JP3475772B2 - Audio encoding device and audio decoding device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To make it possible to transmit non-speech signal of high quality without increasing a transmission speed at the time of transmitting a non-speech signal by changing over a part of processing function blocks based on the judgment result of a speech/non-speech discriminator. SOLUTION: A speech/non-speech signal discriminator 102 permanently monitors whether an input signal is a speech signal or a non-speech signal. Based on the judgment result, an operation mode of an encoder 101 is decided by change-over switches 103, 104. An encoding processing function block 105 is optimized so as to be able to compress and encode subject speech signal with a high degree of efficiency. On the other hand, an encoding processing function block 106 is optimized so as to be able to compress and encode non- speech signal, for example, a DTMF signal, with little distortion. On the receiver side, based on the judgment result of the speech/non-speech signal discriminator 102, separated by a multi-separating part 202, decoding processing function blocks 205, 206 are selected by change-over switches 203, 204 respectively. These blocks 205, 206 correspond to the blocks 105, 106.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a voice encoding / decoding device used in voice digital wire communication and wireless communication, and more particularly to a DTMF (Dual Tone Multi-Frequen).
The present invention relates to a voice encoding / decoding device capable of transmitting a non-voice signal using a voice frequency band such as a cy signal.

[0002]

2. Description of the Related Art Reducing communication costs is the most important issue in corporate communication. In order to realize high-efficiency transmission of voice signals, which occupy most of communication traffic, in recent years 8k
bit / sCS-ACELP (Conjugate-Structure Algebraic-Code-E
xcited Linear Prediction: Conjugate structure algebraic code excitation linear prediction) Applying a high-efficiency speech coder based on speech coding / decoding, as typified by speech coding (conforming to ITU-T Recommendation G.729). The number of cases to do is increasing.

In a voice coding algorithm having a transmission rate of 8 kbit / s class, an input signal is specialized for a voice signal in order to obtain a high quality voice with a small amount of information. This will be described by taking the above 8 kbit / s CS-ACELP method as an example. FIG. 20 shows a block diagram of the encoder, and FIG. 21 shows a block diagram of the decoder. This coding method is a coding algorithm that models a human vocalization mechanism, that is, a synthesis filter 412 (a linear filter corresponding to a spectral envelope of speech) that models human vocal tract information is configured. It is based on the CELP system in which a voice is reproduced by driving with a time-series signal (output of the excitation signal reproducing unit 411) stored in a codebook corresponding to a human vocal cord sound source. For a detailed explanation of the algorithm, see ITU-T Recommendation G.7.
29, ■ Coding of Speech at 8kbit / s using Conjugate-
Structure Algebraic-Code-Excited Linear Prediction
(CS-ACLEP) Please refer to ■.

When the coding algorithm has a structure specialized for voice, a signal other than a voice signal using a voice frequency band (for example, a dual tone configuration) is used in a transmission line using a high-efficiency voice encoding device. The transmission characteristics of DTMF signals used for push buttons, No.5 signaling, modem signals, etc.) tend to deteriorate as the transmission efficiency becomes higher.

As an example showing this, details of the LSP quantizer will be described with reference to FIGS. 22 and 23. Figure 2
2 is a detailed configuration of the LSP quantizer 309 in the encoder shown in FIG. 20, and FIG. 23 is an LSP in the decoder shown in FIG.
3 is a detailed configuration of the inverse quantization unit 407.

In the CS-ACELP speech coding system, the LSP
It is realized by performing three processing steps for coefficient quantization. That is, the LSP quantizer 309 has the following three processing function blocks. (1) Moving average (MA) predictive component calculation unit 308 for efficient quantization by subtracting predictable components between frames (2) Target LSP using a codebook learned by speech First-stage LSP for roughly quantizing Quantization codebook 335 (3) Second-stage LSP for performing fine adjustment with a codebook using a random number sequence on the target LSP roughly quantized at the first stage
Quantization codebook 336

By using the moving average prediction of (1), it is possible to efficiently quantize a signal having a small change in frequency characteristics, that is, a signal having a strong correlation between frames. Also (2)
By using the learning codebook of (1) and the random codebook of (3) together, although the mathematical rigor is lacking, the outline of the spectrum envelope can be efficiently expressed. In addition, the existence of the random number codebook in (3) makes it possible to flexibly follow subtle changes in the spectrum envelope. From the above viewpoint, the LSP quantizer 3
It can be said that 09 is a method that is well suited for efficiently encoding the features of speech spectral envelope information.

On the other hand, in encoding a non-voice signal, especially a DTMF signal, it is necessary to consider the following properties. The spectral envelope is clearly different from the voice signal. There is a sharp change in the spectral characteristics between the signal duration and the pause time. The gain also changes rapidly. However,
If limited to within the signal duration, the amount of change in both spectral characteristics and gain is extremely small. -The LSP quantization distortion is directly reflected in the DTMF signal frequency distortion, so it is necessary to minimize the LSP quantization distortion. -The frequency characteristics are extremely stable in the section where the DTMF signal continues. From the above viewpoints, the LSP quantization section 309 cannot be said to be an effective method for encoding the spectrum envelope information of the DTMF signal.

As shown in the above example, since a non-voice signal such as a DTMF signal has a property different from that of the voice signal in some respects, when encoding the non-voice signal,
Especially under the condition that the transmission rate is low and the redundancy for encoding is small, it is not appropriate to use the same method as that for the voice signal.

By the way, in intra-company communication, signaling lines are transmitted in-channel using DTMF signals, etc. without separately providing a signal line for signaling transmission for call connection in telephone communication. There are many. In this case, if the assigned transmission path is a transmission path using the high-efficiency voice coding described above, the transmission characteristics of the DTMF signal deteriorate, so that a case where call connection cannot be normally made frequently occurs. There is an evil.

As a first means for solving such a problem, for example, the device configuration shown in FIG. 24 as shown in JP-A-9-81199 may be adopted. In this configuration, the transmitting side has a means for distinguishing a voice signal from a non-voice signal such as a DTMF signal, and a memory storing a pattern in which the DTMF signal is encoded in advance. When the input of the DTMF signal is identified by this identifying means, the index of the memory holding the coding pattern corresponding to the DTMF number is transmitted to the receiving side, and the receiving side identifies the index, The DTMF signal corresponding to the number is generated.

