noise superimposing interval detecting unit 1 determines the encoded signal in the noise interval. Specifically, the code book 3a for registering the waveform pattern for each index information, the driving sound source 3b for exciting the waveform pattern read from the code book 3a, and the excitation signal output from the driving sound source 3b are filtered. And a synthesis filter 3c that performs processing.

  The noise control unit 4 has a function of controlling the filter coefficient of the synthesis filter 3c of the noise decoding unit 3 to suppress the frequency characteristic of noise when the noise superimposition section detection unit 1 determines the encoded signal in the noise section. Have. Specifically, the noise control means 4 has a function of determining a positive value of 1 or less to be multiplied by the filter coefficient.

  Further, when the post filter 9 for amplifying the amplitude of the decoded signal output from the synthesis filter 3c is provided at the subsequent stage of the noise decoding unit 3 and the speech decoding unit 2, the post filter 9 is configured to perform noise decoding. It has a function of passing the decoded signal of the noise section output from the means 3 as it is.

  Next, the speech coding system of the present invention will be described. The speech encoding system of the present invention includes a noise superimposition section detecting means 5, speech encoding means 6, noise encoding means 7, and control information generating means 8.

The noise superimposition section detection means 5 monitors the encoded signal encoded on the transmission side, and identifies whether the signal is a speech section signal including speech or a noise section signal including only noise. have. Specifically, by analyzing the waveform characteristics included in the encoded signal, it is possible to determine whether the power of the signal is less than the threshold or whether the gain of the encoding is less than the threshold. A section may be identified.

  Further, the speech encoding means 6 has a function of encoding into index information for specifying the waveform of this section when the noise superimposition section detecting means 5 discriminates the speech section. Specifically, the speech encoding means 6 includes a code book for registering a waveform pattern for each index information.

  The noise encoding unit 7 has a function of encoding, when the noise superimposing section detecting unit 5 determines a noise section, the index information for specifying the waveform of the section. Similar to the voice encoding unit 6, the encoding unit includes a code book for registering a waveform pattern for each index information.

  The control information generation means 8 generates control information for decoding processing of the noise section when the noise superimposition section detection means 5 discriminates the noise section, and adds this control information to the encoded signal of the noise section. And has a function of transmitting to the receiving side. Specifically, it is information for controlling the filter coefficient of the synthesis filter used on the decoding side based on the waveform characteristics of noise. For example, a positive value of 1 or less to be multiplied by the filter coefficient is used as the control information.

(Means for solving the second problem)
Next, means for solving the second problem will be described with reference to the principle diagram of FIG.
This means is applied to an encoding system, and includes a noise superposition section detecting means 10, an inverse filter means 11, a noise removing means 12, a pitch period detecting means 13, and a speech encoding means 14.

  The noise superimposition section detection means 10 monitors a signal input from the transmitter, and identifies a speech section including only speech, a noise section including only noise, and a noise superimposition section where noise is superimposed on speech. have.

  When the noise superimposition section detection means 10 discriminates the noise superimposition section, the inverse filter means 11 performs linear prediction analysis on the noise superimposition section to obtain a linear prediction coefficient, and performs an inverse filtering process using the linear prediction coefficient as a filter coefficient. It has a function to apply. The prediction residual signal output from the inverse filter unit 11 is input to the noise removing unit 12.

The noise removing unit 12 has a function of removing a noise part from the prediction residual signal. For example, a low-pass filter can be used as the noise removing unit 12.
The pitch period detecting unit 13 has a function of obtaining an autocorrelation function of the residual signal output from the noise removing unit 12 and detecting a pitch period at which the autocorrelation function becomes a maximum value. That is, the prediction residual signal is shifted by a specific period, and a specific period at which the correlation between each prediction residual signal and the original prediction residual signal is maximized is detected as a pitch period.

The voice encoding unit 14 has a function of encoding the waveform of the noise superimposed section based on the pitch period detected by the pitch period detection unit 13.
A system for solving the first problem of the present invention will be described.

