JPH0548462A - Linear prediction coding/decoding system - Google Patents

Abstract

PURPOSE:To prevent the quality of a voice signal from being deteriorated even when an analysis frame for linear prediction analysis is extended with respect to the linear prediction coding/decoding system used for applying information compression decoding to the voice signal. CONSTITUTION:This system in which a linear prediction analysis section 11 applies linear prediction analysis to the inputted voice signal to send the extracted linear prediction coefficient at a prescribed transmission speed is provided with a comparison means 3 while the linear prediction analysis is implemented for a prescribed number of times in an analysis frame with a deviated time, and when the comparison means 3 detects it that the inputted linear prediction coefficient is changed by a prescribed value or over than the linear prediction coefficient inputted precedingly, the linear prediction coefficient at the point of detection and analysis location information are sent.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to audio signal information compression and
The present invention relates to a linear predictive coding / decoding method used when decoding.

Generally, by applying a high-efficiency voice coding system to a communication system, it is possible to reduce the line cost by transmitting a voice signal at a low bit rate. Simultaneous communication of a voice signal and a non-voice signal is facilitated, which improves economy and convenience. It is possible to effectively use the radio frequency and to make the voice storage memory economical. The advantages such as are obtained.

For this reason, the above high-efficiency coding system is expected to be applied to in-house communication systems, digital mobile radio systems, etc. However, even if the analysis cycle of linear prediction is lengthened, the quality of the voice signal will be high. It is necessary not to lower it.

[0004]

2. Description of the Related Art FIG. 8 is a block diagram of a conventional example, and FIG. 9 is an explanatory diagram of an analysis method of the conventional example. The operation of FIG. 8 will be described below with reference to FIG.

First, as shown in FIG. 9, the input speech waveform is cut out into an analysis frame of, for example, 40 ms and added to the linear prediction analysis unit 11. The linear predictive analysis unit performs linear predictive analysis on the spectral envelope of the clipped speech waveform, for example, in the 10th order to obtain linear predictive coefficients a _i (i = 1 to 10), and calculates the current input speech waveform s (n ), The input speech waveform s (n) is predicted using the speech waveform s (ni) input i samples before.

At this time, the input speech waveform is expressed by the equation (1) as is well known.

[0007]

[Equation 1]

Here, e (n) is a prediction residual. In addition,
Since the predictive inverse filter 12 has the inverse characteristic of the linear predictive synthesis filter described later, the predictive residual e (n) of the difference between the input speech waveform and the predicted input speech waveform is obtained.

Then, the multiplexing unit 13 multiplexes the linear prediction coefficient a _i and the prediction residual e (n) and transmits them. on the other hand,
In the decoder, the separation unit 21 uses the linear prediction coefficient a _i and the prediction residual e.
After separation into (n), the speech waveform s (n) is reproduced by synthesizing with the predictive synthesizing filter having the characteristic of the first term of the equation (1).

Here, the linear prediction coefficient a _i is constant during the processing of one analysis frame, but in the next analysis frame, the linear prediction coefficient is determined using the next speech waveform and the processing is performed.

In order to efficiently transmit the prediction residual, vector quantization of the prediction residual vector is performed to obtain the optimum index, and the optimum index is transmitted (code-driven linear predictive coding method: CELP) or prediction. A method of modeling the residual vector with a finite number of pulse trains and transmitting the optimum pulse position and pulse amplitude (multi-pulse drive coding method:
There are several methods such as MPC).

The linear prediction analysis is usually performed in the same period as the frame period (for example, 40 ms as shown in FIG. 9), but the one using the autocorrelation method is Weighting that reduces the effect of both ends (the prediction error at both ends becomes large and the continuity between frames becomes long)
Analysis is performed on the (windowed) signal.

As the window function, a Hamming window, a Hanning window,
There is a triangular window (see Fig. 9).

[0014]

As described above, the analysis processing is performed on the weighted signal, but if the analysis frame length is increased, the transmission frequency of the linear prediction coefficient decreases, so the transmission rate should be reduced. You can

However, there is a problem that the prediction gain decreases because the analysis result cannot follow the change of the actual spectrum envelope of the speech waveform. An object of the present invention is to prevent the quality of a voice signal from deteriorating even if the analysis frame of the linear prediction analysis is lengthened.

[0016]

FIG. 1 is a diagram showing the principle of the first and third inventions, FIG. 2 is a diagram showing the principle of the second and third inventions, and FIG. 3 is a diagram showing the structure of the fourth invention. It is a principle block diagram.

In the figure, 3 is comparison means, 4 is prediction gain calculation,
A comparison means, 5 is a linear interpolation means, and 11 is a linear prediction analysis unit. According to the first aspect of the present invention, the linear predictive analysis is performed a predetermined number of times while shifting the time within the analysis frame, and the comparison means is provided.

When the comparison means detects that the input linear prediction coefficient has changed by a predetermined value or more from the previously input linear prediction coefficient, the linear prediction coefficient and the analysis position information at the time of detection are detected. To transmit.