Further, as a second means for solving such a problem, for example, there is a case where the device configuration of FIG. 25 as shown in Japanese Patent Laid-Open No. 9-326772 is adopted. In this configuration, when the input speech is encoded, the auditory weighting filter that obtains the filter coefficient to calculate the speech encoding amount and the reflection coefficient obtained by the auditory weighting filter adaptation unit for enhancing the perceptual quality are set values. Reflection coefficient determination means for comparing with
If the reflection coefficient is out of the specified range, the auditory weighting filter is used, and if the reflection coefficient is in the specified range, the use switching means not using the auditory weighting filter is provided.

Further, a signal level monitoring means for monitoring the change in the level of the input signal is added, and the use switching means switches the use of the auditory weighting filter by the combination of the reflection coefficient and the change in the level of the input signal. . Further, the decryption device side also has a corresponding use switching means.

[0014]

By the way, the voice / DTMF discriminator may make an erroneous discrimination during voice transmission. At this time, in the conventional first device configuration, the receiving side may generate a complete dual tone during the voice transmission, that is, the DTMF "beep" sound may be suddenly caught in the middle of the conversation. Therefore, it is expected that the damage to the voice quality will increase. Moreover, since the input of the decoder becomes indeterminate when it is identified as DTMF, there is a problem that the internal state of the decoder becomes discontinuous and abnormal noise is generated after returning to normal voice transmission.

Therefore, in order to maintain the voice quality at a certain level, the voice / DTMF discriminator is required to have a high performance with an erroneous discrimination rate close to zero. When this is implemented by a signal processor such as a DSP, the normal processing load is high. Since it is necessary to deal with such an identification algorithm, a high-performance processor is required for implementation, which causes a problem that the implementation cost becomes high.

Further, in the conventional second device configuration,
A coding algorithm with a relatively high transmission rate (about 16 kbit / s) can obtain sufficient DTMF transmission characteristics by improving the frequency distortion due to the auditory weighting filter and post filter, but it is still lower than this. In the speed (8 kbit / s or less) speech coding algorithm, since it is specialized for the coding of the speech signal due to the reduced redundancy, functional blocks other than the above filters encode more than distortion by the above filters. It is known that distortion is added. Therefore, the DTMF signal transmission characteristics cannot be sufficiently improved in this device configuration.

The present invention has been conceived in order to solve such a conventional problem, and improves the transmission characteristics of a non-voice signal such as a DTMF signal, and at the same time, the voice originally possessed by the encoding algorithm. It is an object of the present invention to provide a high-efficiency voice encoding / decoding device in which transmission quality is maintained by a simpler method and enable non-voice signals such as DTMF signals to be transmitted in-channel.

[0018]

A speech coding apparatus according to the present invention comprises a speech / non-speech discriminator which discriminates whether an input signal is a speech signal or a non-speech signal and outputs a judgment result. A first functional block that encodes an input signal based on a frequency parameter suitable for audio encoding and outputs an audio encoded signal, and the same encoding method as the first functional block, and the frequency parameter is input signal based on the frequency parameters suitable for the coding of different non-speech sequentially coding <br/> the, and a second functional block for outputting a non-speech coded signal, compresses and encodes the input signal An encoder and a switching means for selecting either a first functional block or a second functional block for supplying an input signal to the encoder according to the determination result of the speech / non-speech discriminator. The output of the above encoder and A determination result of the speech / non-speech discriminator multiplexes, and a multiplexing unit to output to the transmission path, the first
The functional block of LSP (Li
neSpectrumPair: Line spectrum pair) Quantum
The second functional block is suitable for non-speech coding.
LSP quantizer, which has the second functional block
The constituent LSP quantizer is a moving average prediction process of LSP.
To stop or reduce the effect of moving average prediction
It has the ability.

The speech coding apparatus according to the present invention discriminates whether the input signal is a speech signal or a non-speech signal.
And a voice / non-voice discriminator that outputs the determination result
The input signal based on frequency parameters suitable for encoding
A first functional block that encodes and outputs an audio encoded signal.
And the same encoding method as the first functional block above,
Suitable for non-speech coding with different frequency parameters
Sequentially encodes the input signal based on the frequency parameter,
A second functional block for outputting a voice coded signal,
And an encoder that compresses and encodes the input signal, and
The encoder of the input signal according to the determination result of the voice discriminator
Supply to the first functional block or the second functional block
Switch for selecting one of the
Force and the judgment result of the voice / non-voice discriminator are multiplexed.
And a multiplexing unit for outputting to the transmission line,
The functional block is an LSP quantizer suitable for audio coding,
The second functional block is L suitable for non-speech coding.
An SP quantizer, which is the first and second functional blocks described above.
, The weighting factors used for the quantization error calculation are different,
The second functional block in a specific frequency region
Of the weighting coefficient of the first functional block
It should be larger than the number.

The speech decoding apparatus according to the present invention is
Multiplexed data encoded and transmitted by a voice coder
Separation from coded signal into coded signal and voice / non-voice discrimination signal
Demultiplexer and decode the encoded signal into the original signal
And a decoder for transmitting the transmitted voice code.
A third functional block corresponding to the encoder that encodes the encoded signal
And a code that encodes the same non-voice code signal transmitted.
And a fourth functional block corresponding to an encoder,
Of the voice / non-voice discriminator separated by the data separation unit
Depending on the judgment result, the third functional block or the third functional block
There is a switching means for selecting any of the four functional blocks.
It is something.

[0021]

[0022]

[0023]

[0024]

[0025]

[0026]

[0027]

BEST MODE FOR CARRYING OUT THE INVENTION Embodiment 1 Hereinafter, Embodiment 1 of the present invention will be described with reference to the drawings. Figure 1
FIG. 3 is a block diagram showing a configuration of a voice encoding / decoding device according to Embodiment 1 of the present invention. In the transmission side of FIG. 1, 101 is a coder for highly efficient compression coding of a voice signal based on an algorithm, and 102 is a voice signal or a non-voice signal (for example, DTMF signal, No. Five
(Signaling etc.) and output the judgment result, a voice / non-voice signal discriminator, 103 and 104 are changeover switches, 105 and 1
Reference numeral 06 denotes an encoding processing functional block that executes a part of the functions of the encoder 101, and reference numeral 107 denotes a multiplexer that multiplexes the determination result of the discriminator and the output of the encoder and outputs the multiplexed result to the transmission path.
Here, the encoding processing function block 105 is optimized so that the audio signal can be compression-encoded with high efficiency. On the other hand, the encoding processing function block 106 is a non-voice signal,
For example, the DTMF signal is optimized so that it can be compression-encoded with little distortion.

