JPH058839B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to improving the efficiency of multipulse encoding in a multipulse encoding/decoding device.

[Conventional technology]

The sound source information of the input audio signal is modeled using multiple impulse sequences, so-called multipulses, and is encoded together with other audio data such as feature parameters, sent from the analysis side to the synthesis side, and decoded to generate the input audio. Multi-pulse encoding/decoding devices for reproducing signals are well known, and recently, multi-pulse encoding/decoding devices equipped with pitch prediction means have been increasingly used to improve multi-pulse encoding efficiency. It is also well known.

A multi-pulse encoding/decoding device equipped with such a pitch prediction means has a repeatability in that the multi-pulse obtained for each analysis frame is generally repeated at a certain pitch period. This method performs a search and uses the pitch information obtained through this search to reduce the transmission bit rate of encoded audio information that should be sent from the analysis side to the synthesis side.This pitch prediction means uses multipulse encoding. has become an effective means of significantly improving efficiency.

However, the conventional multi-pulse encoding/decoding apparatus equipped with this type of pitch prediction means has the disadvantage that the amount of processing is large and the scale of the hardware is also large.

[Purpose of the invention]

An object of the present invention is to prepare a section corresponding to a maximum pitch period shorter than the length of the analysis frame within one analysis frame, set a certain number of multipulses in this section, and perform analysis in a multi-pulse encoding/decoding device. Other sections within the frame are provided with a means to alternatively represent section information with similar multi-pulse characteristics within the section corresponding to the maximum pitch period, which greatly reduces the amount of processing and greatly simplifies the scale of the hardware. The object of the present invention is to provide a multi-pulse encoding/decoding device that can perform

[Structure of the invention]

The multi-pulse encoding/decoding device of the present invention includes:
A maximum pitch interval corresponding to a maximum pitch period that is preset as a time length exceeding the maximum pitch period that appears in real speech is set within one analysis frame, and a certain number of multipulses are set in this maximum pitch interval, and further, as described above. A multi-pulse setting means is provided for alternatively expressing a multi-pulse to be set in a region other than the maximum pitch section within one analysis frame by the content of a similar section having the highest degree of similarity in the maximum pitch section.

〔Example〕

Next, the present invention will be explained in detail with reference to the drawings.
FIG. 1 is a block diagram showing the configuration of the encoding side (analysis side) of a multi-pulse encoder/decoder according to an embodiment of the present invention, and FIG. 2 is a block diagram showing the configuration of a multi-pulse encoder/decoder according to an embodiment of the present invention. FIG. 2 is a block diagram showing the configuration of the decoding side (analysis side) of the encoding/decoding device.

The encoding side in FIG. 1 includes an LPC analyzer 1, an encoder 2, a cross-correlation function calculator 3, an autocorrelation function calculator 4, a multi-pulse searcher 5, a multi-pulse setter 6, an encoder 7, and a multiplexer. 8, etc., and the decoding side shown in FIG.
It is configured with an LPF (Low Pass Filter) 15, etc.

Now, there are two general methods for obtaining multi-pulses representing a sound source pulse train, as follows. One of them is A-b-S by BSAtal etc.
(Analysis−by−Synthqsis) processing method,
The other one is the correlation function processing method by Ozawa, Araseki, Ono, and others. Of these two processing methods, the former is spectral domain evaluation, and the latter is so-called correlation that utilizes the cross-correlation between the input audio signal and the impulse response of the synthesis filter in the analysis stage, as well as the autocorrelation of the impulse response. This is region evaluation, and the encoding and decoding sides of the embodiments shown in FIGS. 1 and 2 utilize the correlation region evaluation of the latter. There is no problem at all.

In Figure 1, LPC analyzer 1 has input terminal 1.
The input audio signal input through 01 is quantized and analyzed by LPC (Linear Prediction Coe-
fficient, linear prediction coefficient) and extracts LPC coefficients such as the p-order K parameter (partial autocorrelation coefficient),
This is supplied to the encoder 2.

The encoder 2 quantizes and encodes the input LPC coefficients, then sends them to the multiplexer 8, and decodes the quantized LPC coefficients,
An auditory weighted impulse response of a vocal tract filter, which will be described later, is obtained and supplied to a cross-correlation function calculator 3 and an autocorrelation function calculator 4.