In a second aspect of the present invention, the linear prediction analysis is performed a predetermined number of times while shifting the time within the analysis frame, and
Prediction gain calculation / comparison means will be provided. Therefore, the predictive gain calculating / comparing means calculates the predictive gain using the input linear predictive coefficient, and then the calculated predictive gain changes by a predetermined value or more than the predictive gain measure coefficient calculated last time. When it is detected, the linear prediction coefficient and analysis position information at the time of detection are transmitted.

In the third aspect of the present invention, the analysis frame is equally divided into a plurality of subframes, linear prediction analysis is performed for each subframe, and the subframe number is transmitted as the analysis position information.

According to a fourth aspect of the present invention, the linear interpolating means uses the received analysis position information until the time point when the linear prediction coefficient is updated in the analysis frame. A linear prediction coefficient generated by interpolating the prediction coefficient is used.

[0022]

According to the first and second aspects of the present invention, the linear prediction coefficient obtained by performing the linear prediction analysis a predetermined number of times by shifting the time within the speech analysis frame, and the prediction gain calculated using this coefficient are When it is detected that the linear prediction coefficient and the prediction gain obtained last time are changed by a predetermined value or more, the linear prediction coefficient and the analysis position information at the time of detection are transmitted.

According to a third aspect of the present invention, the above-mentioned analysis frame is equally divided into a plurality of subframes and linear prediction analysis is performed for each subframe, but the subframe number is transmitted as analysis position information.

According to a fourth aspect of the present invention, when a linear predictive coded signal generated using the first and second aspects of the present invention is decoded by a decoder, the linear analysis coefficient is received by using the received analysis position information. Interpolated linear prediction coefficients are used in the analysis frame that is not updated.

As described above, according to the present invention, when the change in the spectral envelope of the voice is more than the predetermined value and the position and the changed linear prediction coefficient are transmitted, the ability to follow the change is high. improves.

As a result, the quality of the voice signal does not deteriorate even if the analysis frame of the linear prediction analysis is lengthened.

[0027]

FIG. 4 is a functional block diagram (encoder) of the first and third embodiments of the present invention, and FIG. 5 is a functional block diagram (encoder) of the second and third embodiments of the present invention. FIG. 6 is an explanatory diagram of the analysis method of the embodiment of the present invention, and FIG. 7 is a functional configuration diagram (decoder) of the fourth embodiment of the present invention.

Here, the memory part 31, the comparison part 32, the count part 33, and the switch part SW ₁ are components of the comparison means 3, the memory part 41, the comparison part 42, the count part 43, the predicted gain calculation part 44, and the switch part. SW ₂ is a component of the predictive gain calculation / comparison means 4, and linear interpolation calculation portion 51 is a component of the linear interpolation means 5.

The same reference numerals denote the same objects throughout the drawings. The operation of FIGS. 4 to 7 will be described below, but a brief description will be given for the part described in detail in the conventional example. As shown in FIG. 6, one analysis frame is 40 ms, which is 8 ms.
For example, it is divided into 5 subframes by ms, a linear prediction analysis is performed for each subframe, and the analysis position information transmits a subframe number (0 to 4).

First, referring to FIG. 4, the linear prediction analysis unit 11
Calculates the linear prediction coefficient a _i by performing a linear prediction analysis on the input speech waveform, and compares the prediction coefficient with the comparison part 32 and the memory part 31.
To send to.

Since the linear prediction coefficient obtained this time and the linear prediction coefficient one subframe before read from the memory section 31 are applied to the comparison section, the former is a predetermined amount more than the latter. As described above, it is detected whether or not there is a change.

Here, it is assumed that, for example, when the current linear prediction coefficient is written in the memory portion, the previous linear prediction coefficient is read out. If the detection result has changed by a predetermined amount or more, the switch part SW ₁ is driven by the detection result from the comparison part to be turned on, so that the linear prediction coefficient changed by a predetermined amount or more is turned on. It is sent to the multiplexing unit 13 via the part.

The counting portion counts clocks having a period of, for example, 8 ms, and sends the count value at the time when the above detection result is applied to the multiplexer 13 as the corresponding subframe number.

The multiplexing unit multiplexes the prediction residual signal from the prediction inverse filter 12, the changed linear prediction coefficient, and the corresponding subframe number and sends them to the receiving side. Next, in FIG. 5, as described above, the linear prediction analysis unit 11
Performs a linear prediction analysis on the input speech waveform to obtain a linear prediction coefficient and sends it to the prediction gain calculation section 44.

The prediction gain calculation part calculates the prediction gain G (m) by the following equation (2).

[0036]

[Equation 2]

Note that the denominator of the equation (2) is the prediction residual (m) in the m-th subframe obtained using the prediction coefficient of the previous frame.
Error), the numerator is the power of the input speech waveform at the m-th sub-frame position, and G (m) is the prediction gain at the m-sub-frame position (m = 0-4).