On the receiving side of FIG. 1, 201 is a decoder for decoding a highly efficient compressed voice code into a voice signal based on the same algorithm as the encoder 101, and 202 is multiplexed by a multiplexer 107. The demultiplexing unit that separates and outputs the determination result of the speech / non-speech signal discriminator 102 and the output of the encoder 101, 203 and 204 are changeover switches, and 205 and 206 are partial functions of the decoder 201, respectively. It is a decryption processing function block to be executed. Here, 205 and 206 respectively correspond to the encoding processing functional blocks 105 and 106 of the encoder 101 or are functionally the same.

Next, the operation of this speech encoding / decoding device will be described. On the transmission side of FIG. 1, the voice / non-voice signal discriminator 102 constantly monitors whether the input signal is a voice signal or a non-voice signal, and determines the operation mode of the encoder 101 based on the determination result. decide. When the voice / non-voice signal discriminator 102 determines "voice", the changeover switch 103 is turned to the terminal 103A side and the switch 104 is turned to the terminal 104A side. As a result, the encoding processing function block 105 is selected inside the encoder 101, and the operation mode suitable for encoding the audio signal with high efficiency (hereinafter, referred to as (2) audio mode (3)) is set. In this mode, the encoder 101 executes the encoding process on the audio signal based on the encoding algorithm, and outputs the code corresponding to the input audio.

When the voice / non-voice signal discriminator 102 determines "non-voice", the changeover switch 103 is set to the terminal 103.
Tilt the same 104 to the B side and the terminal 104B side. as a result,
Inside the encoder 101, the encoding processing function block 106
Is selected, and an operation mode (hereinafter referred to as (1) non-voice mode (3)) suitable for compressing and encoding a non-voice signal, such as a DTMF signal, with a small distortion is selected. In this mode, the encoder 101 executes a coding process on a non-voice signal, such as a DTMF signal, based on a coding algorithm, and outputs a code corresponding to the input non-voice signal.

The operation of the voice / non-voice signal discriminator 102 will be described by using a DTMF signal as a non-voice signal to be discriminated as an example. Since the DTMF signal is composed of dual tones and the frequency component of the output signal is fixed to a specific value by regulation, ・ FFT etc. is used for frequency analysis ・ Specified using a bandpass filter The feature amount on the frequency axis is extracted using a method such as filtering the frequency component of
It can be identified by determining whether or not it matches the feature quantity of the signal.

Regarding the level of the DTMF signal as well, since the transmission level is limited to a specific range by regulation and the level fluctuation is small, an audio signal with a relatively large level fluctuation and a wide dynamic range is used. They show distinctly different characteristics. Therefore, by monitoring the level of the input signal, it is possible to improve the detection accuracy of the DTMF signal by using it as auxiliary information for identifying the DTMF signal. The voice / non-voice signal discriminator 102 has a function of uniquely calculating the above parameters using the input signals, making a decision based on them, and outputting the result.

The multiplexing unit 107 encodes the voice signal or the non-voice signal obtained as described above.
(Hereinafter, referred to as voice / non-voice code) and the discrimination result (voice signal or non-voice signal) of the input signal which is the output of the voice / non-voice signal discriminator 102 are multiplexed and sent to the transmission path.

On the other hand, on the receiving side in FIG. 1, first, the signal sequence received from the transmission path is separated in the demultiplexing unit 202 into the voice / non-voice code and the determination result of the voice / non-voice signal discriminator 102. . If the determination result of the voice / non-voice signal discriminator 102 extracted from the signal sequence is “voice”, the changeover switch 203 is set to the terminal 203A side and the switch 204 is set to the terminal 20.
Push down to the 4A side. As a result, the decoding processing function block 205 is selected inside the decoder 201, and the operation mode of the decoder corresponding to the voice mode of the encoder 101 is set. In this mode, the decoder 201 performs a decoding process based on a decoding algorithm and decodes an audio signal. At this time,
Since both the encoding and decoding processes are executed in the voice mode, the quality of the decoded voice signal matches the original performance of the encoding algorithm.

If the determination result of the voice / non-voice signal discriminator 102 extracted from the signal sequence by the demultiplexing unit 202 is "non-voice", the changeover switch 203 is moved to the terminal 203B side,
To the terminals 204B side. As a result, the decoding processing function block 206 is selected inside the decoder 201, and the operation mode of the decoder corresponding to the non-voice mode of the encoder 101 is set. In this mode, the decoder 201 performs a decoding process based on a decoding algorithm to decode a non-voice signal. At this time, since both the encoding / decoding processing is executed in the non-speech mode, the decoded non-speech signal has less distortion than that executed in the speech mode.

As described above, according to the present embodiment, a normal voice encoding / decoding algorithm suitable for voice encoding is used during voice signal transmission, and a non-voice signal, especially DTMF is used. When transmitting signals, etc., some processing function blocks are switched to a method more suitable for encoding non-voice signals and encoding / decoding processing is executed, so without increasing the transmission rate during non-voice signal transmission, High quality non-voice signal can be transmitted.

Further, in the present embodiment, encoding /
Since it changes a part of the decryption process and does not perform switching related to the essence of the algorithm,
For example, during voice signal input, the voice / non-voice signal discriminator 10
Even if it is erroneously identified as "non-voice" in 2, there is some degradation, but some voice transmission quality can be maintained, so
There is also an advantage that it is possible to suppress the harmful effects that may be felt during a call.