The input audio signal for each analysis frame quantized by the LPC analyzer 1 is sent to the cross-correlation function calculator 3.
Also supplied.

The input audio signal waveform S(Z) for each analysis frame supplied to the cross-correlation function calculator 3 has a transfer function W
Perform the convolution integral S(Z)*W(Z) by passing through a noise filter of (Z). In this case, the transfer function W(Z) can be set using the LPC coefficient of the input audio signal and its order, as well as an auditory weighting coefficient, etc., and by this processing, the input audio signal is After weighting is applied, a cross-correlation with an impulse response of H(Z)*W(Z) is performed and this is supplied to the multipulse searcher 5. This H(Z)*W(Z) is obtained by giving auditory weighting to the transfer function H(Z) of the vocal tract filter through convolution with the transfer function W(Z) of the noise filter.

The autocorrelation function calculator 4 calculates the autocorrelation function of the weighted impulse response of the vocal tract filter mentioned above and supplies it to the multipulse searcher 5.

The multi-pulse searcher 5 utilizes the cross-correlation function and auto-correlation function for each analysis frame thus input, and basically performs the calculation shown in the following equation (1) to determine the multi-pulse.

In equation (1), m _i is i within the analysis frame.
The time position of the th pulse from the frame edge, g _i is its amplitude, _hx (m _i ) is the interaction function at time delay m _i , g _l is the amplitude of the l th pulse in the analysis frame, R _hh (|m _l −m _i |) is an autocorrelation function.
Equation (1) means that the amplitude g _i (m _i ) can be obtained by finding the difference between the cross-correlation function and the autocorrelation function, and that if such a pulse is generated at the time position m _i , the amplitude g _i (m _i ) can be determined as the optimal one. In other words, among the pulses obtained by focusing on a certain sound source pulse and calculating its amplitude using equation (1) at various time positions, the one with the maximum absolute value of the amplitude according to equation (1) is the closest approximation to the sound source pulse. This process is repeated to obtain a plurality of sound sources, which are used as multipulses.

In this example, the length of one analysis frame is 22.5 m.
While performing a multi-pulse search based on equation (1) over SEC, a preset constant value is searched within an interval of 15 mSEC as an intra-frame interval corresponding to the maximum pitch period of the input audio signal set within this analysis frame. number of multi-pulses, in this example 8
A multipulse search is performed over the entire frame interval until 100 multipulses are set, and the multipulse for each analysis frame thus obtained is supplied to the multipulse setting device 6. In this case, multi-pulses are set approximately on average for all analysis frames in all frame sections including the 7.5 mSEC section of the analysis frame.

Now, the maximum pitch period included in the input audio signal is at most 10 mSEC, and as in this example, it is 15 mSEC.
If SEC is also provided, it is safe to assume that any pitch period will be included within this interval. In addition, the section of the analysis frame, in this example, is 22.5 m.
As mentioned above, the multipulse in the section of SEC - 15mSEC = 7.5mSEC has multipulse characteristics that are very similar to any section of the multipulse set in the section of 15mSEC,
Therefore, the multipulse information in this section is 15mSEC
It becomes possible to replace the interval with the similar interval.

Note that the above-mentioned 15mSEC interval has a fixed number of multipulses for each analysis frame, eight multipulses in this example, but this number depends on the transmission capacity from the encoding side to the decoding side and the amount of distortion. This can be set arbitrarily, taking into account the following conditions. In this way, the time length of 15
In the mSEC section, eight multipulses are expressed, and the eight multipulses express the multipulse train of all patterns up to the maximum pitch period. Remaining interval of analysis frame 7.5
15 as part of the multipulse search range for mSEC as well.
Multipulses are set in the same way as the mSEC section, but the multipulse that should be set in this 7.5mSEC section is necessarily similar to a part of the multipulse that is set in the 15mSEC section. It is obvious that this characteristic has the following characteristics when considering the repeatability in multi-pulses.

The multipulses for each analysis frame searched by the multipulse searcher 5 are sequentially supplied to the multipulse setter 6.