(Operation of the system that solves the first problem)
In the speech decoding system of the present invention, when an encoded signal encoded on the transmission side is received, the noise superimposition section detecting means 1 is a coded signal in a noise section including only noise or a speech section including speech. It is determined whether the signal is an encoded signal.

Here, if the encoded signal is an encoded signal of a speech section, this encoded signal is input to the speech decoding means 2. The voice encoding unit 2 encodes the encoded signal into a waveform signal.
If the encoded signal is an encoded signal in a noise interval, this encoded signal is input to the noise decoding means 3. The noise decoding means 3 detects a waveform pattern corresponding to the index information from the codebook 3a, and excites this waveform pattern via the driving sound source 3b. The excitation signal is input to the synthesis filter 3c. At the same time, the noise control means 4 multiplies the filter coefficient of the synthesis filter 3c by a positive value of 1 or less and notifies the synthesis filter 3c. The synthesis filter 3c performs a filtering process on the excitation signal based on the filter coefficient notified from the noise control unit 4, and outputs a decoded signal. As a result, the waveform of the noise section is reproduced without unnaturally enhancing the frequency characteristics.

  Further, the noise decoding unit 3 may process the gain of the noise section as “0”. Further, when the post filter 9 is provided in the subsequent stage of the synthesis filter 3c, the post filter 9 passes the noise waveform peak output from the synthesis filter 3c without being emphasized (no processing is performed).

  Next, in the speech coding system of the present invention, when a signal is input from the transmitter, the noise superimposition section detecting means 5 is a signal in a speech section including speech or includes only noise. Identify whether the signal is in the noise interval.

  Here, if the input signal is a speech section signal, the speech encoding means 6 discriminates a waveform pattern similar to the speech section waveform, encodes this waveform pattern into index information for identification, and transmits it to the receiving side. To do.

  If the input signal is a signal in the noise interval, the noise encoding means 7 discriminates a waveform pattern similar to the waveform in the noise interval, and encodes this waveform pattern into index information for specifying the waveform pattern. At the same time, the control information generation means 8 generates control information related to the noise section decoding process and adds it to the index information. Specifically, the control information generation means 8 performs linear prediction analysis on the input signal in the noise interval to determine the frequency characteristic, multiplies the obtained filter coefficient by a positive value of 1 or less, and performs synthesis used on the receiving side. Determine the filter coefficients of the filter. Then, this filter coefficient is transmitted as control information to the receiving side together with index information.

Hereinafter, a system for solving the second problem of the present invention will be described.
(Operation of the system that solves the second problem)
In the speech coding system of the present invention, the noise superimposition section discriminating means 10 monitors the signal input from the transmitter, and is a speech section including only speech, a noise section including only noise, or It is determined whether or not it is a noise superimposition section in which noise is superimposed on speech.

  Here, when the noise superimposition section is determined, the inverse filter unit 11 obtains a prediction coefficient of the noise superimposition section, performs a filtering process using the prediction coefficient as a filter coefficient, and outputs a prediction residual signal. This prediction residual signal is input to the noise removing unit 12 and the noise portion is removed.

  The prediction residual signal from which the noise part has been removed by the noise removing unit 12 is input to the pitch period detecting unit 13. The pitch period detection means 13 obtains an autocorrelation function of the prediction residual signal and detects a pitch period at which the autocorrelation function becomes a maximum value.

Then, the voice encoding unit 14 discriminates a waveform pattern similar to the waveform of the noise superimposition section based on the pitch period detected by the pitch period detection unit 13, and encodes the waveform pattern into index information for specifying the waveform pattern. As a result, the audio signal can be encoded without being affected by noise.

According to the present invention, it is possible to prevent an unnatural frequency characteristic from being added to noise with a small change in frequency characteristic, and to reduce the uncomfortable feeling during reproduction.
Furthermore, when coding a signal in which noise is superimposed on speech, high-quality coding can be performed by removing a noise component and detecting an accurate pitch period.