Second Embodiment Hereinafter, a second embodiment according to the present invention will be described with reference to FIGS. 2 and 3. The present embodiment relates to the mode switching of the first embodiment,
One operation example of the encoding processing function blocks 105 and 106 has been described in detail. Here, in order to make the description easy to understand, the CS-ACELP system (conforming to ITU-T recommendation G.729) is used as an example of the encoding algorithm. The detailed block diagram of the encoder based on the CS-ACELP system is as shown in FIG. 20 described above, and the detailed block diagram of the decoder is also shown in FIG. 21. Note that the constituent elements denoted by the same reference numerals as those in FIG. 1 are constituent elements having the same functions as those described in the section of the above-described first embodiment, and therefore redundant description will be omitted.

In this embodiment, the L side learned by a non-voice signal, for example, a DTMF signal, on the transmission side (encoder side).
SP Quantization Codebook 2 (310A) is prepared separately. That is, as shown in FIG. 2, the LSP quantized codebook learned by voice is included in the encoding processing function block 105 optimized for voice input.
1 (310) corresponds, and the LSP quantization codebook 2 (310A) corresponds to the encoding processing function block 106 optimized for non-voice signals. When the voice / non-voice signal discriminator 102 determines that it is “voice”, the changeover switch 103 is set to the terminal 103A and the switch 104 is set to the terminal 10A.
LSP based on LSP Quantization Codebook 1 (310)
Perform vector quantization of. On the other hand, when the voice / non-voice signal discriminator 102 determines that it is “non-voice”, the changeover switch 103 is set to the terminal 103B, and the same switch 104 is set to the terminal 104B.
Executes vector quantization of LSP based on LSP quantization codebook 2 (310A).

Similarly, on the receiving side (decoder side), as shown in FIG.
In addition to the LSP quantization codebook 1 (406), the LSP quantization codebook 2 (406A) that is exactly the same as the LSP quantization codebook 2 (310A).
Prepare A). That is, the decoding processing function block 205 optimized for voice input corresponds to the LSP quantization codebook 1 (406), and the decoding processing function block 20 optimized for non-voice signals is used.
LSP quantized codebook 2 (406A) corresponds to 6, respectively.
In the demultiplexing unit 202, if the extracted result of the voice / non-voice signal discriminator 102 is “voice”, the changeover switch 203 is turned to the terminal 203A side, the switch 204 is turned to the terminal 204A side, and the LSP quantum The LSP vector corresponding to the transmitted index is extracted from the coded codebook 1 (406). On the other hand, the voice / non-voice signal discriminator 102 extracted by the demultiplexing unit 202
If the result of the determination is “non-voice”, the changeover switch 203
To the terminal 203B side and the same 204 to the terminal 204B side.
The LSP vector corresponding to the transmitted index is extracted from SP quantization codebook 2 (406A).

As described above, according to the present embodiment, when the voice signal is input, the LSP is quantized using the codebook learned by the voice, and the DTMF signal is input. In this case, the LSP is quantized by using the codebook learned by the DTMF signal. Therefore, when transmitting a voice signal, encoding that is more suitable for the voice signal, and when transmitting the DTMF signal, a more DTMF signal is used. Appropriate encoding can be performed. By this means, the spectral envelope information of the DTMF signal can be quantized accurately and efficiently,
It is possible to further reduce quantization distortion during DTMF signal transmission.

Third Embodiment Hereinafter, a third embodiment according to the present invention will be described with reference to FIGS.
The present embodiment relates to the mode switching of the first embodiment and describes in detail one operation example of the coding processing functional blocks 105 and 106. Here, in order to make the explanation easy to understand, the CS-ACELP method (I
TU-T Recommendation G.729). Note that the constituent elements denoted by the same reference numerals as those in FIG. 1 are constituent elements having the same functions as those described in the section of the above-described first embodiment, and therefore redundant description will be omitted.

The present embodiment focuses on the fact that no subtle fluctuations in the spectrum envelope occur during normal DTMF signal transmission. The operation of the encoder will be described with reference to FIG. FIG. 4 shows a part of processing functional blocks of the encoder of the present invention based on the CS-ACELP system. In this embodiment, there are two types of LSP quantizers having different quantizing methods. That is, the LSP quantizer 1 (309)
For example, a codebook search method specialized for voice signal input as shown in FIG. 22 is realized. On the other hand, LSP
The quantizer 2 (309A) realizes a codebook search method specialized for the input of a non-voice signal. That is, the LSP quantizer is included in the encoding processing function block 105 optimized for the audio signal.
1 (309) corresponds, and the LSP quantizer 2 (309A) corresponds to the encoding processing functional block 106 optimized for non-voice signals.
When the voice / non-voice signal discriminator 102 determines that it is “voice”, the changeover switch 103 is set to the terminal 103A and the switch 104 is set to the terminal 104A, respectively, and the LSP based on the quantization method of the LSP quantization unit 1 (309) is used. Performs quantization. On the other hand, when the voice / non-voice signal discriminator 102 determines “non-voice”, the changeover switch 103
To terminal 103B and terminal 104 to terminal 104B, respectively, to execute LSP quantization based on the quantization method of LSP quantization section 2 (309A).

Similarly, on the receiving side, as shown in FIG. 5, in addition to the LSP dequantization unit 1 (407), the LSP quantization unit 2 (309
An LSP dequantization unit 2 (407A) corresponding to A) is prepared. That is,
The decoding processing function block 205 optimized for voice input has an LSP dequantization unit 1 (407), and the decoding processing function block 206 optimized for non-voice signals has an LSP dequantization unit 2 (407A).
Correspond to each. The determination result of the voice / non-voice signal discriminator 102 extracted by the demultiplexing unit 202 is “voice”
If so, the changeover switch 203 to the terminal 203A side, the same 204 terminal
The LSP is extracted to the 204A side by a method based on the LSP dequantization unit 1 (407). LSP dequantization unit 1 (407), for example,
Inverse quantization corresponding to the LSP quantization unit 1 (309) as shown in FIG. 23 is realized. On the other hand, if the determination result of the voice / non-voice signal discriminator 102 extracted by the demultiplexing unit 202 is “non-voice”, the changeover switch 203 is turned to the terminal 203B side and the switch 204 is turned to the terminal 204B side, and the LSP reverse The LSP is extracted by the method based on the quantizer 2 (407A).