Multi-pulse setting device 6 has an analysis frame length of 22.5m.
The built-in correlation calculation circuit takes the autocorrelation between the multi-pulse train set in the 15mSEC section and the multipulse train set in the 7.5mSEC section, and finds a similar section to 7.5mSEC that exists in the 15mSEC section. The interval information is obtained when the autocorrelation function obtained by performing convolution integration of both while temporally shifting 7.5m corresponding to the 15mSEC interval is maximum, that is, when the similarity is maximum. is used as alternative multipulse information for the 7.5mSEC section, and this and the above 15m
The multipulse information of the SEC section is supplied to the encoder 7 via an output line 601 as multipulse setting information.

In addition, when calculating the autocorrelation between the 15mSEC interval and the 7.5mSEC interval, it is possible to easily calculate the pitch period of the multipulse train and the ratio of the pitch period to the analysis frame length, the so-called pitch gain, in this process. These are supplied as equivalent pitch prediction information to the encoder 7 via output line 602, encoded together with the multipulse setting information supplied via output line 601, and sent to the multiplexer via output lines 701 and 702, respectively. 8 is supplied.

The multiplexer 8 combines the thus supplied LPC coefficient information, multipulse setting information, and equivalent pitch prediction information, performs multiplexing processing in a predetermined manner, and then transmits the information via a transmission path 801 as shown in FIG. Transmit to the decoding side.

On the decoding side, a demultiplexer 9 demultiplexes the multiplexed signal, and sends LPC coefficient information to a decoder 10, equivalent pitch prediction information to a decoder 11, and multipulse setting information to a decoder. Vessel 1
2 and decoded.

The multi-pulse regenerator 13 regenerates a multi-pulse train over analysis frames based on the decoded equivalent pitch prediction information and multi-pulse setting information, and supplies this to the LPC synthesizer 14.

The LPC synthesizer 14 is an all-pole type digital filter, which uses the LPC coefficients based on the LPC coefficient information received from the decoder 10 as filter coefficients, and uses the multipulse supplied from the multipulse regenerator 13 as a filter driving sound source to receive the input audio signal. After playing and converting it to D/A (digital/analog),
The signal is supplied to the LPF 15, subjected to predetermined filtering, and then output to the output terminal 1501.

In this way, highly efficient encoding and decoding can be achieved with a significant improvement in processing amount and hardware scale.

In addition, in the above-mentioned embodiment, in order to substitute and express multi-pulses in the section corresponding to the maximum pitch period from one analysis section, autocorrelation is used to determine the similar section between the two sections. However, this method uses other methods that have almost the same effect, such as AMDF (Average Magnitude
It is clear that it can be implemented in the same way even if it is replaced with (Differential Function) etc.

[Effects of the Invention] As explained above, according to the present invention, in a multi-pulse encoding/decoding device having pitch prediction means, an interval corresponding to a maximum pitch cycle shorter than the analysis frame length is set within one analysis frame. By preparing and setting a certain number of multi-pulses in this, and providing means for alternatively representing this with section information having similar multi-pulse characteristics within the section corresponding to the maximum pitch period for other sections within the analysis frame. This has the effect of realizing a multi-pulse encoding/decoding device that can significantly reduce the amount of processing and greatly simplify the scale of hardware.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing the configuration of the encoding side of the multi-pulse encoding/decoding apparatus according to the present invention, and FIG. 2 is a block diagram showing the configuration of the decoding side. 1...LPC analyzer, 2...encoder, 3...
Cross-correlation function calculator, 4... Autocorrelation function calculator, 5... Multi-pulse searcher, 6... Multi-pulse setter, 7... Encoder, 8... Multiplexer, 10, 11, 12... Decoder, 13...
Multipulse regenerator, 14...LPC synthesizer, 1
5...LPF.

Claims (1)

[Claims] 1. A maximum pitch interval corresponding to the maximum pitch period, which is preset as a time length exceeding the maximum pitch period that appears in real speech, is set within one analysis frame, and a certain number of multipulses are set in this maximum pitch interval, and The multi-pulse type is characterized by comprising a multi-pulse setting means for alternatively expressing a multi-pulse to be set in a region other than the maximum pitch section within the one analysis frame with the content of a similar section having the highest degree of similarity in the maximum pitch section. Encoding/decoding device.