  Therefore, according to the present invention, it is possible to contribute to an improvement in voice quality of a mobile communication system such as a mobile phone or a car phone.

Embodiments of the present invention will be described with reference to the drawings.
<Example 1>
A first embodiment of the present invention will be described with reference to the drawings.

  FIG. 3 is a block diagram illustrating the configuration of the speech decoding system according to the first embodiment. The speech encoding system according to the first embodiment includes a noise superimposition detection / determination unit 1 as noise superimposing section detection means, a speech decoder A (2) as speech encoding means, and a speech decoder B as noise decoding means. (3) and a received code separating unit 15 are provided.

  Note that the speech decoder A (2) and the speech decoder B (3) adopt the CELP method as a decoding method. The reception code separation unit 15 has a function of separating the encoded signal received from the transmission side into power information, index information, and synthesis filter coefficients.

  The noise superimposition detection determination unit 1 compares the power information separated by the received code separation unit 15 with a preset threshold value. If the power information is equal to or greater than the threshold value, the encoded signal is determined to be a speech section. If it is less than the threshold, it has a function of determining the encoded signal as a noise interval. Furthermore, the noise superimposition detection / determination unit 1 has a function of causing the encoded signal in the speech section to be input to the speech decoder A (2) and the encoded signal in the noise section to be input to the speech decoder B (3). ing.

The audio decoder A (2) decodes the encoded signal in the audio section. Specifically, it has the same configuration and function as a conventional CELP decoder and will not be described.
The audio decoder B (3) decodes the encoded signal in the noise section. Here, FIG. 4 shows an internal configuration and a peripheral configuration of the speech decoder B (3). In the figure, the speech decoder B (3) includes an adaptive codebook 30a, a noise codebook 31a, a driving sound source 3b, and a synthesis filter 3c. The synthesis filter 3c is connected to an LPC coefficient correction unit 4 as noise control means of the present invention.

  The adaptive codebook 30a registers the waveform pattern of the waveform signal having periodicity and the index information, and has a function of updating the waveform pattern with the decoded waveform signal.

  The noise codebook 31a is for registering the waveform pattern and index information of a waveform signal having no periodicity. The adaptive codebook 30a and the noise codebook 31a define the amplification factor (gain) of the waveform pattern read from each, and the driving sound source 3b is read from the adaptive codebook 30a and the noise codebook 31a. It has a function of exciting the waveform pattern according to each gain.

The synthesis filter 3c performs a filtering process on the excitation signal output from the driving sound source 3b and decodes it into a waveform signal. The filter coefficient of the synthesis filter 3c is determined on the transmission side. That is, the transmission side performs linear prediction analysis on the original waveform signal, calculates a linear prediction coefficient, and transmits this linear prediction coefficient as a filter coefficient to the reception side. Thereby, the speech decoder B (3) detects the filter coefficient from the encoded signal, and uses this filter coefficient as the filter coefficient of the synthesis filter 3c.

  The LPC coefficient correction unit 4 has a function of correcting the filter coefficient of the synthesis filter 3c in response to the determination result of the noise superimposition detection determination unit 1. Specifically, as shown in the following equation, the filter coefficient of the synthesis filter 3c is multiplied by a positive number of 1 or less to correct the filter coefficient.

α′i = gi × αi (0.0 <g ≦ 1.0)
As a result, the frequency characteristic of the synthesis filter 3c can be made substantially flat (see FIG. 12).

Hereinafter, the operation of the speech decoding system will be described. In the speech decoding system, the reception code separation unit 15 receives the encoded signal encoded on the transmission side.
The reception code separation unit 15 separates the encoded signal into power information, index information, and filter coefficients, and inputs the power information to the noise superimposition detection determination unit 1. The noise superimposition detection determination unit 1 determines whether the power information is greater than or equal to a threshold value or less than the threshold value. If the power information is equal to or greater than the threshold value, the noise superimposition detection / determination unit 1 determines that the encoded signal is a speech section signal, and uses the power information, index information, and filter coefficient separated by the received code separation unit 15 as speech. Input to the decoder A (2). The voice decoder A (2) decodes the encoded signal into a voice waveform based on these pieces of information.