As an example of the LSP quantizer 2 (309A), for example, the means shown in FIG. 6 is used. This is the LSP quantizer 1 (3
The signal input from the second stage LSP quantization codebook 336 in 09) is invalidated. For non-voice signals such as DTMF signals, the spectral envelope is known in advance,
In the first stage LSP quantization, a vector capable of expressing the spectrum envelope of a non-voice signal expected to be input with a small quantization distortion is embedded in the first stage LSP quantization codebook 335 in advance. As a result, the first-stage quantization can be quantized accurately. In addition, when the DTMF signal is input, the envelope information usually does not change subtly. Since a certain degree of accuracy has already been ensured by the first-stage quantization, the second-stage fine adjustment is unnecessary, and conversely it may cause frequency distortion, so it is disabled.

When the method shown in FIG. 6 is used on the transmission side, as an example of the LSP dequantization unit 2 (407A),
For example, the method shown in FIG. 7 is used. This is also LS
This is a signal in which the signal input from the second stage LSP quantization codebook 336 of the P inverse quantization unit 1 (407) is invalidated.

As described above, according to the third embodiment of the present invention, at the time of transmitting a DTMF signal, the first LSP quantizing means accurately quantizes the spectrum envelope information of the DTMF signal,
Also, since the second LSP quantizing means is omitted, paying attention to the fact that the fluctuation of the spectrum envelope information is small at the time of transmitting the DTMF signal.
DTMF transmission distortion can be reduced.

Alternatively, as shown in FIG.
LSP quantized codebook 2 (3
10A) is separately prepared and the means for switching the codebook to be used according to the determination result of the voice / non-voice signal discriminator 102 is also used, it is possible to obtain the same effect as the above-mentioned embodiment.

Embodiment 4. Embodiments 2 and 3 described above
Then, when transmitting DTMF signals, it was devised to reduce frequency distortion due to quantization distortion.
An embodiment for enabling to follow abrupt changes in spectrum envelope information occurring at the rise and fall of the DTMF signal will be shown. FIG. 9 shows the LSP according to the fourth embodiment.
FIG. 3 is a block configuration diagram showing an internal configuration of a quantization unit. Note that, in FIG. 9, the constituent elements denoted by the same reference numerals as those in FIG. 4 are constituent elements having the same functions as those described in the section of the above-mentioned third embodiment, and therefore redundant description will be omitted.

Next, the operation of the encoder will be described. In this embodiment, the operation of the block group related to LSP quantization shown in FIG. 4 is exactly the same as that according to the third embodiment. The difference from the third embodiment is only the inside of the LSP quantizer 2 (309A). According to FIG. 9, the LSP quantizer 2 (309A) has a configuration without the function of moving average prediction. Only the first stage LSP codebook is used for LSP quantization. As a result, when the voice / non-voice signal discriminator 102 discriminates the DTMF signal from the input signal, the same effect as stopping the LSP moving average prediction process can be obtained.

On the other hand, the decoder 201 also has a simple structure in which the moving average prediction function is removed from the LSP dequantization unit 2 (407A) shown in FIG. 7 in order to correspond to the encoder 101. , LSP
Can be decrypted.

Alternatively, as shown in FIG. 10, as described in the second embodiment, a means for separately preparing the first stage LSP quantization codebook 2 (335A) can be used together. At this time, by incorporating the moving average prediction coefficient codebook 2 (337A) configured to reduce the weighting of the past LSP of the moving average into the LSP quantization codebook 2 (310A), the moving average prediction effect is reduced. Can be realized.

At this time, the decoder 201 also has a configuration as shown in FIG.
Decoding of P can be realized.

As described above, according to the fourth embodiment of the present invention, the effect of moving average prediction in LSP quantization is eliminated or reduced during DTMF signal transmission. At the rise and fall of
It is possible to improve the characteristics of the DTMF signal waveform to be decoded.

Embodiment 5. Next, an embodiment will be described which focuses on more accurately encoding information in the vicinity of the frequencies of the dual tones forming the DTMF signal. FIG. 12 is a block configuration diagram showing an internal configuration of the LSP quantization unit 2 (309A) in the fifth embodiment. In addition, in FIG.
The constituent elements denoted by the same reference numerals as those in FIG. 6 are the same as those in the third embodiment.
Since it is a component having the same function as that described in the section 1), duplicate description will be omitted.

Next, the operation of the encoder will be described. In the present embodiment, the operation of the LSP quantization unit 2 (309A) shown in FIG. 12 is different from that of the third embodiment in the internal operation of the quantization error weighting coefficient calculation unit 2 (338A), but other The operation is the same as in the third embodiment.

The weighting coefficient used for the quantization error calculation is
According to the CS-ACELP method, it is calculated by the method shown in the following formula.

[0058]

[Equation 1]

Here, ωi is an i-th order LSP. That is, the weighting coefficient is made heavier in the frequency range in which the peak of the spectrum comes, and the weighting coefficient is made lighter in the frequency range at the "valley" of the spectrum. This has the effect of making the contribution of the quantization error heavy in the frequency range showing the peak of the spectrum and sharpening the sensitivity to the error.

It is known that the dual-tone spectrum information of the DTMF signal is concentrated in the low-order LSP around the 1st to 6th orders. The quantization error weighting coefficient calculation unit 2 (338A) is a quantization error weighting coefficient calculation unit (338)
It has been improved by utilizing this fact. That is, in the vector quantization of the 10th-order LSP coefficient, when calculating the vector distance, the weighting is further weighted for the error of the low-order LSP around the 1st to 6th order (that is, even a slight error is sensitive to the error). To increase the coefficient sensitivity), higher order
The LSP is configured to be lightly weighted (that is, the coefficient sensitivity is kept low to allow some error). This can be easily realized, for example, by multiplying w1 to w6 in the above equation by a coefficient of 1 or more.

As described above, according to the fifth embodiment of the present invention, since the LSP quantization can be made more accurate in the vicinity of the dual tone frequency forming the DTMF signal, the LSP quantization error can be increased. Can be reduced, and thus the decrypted D
It is possible to reduce the frequency distortion of the TMF signal.