As can be seen from equation (2), the prediction power increases as the prediction error power decreases, and the prediction gain decreases as the prediction error power increases. Now, the prediction gain calculation part 44 sends the calculated prediction gain to the comparison part 42 and the memory part 41. Therefore, the prediction gain calculated this time and the prediction gain one subframe before are sent to the comparison part. Is applied. Therefore, the comparison part detects whether the former has changed by a predetermined amount or more than the latter.

If the detection result has changed by a predetermined amount or more, the switch part SW ₂ is driven by the detection result from the comparison part to be turned on, so that the linear prediction coefficient changed by a predetermined amount or more. Is sent to the multiplexing unit 13.

The counting section 43 counts, for example, the clock with a period of 8 ms from the prediction gain calculating section 44, but the count value at the time when the above detection result is applied is multiplexed as the corresponding subframe number. It is sent to the conversion unit 13.

The multiplexing unit multiplexes the prediction residual signal from the prediction inverse filter, the subframe number and the linear prediction coefficient and sends them to the receiving side. Further, in FIG. 7, in the decoder,
The demultiplexer 51 converts the multiplexed signal into a prediction residual signal, a linear prediction coefficient,
Separated into subframe numbers, but linear catch calculation part 51
Then, until the linear prediction coefficient is updated in the frame by using the separated sub-frame number, the coefficient obtained by linearly interpolating the linear prediction coefficient of the previous frame and the current frame is sent to the prediction synthesis filter 22.

Therefore, the predictive synthesis filter reproduces a voice waveform by using the coefficient and the prediction residual signal which are linearly interpolated while referring to the subframe number. It is also possible to reproduce the speech waveform using the linear prediction coefficient of the previous frame until the time point when the linear prediction coefficient is updated in the frame without using the linear interpolation calculation part.

That is, since the time when the change of the speech spectrum envelope is equal to or more than a predetermined value is detected and the position and the changed linear prediction coefficient are transmitted, the ability to follow the change is improved.

[0044]

As described above in detail, according to the present invention, it is possible to prevent the quality of the voice signal from being deteriorated even if the analysis frame of the linear prediction analysis is lengthened.

[Brief description of drawings]

FIG. 1 is a principle configuration diagram of first and third aspects of the present invention.

FIG. 2 is a principle configuration diagram of second and third present inventions.

FIG. 3 is a principle configurational diagram of a fourth invention.

FIG. 4 is a functional configuration diagram (encoder) of the first and third embodiments of the present invention.

FIG. 5 is a functional configuration diagram (encoder) of the second and third embodiments of the present invention.

FIG. 6 is an explanatory diagram of an analysis method according to an embodiment of the present invention.

FIG. 7 is a functional configuration diagram of a fourth embodiment of the present invention (decoder).
Is.

FIG. 8 is a configuration diagram of a conventional example.

FIG. 9 is a diagram illustrating an analysis method of a conventional example.

[Explanation of symbols]

3 Comparison Means 4 Prediction Gain Calculation / Comparison Means 5 Linear Interpolation Means 11 Linear Prediction Analysis Unit

 ─────────────────────────────────────────────────── ─── Continuation of the front page (72) Inventor Hideaki Kurihara 1015 Kamiodanaka, Nakahara-ku, Kawasaki-shi, Kanagawa Fujitsu Limited

Claims (4)

[Claims] 1. A linear predictive coding method for transmitting a linear predictive coefficient extracted by performing a linear predictive analysis of an input speech signal in a linear predictive analysis unit (11), wherein Is performed a predetermined number of times in the analysis frame with a time shift, and a comparison means (3) is provided, in which the input linear prediction coefficient changes by a predetermined value or more than the previously input linear prediction coefficient. A linear predictive coding method characterized by transmitting a linear predictive coefficient and analysis position information at the time of detection when it is detected. 2. In the above-described linear predictive coding system, linear predictive analysis is performed a predetermined number of times by shifting the time within the analysis frame, and a predictive gain calculating / comparing means (4) is provided to calculate the predictive gain. The comparing means calculates the prediction gain using the input linear prediction coefficient, and then the calculated prediction gain is a predetermined value or more than the previously calculated prediction gain measurement coefficient.
A linear predictive coding method characterized by transmitting a linear prediction coefficient and analysis position information at the time of detection when a change is detected. 3. The linear predictive coding system according to claim 1, wherein the analysis frame is equally divided into a plurality of subframes,
A linear predictive coding method characterized by performing a linear predictive analysis for each subframe and transmitting a subframe number as the analysis position information. 4. A decoder for reproducing a speech waveform from a linear predictive coded signal generated by using the method according to claim 1, wherein a linear interpolating means (5) is provided, and the linear predictive coefficient is updated in the analysis frame. Up to the point of time, the linear interpolating means uses the linear predictive coefficient generated by interpolating the linear predictive coefficient before and after the update by utilizing the received analysis position information. Decryption method.