  On the other hand, when the power information is less than the threshold value, the noise superimposition detection / determination unit 1 determines that the encoded signal is a signal in the noise section, and uses the power information and index information separated by the received code separation unit 15 as speech decoder B. At the same time as inputting to (3), the filter coefficient is notified to the LPC coefficient correction unit 4.

  The speech decoder B (3) searches the adaptive codebook 30a or the noise codebook 31a based on the index information, and detects a corresponding waveform pattern. Then, the driving sound source 3b excites the waveform pattern according to the gain of each codebook, and inputs the excitation signal to the synthesis filter 3c.

Here, the LPC coefficient correction unit 4 corrects the filter coefficient by multiplying the filter coefficient by a positive value of 1 or less. Then, the corrected filter coefficient is notified to the synthesis filter 3c.
The synthesis filter 3c performs a filtering process on the excitation signal output from the driving sound source 3b according to the filter coefficient notified from the LPC coefficient correction unit 4, and decodes the excitation signal into a noise waveform. As described above, according to the first embodiment, when the signal in the noise section is encoded, the spectrum of the synthesis filter 3c can be made substantially flat by controlling the filter coefficient, and the noise waveform characteristic can be improved. It is possible to prevent unnatural emphasis and to suppress the reproduction of noise that is audibly annoying. Therefore, it is possible to improve the voice quality of portable mobile communication such as a mobile phone or a car phone.

<Example 2>
In the second embodiment, an example in which the system of the present invention is applied to an encoder will be described. FIG. 5 is a schematic configuration diagram of a speech encoding system.

In the figure, the speech coding system includes a speech encoder A (6), a speech encoder B (7), and a noise superimposition detection / determination unit 5. The noise superimposition detection / determination unit 5 detects the power of the waveform signal input from the transmitter. If the power is greater than or equal to a threshold value, the noise superimposition detection / determination unit 5 determines that the waveform signal is in a voice section including speech. It has a function to determine the waveform signal of the noise section including only the. The noise superimposition detection / determination unit 5 has a function of inputting the waveform signal in the speech section to the speech encoder A (6) and inputting the waveform signal in the noise section to the speech encoder B (7).

  The speech encoder A (6) has a function of encoding a waveform signal in a speech section and is a conventional CELP encoder. The speech encoder B (7) has a function of encoding a waveform signal in a noise interval.

  Here, FIG. 6 shows an internal configuration and a peripheral configuration of the speech encoder B (7). In the figure, a speech encoder B (7) includes an adaptive codebook 70a, a noise codebook 71a, a driving sound source 7b, a synthesis filter 7c, an LPC analysis unit 7e, and an error minimization unit 7d.

  The adaptive codebook 70a registers a waveform pattern having a periodicity and index information for specifying each waveform pattern. The noise codebook 71a registers a waveform pattern of a waveform having no periodicity and index information for identifying each waveform pattern.

The driving sound source 7b has a function of exciting the waveform pattern detected from the adaptive codebook 70a and the waveform pattern detected from the noise codebook 71a according to the gain of each codebook.
The synthesis filter 7c has a function of performing a filtering process using a linear prediction coefficient of a waveform signal in a noise interval as a filter coefficient. The error minimizing unit 7d compares the waveform signal output from the synthesis filter 7c with the waveform of the input noise signal, optimizes the index information and the amplification factor (gain) of the waveform pattern, and generates the noise codebook 71a. It has a function to update the contents.

The LPC analysis unit 7e has a function of calculating a linear prediction coefficient by performing linear prediction analysis on the input waveform, and inputting the linear prediction coefficient as a filter coefficient to the synthesis filter 7c.
Furthermore, the code encoder 16 and the LPC coefficient correction unit 8 are connected to the speech encoder B (7). The code transmission unit 16 has a function of transmitting the power information, index information, and filter coefficient encoded by the speech encoder B (7) to the reception side.