Embodiment 6 Hereinafter, Embodiment 6 according to the present invention will be described with reference to FIG. The present embodiment is a voice / non-voice signal discriminator of the first embodiment.
The present invention relates to the multiplexing of signal identification information which is the output of 102, that is, the multiplexer 107 and the demultiplexer 202.

Next, the operation will be described. Figure 13
It is a block diagram which shows one example of a structure of the quantizer for quantizing a certain parameter as a target. Signals that are candidates for quantized values are stored in the quantized codebook 340 in advance. When the target signal to be quantized is input, the least square error calculator 342 calculates the error from the target signal for each candidate of the quantization codebook. In this way, a codebook candidate that gives the smallest squared error is searched, and the index attached to the found candidate is extracted and output from the quantizer as the optimum code.

In this embodiment, a specific constraint condition is used when searching for the optimum code. As an example, when transmitting a voice signal, only the white parts (indexes 0000 to 1010) in the quantization codebook of FIG. 13 are searched. Also, during DTMF signal transmission, only the shaded portions (indexes 1011-1111) are searched for. By providing such a limitation, the demultiplexing unit on the receiving side can know the determination result of the voice / non-voice signal discriminator 102 by checking the quantized value of the parameter, and the discrimination pattern can be multiplexed. It can have the same effect as that done.

As described above, according to the sixth embodiment,
Since it is not necessary to provide a bit for identification, the determination result of the voice / non-voice identification device can be sent to the receiving side without increasing the transmission rate.

Seventh Embodiment In the sixth embodiment, the determination result information of the voice / non-voice signal discriminator 102 is embedded by using different quantized codes during voice signal transmission and during DTMF signal transmission. Is possible, but next,
An actual example in which one of the quantization tables is used as a voice / non-voice identification pattern will be described in detail with reference to FIGS. The constituent elements denoted by the same reference numerals as those in FIGS. 1 to 13 are constituent elements having the same functions as those described in the above-described first to sixth embodiments, and therefore redundant description will be omitted.

In the present embodiment, more efficient voice / non-voice signal transmission can be realized by using, for example, the means of the third embodiment. Here, FIG. 14 shows the LSP quantizer 1 (30
9) and a block diagram showing the internal structure of LSP quantized codebook 1 (310), and FIG. 15 is also used during DTMF signal transmission.
FIG. 3 is a block diagram showing an internal structure of an LSP quantization unit 2 (309A) and an LSP quantization codebook 2 (310A). In the quantized codebook according to the sixth embodiment, the second-stage LSP in the present embodiment is used.
Corresponds to the quantization codebook 336. The quantization candidates assigned during DTMF signal transmission are the second stage LSP quantization codebook 3
There is only one candidate, which is shaded in 36. The remaining candidates are assigned at the time of voice signal transmission, and the above candidates become the prohibition code in this mode. Since there is only one candidate assigned during DTMF signal transmission, its index can be treated as it is as a pattern for identifying the DTMF transmission mode.

Here, although only one second-stage LSP code is assigned at the time of DTMF signal transmission, as shown in FIG. 15, the first-stage LSP quantization codebook 2 (335A) is used for DT.
Since the spectrum envelope information of the MF signal can be sufficiently expressed, it is not necessary to execute the second stage LSP quantization, and therefore, there is originally no need to send the information of the second stage LSP quantization. Therefore, problems such as deterioration of characteristics do not occur.

As a result of the operation of the encoder described above, the frame format output from the encoder formed by the multiplexing unit 107 is as shown in FIG. The DTMF signal transmission mode is characterized in that the index used as the inhibition code in the second stage LSP quantization is sent as the DTMF signal transmission identification pattern. In the demultiplexing unit 202 on the receiving side, the second stage L
The SP is constantly monitored, and if it matches this identification pattern, it is interpreted as the DTMF signal transmission mode and the decoder 201 is controlled.

As described above, according to the seventh embodiment,
Since the number of prohibited codes (quantization steps) that cannot be selected during voice transmission is suppressed to only one, the signal sequence for transmitting the determination result of the voice / non-voice signal discriminator 102 hardly deteriorates the quality during voice transmission. It has the advantage that it can be embedded. Also, during DTMF signal transmission, it is an algorithm specialized for encoding and transmitting DTMF signals, and a sufficient number of bits for transmitting the selection prohibition code that is the identification pattern is secured. There is an excellent advantage that there is no deterioration of the DTMF signal.

Eighth Embodiment In the above seventh embodiment, the LSP is used as the quantization codebook to be divided.
The case of using the quantization codebook has been described. Next, the gain quantization unit 316 and the gain quantization codebook 317 in FIG.
An actual example of using will be described in detail with reference to FIGS. 17 and 18.

The eighth embodiment is a means focusing on the fact that the fluctuation of the gain is extremely small during the transmission of the DTMF signal except when the DTMF signal rises and falls. The present embodiment is the same as the case of the seventh embodiment except that one of the gain-quantized vector codebooks is prohibited as the identification information multiplexing means and the index is used for the identification pattern. At the time of voice signal transmission, the gain quantization performed every subframe (5 msec) is thinned out every frame (10 msec) to reduce the redundancy when the DTMF signal transmission with small gain fluctuation is performed. Characterized by embedding.

FIG. 18 shows a time schedule for gain quantization executed by the gain quantization unit 316. In the voice signal transmission mode, gain quantization is performed twice in one frame (that is, every subframe). A normal codebook search is performed in the first subframe. In the gain codebook search of the second subframe, the quantization method shown in Embodiment 6 is used. That is, when transmitting a voice signal, a full search is performed except for the only forbidden vector, and the index of the optimum vector is output from the gain quantizer 316.

At the time of DTMF signal transmission, gain quantization is executed only once per frame. This is, for example,
This can be achieved by quantizing the gain for all frames. In the codebook search, the full search is executed as in the first subframe in the voice signal transmission mode. Then, instead of the gain quantization information for one subframe, the gain quantization unit 316 outputs the index of the vector for which the search is prohibited at the time of voice input as identification information.

As a result of the operation of the encoder described above, the frame format output from the encoder formed by the multiplexing unit 107 is as shown in FIG. In the DTMF signal transmission mode, the gain quantization code for all frames is inserted in the portion where the first subframe gain quantization information was sent in the voice signal transmission mode. Further, it is characterized in that the index used as the inhibition code in the quantization of the gain code in the second subframe is sent as the DTMF signal transmission identification pattern. In the demultiplexing unit 202 on the receiving side, the second frame gain code is monitored, and if it matches this identification pattern, it is interpreted as the DTMF signal transmission mode and the decoder 20
Control 1

In the voice transmission mode, the decoder 201 performs inverse quantization on the basis of the gain code sent thereto, updates the gain value for each subframe, and performs decoding processing. On the other hand, in DTMF signal transmission mode, 1 per frame
Since only one gain code is sent, the gain is updated in frame units.

As described above, according to the eighth embodiment,
Since the number of prohibited codes (quantization steps) that cannot be selected during voice transmission is suppressed to only one, the signal sequence for transmitting the determination result of the voice / non-voice signal discriminator 102 hardly deteriorates the quality during voice transmission. It has the advantage that it can be embedded. Also, during DTMF signal transmission, it is an algorithm specialized for encoding and transmitting DTMF signals, and a sufficient number of bits for transmitting the selection prohibition code that is the identification pattern is secured. , It is possible to eliminate the deterioration of the DTMF signal.

Ninth Embodiment The ninth embodiment according to the present invention will be described below with reference to FIG. The present embodiment describes in detail one operation example of the voice / non-voice signal discriminator 102. Note that the constituent elements denoted by the same reference numerals as those in FIG. 1 are constituent elements having the same functions as those described in the section of the above-described first embodiment, and therefore redundant description will be omitted.

Next, the operation will be described with reference to FIG. In the first embodiment, several examples of the DTMF signal identification means are presented, but it can be said that they are roughly divided into two characteristics: (1) frequency component and (2) gain component. Here, in a speech encoding / decoding device using, for example, the CS-ACELP scheme (FIGS. 20 and 21) as an encoding scheme, a linear prediction method is used in order to analyze and efficiently encode an input speech signal. An LP analysis unit 302, which is a coding processing functional block that executes frequency analysis of a voice signal based on Information indicating the characteristics of the DTMF signal can be derived from the parameters (eg, reflection coefficient) calculated here. The voice / non-voice signal discriminator 102 has this function, and the voice / non-voice signal discriminator 102 itself has no means for performing frequency analysis. Since the details of the voice / non-voice signal discriminator 102 are described in detail in Japanese Patent Laid-Open No. 9-326772, the description thereof is omitted here.

As described above, according to the ninth embodiment,
Since the parameters required for the encoding process are diverted to be used for identifying the voice / non-voice signal, the parameters required for the signal identification can be easily obtained. Therefore, the processing load for signal identification can be reduced.
Since it can be realized by simple processing, it can be expected that the device configuration can be simplified, and there is an advantage that the cost for realizing the device can be reduced.

[0081]

As described above, according to the present invention, when a voice signal is transmitted, a method using a normal voice encoding / decoding algorithm, which is more suitable for voice encoding, is used. When transmitting DTMF signals, etc., some processing function blocks are switched to a method more suitable for encoding non-voice signals, and encoding / decoding processing is executed. High quality non-voice signals can be transmitted without raising.

In addition, in the present invention, a part of the encoding / decoding process is changed, and switching that does not relate to the essence of the algorithm is not performed. / Even if the non-voice signal discriminator 102 erroneously identifies it as "non-voice", although there is some deterioration, it is possible to maintain a certain level of voice transmission quality, so that it is possible to suppress the harmful effect on the ear during a call. There are also advantages.

Further, even if a voice / non-voice signal discriminator having a poor discriminating performance constructed by a simple method is applied, the voice quality can be maintained to a certain extent, so that the device constitution can be simplified. There is an excellent effect that the cost for realization can be reduced.

[Brief description of drawings]

FIG. 1 is a block configuration diagram of a speech encoding / decoding device according to a first embodiment.

[Fig. 2] Fig. 2 is a block configuration diagram of a portion related to LSP quantization in the speech encoder according to the second embodiment.

[Fig. 3] Fig. 3 is a block configuration diagram of a portion related to LSP dequantization in the speech decoder according to the second embodiment.

[Fig. 4] Fig. 4 is a block configuration diagram of a portion related to LSP quantization in the speech coder according to the third embodiment.

[Fig. 5] Fig. 5 is a block configuration diagram of a portion related to LSP dequantization in the speech decoder according to the third embodiment.

FIG. 6 is a second LS in the speech coder according to the third embodiment.
FIG. 3 is a detailed configuration diagram of a P quantization unit and an LSP quantization codebook.

FIG. 7 is a second LS in the speech decoder according to the third embodiment.
FIG. 3 is a detailed configuration diagram of a P inverse quantization unit and an LSP quantization codebook.

[Fig. 8] Fig. 8 is a block configuration diagram of a part related to LSP quantization in the speech encoder according to the third embodiment.

FIG. 9 is a second LS in the speech coder according to the fourth embodiment.
FIG. 3 is a detailed configuration diagram of a P quantization unit and an LSP quantization codebook.

FIG. 10 is a block diagram showing a second embodiment of the speech coder according to the fourth embodiment.
It is a detailed block diagram of an LSP quantization part and an LSP quantization codebook.

FIG. 11 is a block diagram showing a second embodiment of the speech decoder according to the fourth embodiment.
FIG. 3 is a detailed configuration diagram of an LSP dequantization unit and an LSP quantization codebook.

FIG. 12 is a block diagram illustrating a second embodiment of the speech coder according to the fifth embodiment.
It is a detailed block diagram of an LSP quantization part and an LSP quantization codebook.

[Fig. 13] Fig. 13 is a diagram illustrating one example of a configuration of a quantization codebook in the speech coder according to the sixth embodiment.

FIG. 14 is a diagram showing a first embodiment of the speech coder according to the seventh embodiment.
It is a detailed block diagram of an LSP quantization part and a 1st LSP quantization codebook.

FIG. 15 is a block diagram illustrating a second embodiment of the speech coder according to the seventh embodiment.
It is a detailed block diagram of an LSP quantization part and a 2nd LSP quantization codebook.