  The LPC coefficient correction unit 8 has the same function as that of the first embodiment, and has a function of correcting the filter coefficient of the synthesis filter 7c used when decoding the encoded signal in the noise section. ing. Specifically, correction is performed by multiplying the filter coefficient by a positive value of 1 or less. In response to this, the code transmission unit 16 transmits the filter coefficient corrected by the LPC coefficient correction unit 8 together with other encoded signals.

  The operation of the speech encoding system in the second embodiment will be described below. When a waveform signal is input from the transmitter, the noise superimposition detection / determination unit 5 detects the power of the waveform signal and determines whether it is greater than or less than a threshold value. Here, if the power of the waveform signal is equal to or greater than the threshold value, the noise superimposition detection / determination unit 5 determines that the waveform signal is a waveform signal in the speech section, and inputs this waveform signal to the speech encoder A (6).

  Speech encoder A (6) encodes waveform information into index information, power information, and filter coefficients using a codebook, and transmits the encoded information to the receiving side. If the power of the input waveform is less than the threshold value, the noise superimposition detection / determination unit 5 determines that the waveform signal is a waveform signal in the noise interval, and inputs this waveform signal to the speech encoder B (7). .

  The speech encoder B (7) has a function of searching the adaptive codebook 70a and the noise codebook 71a based on the waveform of the noise section and detecting a similar waveform pattern. Furthermore, the speech encoder B (7) inputs the waveform pattern read from the adaptive codebook 70a or the noise codebook 71a to the driving sound source 7b.

  The driving sound source 7b excites the waveform pattern and inputs it to the synthesis filter 7c. Here, the LPC analysis unit 7e performs linear prediction analysis on the input waveform signal and calculates a linear prediction coefficient. Then, the LPC analysis unit 7e notifies the synthesis filter 7c of the linear prediction coefficient.

  The synthesis filter 7c performs a filtering process using the linear prediction coefficient as a filter coefficient on the excitation signal input from the driving sound source 7b. The error minimizing unit 7d compares the decoded signal output from the synthesis filter 7c with the input waveform signal, and applies the optimum index information and waveform pattern gain to minimize the error between the two. Notification is made to the book 70a and the noise codebook 71a. Each codebook updates the registration contents and gain based on the index information and gain notified from the error minimizing unit 7d, and notifies the code transmitting unit 16 of the updated index information. Further, the LPC coefficient correction unit 8 corrects the filter coefficient by multiplying the linear prediction coefficient (filter coefficient) calculated by the LPC analysis unit 8 by a positive value of 1 or less. Then, the LPC coefficient correction unit 8 notifies the code transmission unit 16 of the corrected filter coefficient.

The code transmission unit 16 notifies the reception side of the index information and power information notified from the speech encoder B (7) and the filter coefficient notified from the LPC coefficient correction unit 8.
As a result, on the receiving side, the decoding process is performed using the corrected filter coefficient, so that the spectrum of the synthesis filter can be made flat, and the waveform in the noise section is prevented from being unnaturally decoded. can do.

  As described above, according to the second embodiment, when performing the decoding process of the noise section, the spectrum of the synthesis filter can be made to have a flat characteristic, and the frequency characteristic of the noise section is not unnatural and can be heard acoustically. Unpleasant noise can be suppressed.

<Example 3>
A third embodiment of the present invention will be described below with reference to the drawings. FIG. 7 shows the internal configuration of the speech encoder B according to the third embodiment.

  In the figure, the speech encoder B (7) is different from the speech encoder B (7) of the second embodiment described above in that it includes an adaptive codebook 70a, a noise codebook 71a, a driving sound source 7b, a synthesis filter 7c, and LPC analysis. A section 7e and an error minimizing section 7d are provided. Furthermore, the code transmitter 16 is connected to the speech encoder B (7).

  The code transmitting unit 16 has a function of transmitting “0” as index information of the adaptive codebook 70a when transmitting the encoded signal in the noise section. Other configurations and functions are the same as those of the second embodiment described above, and a description thereof will be omitted.

  FIG. 8 is a block diagram showing a configuration of speech decoder B (3) corresponding to speech encoder B (7) in FIG. The speech decoder B (3) is provided with an adaptive codebook 30a, a noise codebook 31a, a driving sound source 3b, a synthesis filter 3c, and an adaptive postfilter 17 with respect to the configuration of the first embodiment.

The adaptive post filter 17 has a function of amplifying the amplitude value without changing the waveform period. When the adaptive codebook 30a receives the index information “0” of the adaptive codebook 30a from the transmission side, the adaptive codebook 30a sets the gain of the adaptive codebook 30a to “0”. Thus, when a waveform signal in the noise section is input, the noise codebook 31a is searched based on the index information of the noise codebook 31a, and the corresponding waveform pattern is read out. Further, when the waveform signal in the noise interval is input, the adaptive post filter 17 passes the waveform signal without performing any processing.

  According to the third embodiment, an unnatural periodicity is added to a noise signal having a flat characteristic having no periodicity during decoding processing by encoding and decoding a noise waveform having no periodicity using a noise codebook. Without being decoded into an acoustically natural waveform signal.

<Example 4>
FIG. 9 shows the configuration of the encoder B in the fourth embodiment. The encoder B (7) includes an adaptive codebook analyzer 18, a noise codebook analyzer 19, a drive excitation generator 20, and an open loop pitch analyzer 21.

  The adaptive codebook analysis unit 18 has a function of performing a filtering process on the waveform signal detected from the noise codebook 71a by the long-term prediction synthesis filter 72 and performing a closed loop process for calculating the pitch period of the waveform signal. (See FIG. 10).

  On the other hand, the open loop pitch analysis unit 21 is activated when encoding a noise superimposed section in which a noise waveform is superimposed on a speech waveform, and includes a short-term prediction inverse filter 11, a low-pass filter LPF 12, an autocorrelation detection unit 13b, A correlation maximum value detection unit 13c and a delay unit 13a are provided (see FIG. 11).

  The short-term prediction inverse filter 11 has a function of performing an inverse filtering process using a linear prediction coefficient of a waveform signal as a filter coefficient and outputting a prediction residual signal. The low pass filter LPF 12 has a function of removing the waveform of the noise portion from the prediction residual signal.

  The delay unit 13a has a function of shifting the period of the prediction residual signal by a specific period. The autocorrelation detection unit 13b has a function of detecting a correlation value between the original prediction residual signal and the prediction residual signal shifted by a specific period by the delay unit 13a.

  The correlation maximum value detection unit 13c has a function of detecting the delay amount (cycle) having the largest correlation by the delay unit 13a being shifted by a specific period. This delay amount is notified to the driving sound source 7b as a pitch period. Then, the driving sound source 7b excites the waveform pattern read from the applied codebook 70a based on this pitch period.

  As described above, according to the fourth embodiment, it is possible to accurately detect the pitch period of a speech waveform on which noise is superimposed, to perform high-quality encoding processing independent of the presence or absence of noise, and to improve the quality of reproduced speech Can be improved.

Principle of the present invention (1) Principle of the present invention (2) 1 is a schematic configuration diagram of a speech decoding system according to Embodiment 1. Internal configuration block diagram of speech decoder B Schematic configuration diagram of a speech encoding system in Embodiment 2. Internal configuration block diagram of speech encoder B Block diagram of internal configuration of speech encoder B in Embodiment 3 Block diagram of internal configuration of speech decoder B in the third embodiment. Schematic configuration diagram of a speech encoding system in Embodiment 4. Block diagram showing the internal configuration of the adaptive codebook analyzer Internal configuration block diagram of the open loop analysis unit Spectrum showing frequency characteristics of synthesis filter The figure which shows the spectrum of the source sound of the air conditioning sound The figure which shows the spectrum of the reproduction sound of the air conditioning sound Spectrum showing frequency characteristics of synthesis filter