FIG. 16 is a diagram showing a frame format at the time of signal transmission of the speech encoding / decoding device according to the seventh embodiment.

FIG. 17 is a diagram showing a frame format at the time of signal transmission of the audio encoding / decoding device according to the eighth embodiment.

FIG. 18 is a table showing a time schedule of gain quantization processing blocks of the speech encoding / decoding device according to the eighth embodiment.

FIG. 19 is a block configuration diagram of a speech encoding / decoding device according to a ninth embodiment.

FIG. 20 is a block diagram showing an internal configuration of an encoder based on the CS-ACELP system.

FIG. 21 is a block diagram showing an internal configuration of a decoder based on the CS-ACELP method.

[Fig. 22] Fig. 22 is a block diagram illustrating an internal detailed configuration of an LSP quantization unit and an LSP quantization codebook of an encoder in the CS-ACELP system.

[Fig. 23] Fig. 23 is a block diagram illustrating an internal detailed configuration of an LSP dequantization unit and an LSP quantization codebook of a decoder in the CS-ACELP system.

FIG. 24 is a block diagram showing a configuration of a conventional speech encoding / decoding device.

FIG. 25 is a block diagram showing a configuration of a conventional speech encoding / decoding device.

[Explanation of symbols]

101: encoder, 102: voice / non-voice signal discriminator, 10
3, 104: Changeover switch, 105, 106: Encoding processing functional block, 107: Multiplexing unit, 201: Decoder, 202: Demultiplexing unit, 203, 204: Changeover switch, 205, 206: Decoding processing functional block, 301: precoding unit, 302: LP analysis unit,
303: Linear prediction calculation unit, 304: Auditory weighting filter, 305: LP-LSP conversion unit, 306: Open loop pitch search unit, 307: Non-quantized LSP-LP conversion unit, 308: Moving average prediction component calculation unit, 309 , 309A: LSP quantizer, 310, 310A: L
SP quantized codebook, 311: Quantized LSP-LP converter, 312: LP inverse filter, 313: Pitch component calculator, 314: Excitation signal
(Periodic component) Quantizer (closed-loop pitch searcher), 315, 32
2: Impulse response calculator, 316: Gain quantizer, 317:
Gain quantized codebook, 318: target signal calculation unit for codebook search, 319: pulse preliminary selection unit, 320: pitch prefilter, 321: excitation signal (noise component) quantization unit (fixed codebook search unit), 323 : Gain moving average predictor, 324: Code calculator, 325: Pitch post filter, 326, 331, 33
2, 333: adder, 327: quantized excitation signal calculation unit,
330: Gain multiplier, 334: Least square error calculator, 335,
335A: First stage LSP quantization codebook, 336: Second stage LSP quantization codebook, 337, 337A: Moving average prediction coefficient codebook, 338: Quantization error weighting coefficient calculator, 340: General quantization Codebook, 341: adder, 342: least square error calculator,
401: pitch lag decoding unit, 402: gain quantization codebook, 40
3: Gain dequantization unit, 404: Gain moving average prediction unit, 405:
Excitation signal (noise component) dequantization unit, 406, 406A: LSP quantization codebook, 407, 407A: LSP dequantization unit, 408: Pitch post filter, 409: LSP interpolation unit, 410: LSP-LP conversion Part, 411: excitation signal calculation part, 412: synthesis filter,
413: Post filter, 414: Post-processing unit.

Claims (3)

(57) [Claims] 1. A voice / non-voice discriminator that discriminates whether an input signal is a voice signal or a non-voice signal and outputs a determination result, and an input signal based on a frequency parameter suitable for voice encoding. A first functional block that encodes and outputs a speech-encoded signal, and an input based on a frequency parameter suitable for non-speech encoding that is different from the above-mentioned frequency parameter by the same encoding method as the first functional block. Sequentially encode signal
And a second functional block for outputting a non-voice coded signal, which compresses and encodes the input signal, and the code of the input signal according to the determination result of the voice / non-voice discriminator. Switching means for selecting either the first functional block or the second functional block for supply to the audio encoder, the output of the encoder, and the determination result of the voice / non-voice discriminator, and transmission. And a multiplexing unit for outputting to a channel, the first functional block is an LS suitable for audio coding.
P (LineSpectrumPair): Line spectrum
Pair) quantizer, the second functional block is a non-voice code
Is an LSP quantizer suitable for encoding, and the LSP quantizer that constitutes the second functional block is
Stop the LSP moving average prediction process, or
A speech coding apparatus having a function of reducing the effect of uniform prediction . 2. A voice / non-voice discriminator that discriminates whether an input signal is a voice signal or a non-voice signal and outputs a determination result, and an input signal based on a frequency parameter suitable for voice encoding. A first functional block that encodes and outputs a speech-encoded signal, and an input based on a frequency parameter suitable for non-speech encoding that is different from the above-mentioned frequency parameter by the same encoding method as the first functional block. Sequentially encode signal
And a second functional block for outputting a non-voice coded signal, which compresses and encodes the input signal, and the code of the input signal according to the determination result of the voice / non-voice discriminator. Switching means for selecting either the first functional block or the second functional block for supply to the audio encoder, the output of the encoder, and the determination result of the voice / non-voice discriminator, and transmission. And a multiplexing unit for outputting to a channel, the first functional block is an LS suitable for audio coding.
P quantizer, the second functional block is a non-voice code
Is an LSP quantizer suitable for quantization, and the first and second functional blocks described above are used for quantization error calculation.
The weighting factors used are different, and
In addition, the weighting coefficient of the second functional block is
A speech coding apparatus, characterized in that the weighting coefficient is made larger than the weighting coefficient of the first functional block . 3. A demultiplexer for separating a coded signal and a speech / non-speech discrimination signal from a multiplexed data sequence coded and transmitted by the speech coder according to claim 1 or 2. And a decoder for decoding the coded signal into an original signal, the decoder including a third functional block corresponding to the encoder for coding the transmitted voice coded signal, and the same transmission. A fourth corresponding to an encoder that encodes the encoded non-voice coded signal
And a third functional block separated according to the determination result of the voice / non-speech classifier separated by the multiplex data separation unit.
Alternatively, a voice decoding device having a switching means for selecting one of the fourth functional blocks.