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 .... Noise superimposition section detection means 2 .... Speech decoding means 3 .... Noise decoding means 3a ... Codebook 3b ... Driving sound source 3c ... Synthetic filter 4 .... Noise control means (LPC coefficient correction unit)
5. ・ Noise superimposition section detection means (noise superimposition detection / determination unit)
6 .. Speech encoding means 7.. Noise encoding means 7 b... Driving sound source 7 c... Synthesis filter 7 d .. Error minimizing section 7 e .. LPC analysis section 8 .. Control information generating means (LPC coefficient correcting section)
9..Post filter 10..Noise superimposing section detection means 11..Inverse filter means (short-term prediction inverse filter)
12. Noise removal means (low pass filter LPF)
13. · Pitch period detection means 13a ·· Delay portion 13b ·· Autocorrelation detection portion 13c ·· Maximum correlation value detection portion 14 ·· Voice encoding means 15 ·· Reception code separation portion 16 ·· Code transmission portion 17 ··· Adaptive post filter 18 .... Adaptive codebook analysis unit 19 .... Noise codebook analysis unit 20 .... Driving sound source generation unit 21 ... Open loop pitch analysis unit 30a ... Adaptive code book 31a ... Noise codebook 70a ... Codebook 71a ... Noise codebook 72 ... Long-term prediction synthesis filter

Claims (4)

Detecting means for determining whether the signal input from the transmitter is a signal in a voice section including voice or a signal in a noise section other than the voice section;
A first speech encoder that encodes the signal of the speech interval;
A second speech encoder that encodes the signal in the noise interval ;
Means for calculating a linear prediction coefficient by performing linear prediction analysis on a signal input from the transmitter, and outputting the linear prediction coefficient, provided in the second speech encoder;
Means for correcting the linear prediction coefficient to a linear prediction coefficient such that temporal variation of the frequency characteristic is not seen in the output of the synthesis filter used on the receiving side when the detection means determines a noise interval ;
Means for transmitting the corrected linear prediction coefficient to the receiving side;
Speech coding apparatus comprising: a. Detecting means for determining whether the signal input from the transmitter is a signal in a voice section including voice or a signal in a noise section other than the voice section;
A speech encoder that encodes the signal of the speech section or encodes the signal of the noise section ;
Means for providing a linear prediction coefficient by performing linear prediction analysis on the signal input from the transmitter, and outputting the linear prediction coefficient, provided in the speech encoder;
Means for correcting the linear prediction coefficient to a linear prediction coefficient such that the fluctuation of the spectrum between frames is reduced in the output of the synthesis filter used on the receiving side when the detection means determines a noise interval ;
Means for transmitting the corrected linear prediction coefficient to the receiving side;
Speech coding apparatus comprising: a. Determining whether the signal input from the transmitter is a signal in a speech section including speech or a signal in a noise section other than that,
Encoding the speech interval signal or encoding the noise interval signal;
Linear prediction analysis of the signal input from the transmitter to calculate a linear prediction coefficient, and outputting the linear prediction coefficient;
When encoding the signal in the noise interval, the linear prediction coefficient is used as a synthesis block used on the receiving side.
Correcting to a linear prediction coefficient such that temporal fluctuation of the frequency characteristic is not seen in the output of the filter;
Transmitting the corrected linear prediction coefficient to the receiving side . Determining whether the signal input from the transmitter is a signal in a speech section including speech or a signal in a noise section other than that,
Encoding the speech interval signal or encoding the noise interval signal;
Linear prediction analysis of the signal input from the transmitter to calculate a linear prediction coefficient, and outputting the linear prediction coefficient;
Correcting the linear prediction coefficient to a linear prediction coefficient that reduces the variation in the spectrum between frames at the output of the synthesis filter used on the receiving side when encoding the signal in the noise section;
Transmitting the corrected linear prediction coefficient to the receiving side;
A speech encoding method characterized by comprising:

Images