JP3074703B2 - Multi-pulse encoder - Google Patents

Description

Description: BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a multi-pulse encoding device for performing high-efficiency encoding of a speech signal.

[Summary of the Invention]

The present invention compares a plurality of audio information obtained by supplying a drive pulse to a plurality of synthesis filters with input audio information, selects one synthesis filter and a pair of corresponding drive pulses, and By encoding information related to the synthesis filter and the driving pulse, it is possible to compress and encode the audio signal at a low bit rate, and to obtain a high quality synthesized sound by voice synthesis even at a low bit rate. It is intended to provide a multi-pulse encoding device capable of performing the following.

[Conventional technology]

As a conventional speech analysis / synthesis system (vocoder), for example, there is a speech analysis / synthesis system using so-called linear prediction analysis (LPC). In the coding of a speech signal using the LPC analysis / synthesis system, a so-called multi-pulse drive linear prediction coding (MPC or MPEC) is an example of a coding method that can obtain a high-quality synthesized voice. is there. This M
In the above-mentioned LPC analysis / synthesis system, the PC avoids the modeling of the sound source generally performed by the pulse and the noise, and expresses the sound source by a plurality of pulses irrespective of voiced and unvoiced sounds, thereby driving the LPC synthesis filter. Is the way.

FIG. 2 shows a block diagram of a conventional speech synthesis circuit using the MPC.

In FIG. 2, a signal P (n) indicating the amplitude and position of a pulse described later is supplied to an input terminal 101 and sent to a multi-pulse generation circuit 102. In the multi-pulse generation circuit 102, the signal P (n) of the amplitude and position of the pulse
, A multi-pulse train V (n) is generated. By driving the long-term prediction synthesis filter 103 and the short-term prediction synthesis filter 104, which are LPC synthesis filters, with the multi-pulse train V (n), a synthesized speech signal Q (n) can be obtained. An output terminal of the synthesized voice signal Q (n)
Output from 105.

Here, the filter characteristic F _SL (Z) of the long-term prediction synthesis filter 103 can be expressed by F _SL (Z) = 1 / (1 + βZ− ^M ) (1) In the equation (1), M is the number of delay taps of the long-term prediction synthesis filter 103, and β is a prediction coefficient. Also,
Filter characteristics F _SS (Z) of the short-term prediction synthesis filter 104
_{Can be} expressed as F _SS (Z) = 1 / (1 + α ₁ Z ⁻¹ + α ₂ Z ⁻² ...) (2) In the equation (2), α is a prediction coefficient.

However, in the case of speech synthesis using general MPC, the long-term prediction synthesis filter 103 is not always necessary, but by using the long-term prediction synthesis filter 103 as shown in FIG. 2, a long-term correlation of speech is obtained. Therefore, the sound quality can be improved, and the energy of the multi-pulse can be reduced.

By the way, each filter characteristic F _SL (Z) of the long-term prediction synthesis filter 103 and the short-term prediction synthesis filter 104 described above and
F _SS (Z) is determined using, for example, a configuration as shown in FIG.

An input audio signal S (n), which is an original audio signal, is supplied to the input terminal 111 in FIG. Here, the prediction coefficient α _i of the short-term prediction synthesis filter 104 in FIG.
(I is 1, 2, 3,...) Is, for example, the input audio signal S (n)
It is obtained by performing LPC analysis. Further, the number M of delay taps and the prediction coefficient β of the long-term prediction synthesis filter 103 in FIG.
The above-mentioned input audio signal S (n) is input to a short-term prediction analysis filter 112 having a filter characteristic F _AS (Z) expressed by F _AS (Z) = 1 + α ₁ Z ⁻¹ + α ₂ Z ⁻² (3) It is obtained by analyzing the output R ₁ (n) when input. As an analysis method at this time, for example, a long-term prediction analysis filter 113 of a filter characteristic F _AL (Z) of F _AL (Z) = 1 + βZ− ^M (4)
In order to minimize the sum of squares of the output R ₂ (n) when the output R ₁ (n) of the short-term prediction analysis filter 112 is input,
That is, Is minimized, and the prediction coefficient β and the number M of delay taps are obtained. The output R ₂ (n) is connected to the output terminal 11
Output from 4.

Thus, the filter characteristics F _AL (Z) and F
Long-term prediction synthesis filter 10 of the second _{view SS} (Z) is determined
3 and the short-term prediction synthesis filter 104
The MPC is driven by (n).

Further, in the above-mentioned speech synthesizer, for example, the synthesized speech signal Q
Multi-pulse train V such that (n) approaches the desired value
The search of (n) is performed, and speech is synthesized by the multi-pulse train V (n) obtained by the search. An algorithm for searching for the multi-pulse train V (n) at this time is, for example, a so-called AbS
There is a method using the principle of the (analysis-by-synthesis) method. In this method, the input audio signal S (n) and the synthesized audio signal Q
After calculating an error from (n) and performing weighting based on auditory characteristics, a search for the multi-pulse train V (n) that minimizes the mean square error is performed.

That is, in the configuration of FIG. 4, assuming that a certain number of pulses (multi-pulse train V (n)) have already been determined as an initial state, the multi-pulse train V (n) becomes
The signal is converted into a synthesized speech signal Q (n) through the LPC synthesis filter 123 composed of the above-described long-term prediction synthesis filter and short-term prediction synthesis filter. This synthesized speech signal Q (n)
Is the input audio signal S (n) supplied from the terminal 126
Is subtracted from the synthesized speech signal Q (n) and sent to a subtractor 124 that obtains an error signal e (n). Thereafter, the error signal e (n) is sent to a weighting filter 125 for performing perceptual weighting (weighting with a weighting coefficient W (z)), and based on the output e _w (n), a square error minimizing circuit is used. At 121, a new pulse amplitude and position signal P (n) is determined so that the square error is minimized, and sent to the multi-pulse generation circuit 122. That is, in the configuration of FIG. 4, an error signal e () between the synthesized voice signal Q (n) based on the multi-pulse train V (n) determined so far (for example, in the initial state) and the input voice signal S (n). n)
A new multi-pulse train V (n) is added, and the error signal e (n) becomes smaller than a preset value or until a predetermined number of pulses (multi-pulse train V (n)) is determined. The processing is repeated.

[Problems to be solved by the invention]

In the conventional MPC system, as described above, for example, prior to searching for the multi-pulse train V (n) using the principle of the AbS method, the filter characteristics of the long-term prediction synthesis filter and the short-term prediction synthesis filter are used. However, there is no optimal method for determining the filter characteristics of the long-term prediction synthesis filter, and it is particularly difficult to determine the number M of delay taps. For this reason, it is difficult to say that the synthesized speech obtained by the speech synthesizer is good.

Therefore, the present invention has been proposed in view of the above-described circumstances, and provides a multi-pulse encoding device capable of performing compression encoding of an audio signal and easily obtaining an optimal synthesized audio signal. It is intended to do so.

[Means for solving the problem]

A multi-pulse encoding apparatus according to the present invention has been proposed to achieve the above-described object, and includes a plurality of synthesis filters for performing long-term prediction synthesis and short-term prediction synthesis, and a driving pulse corresponding to the plurality of synthesis filters. Generating means for comparing the input audio information with audio information obtained by supplying the driving pulse generated by the driving pulse generating means to the corresponding synthesis filter, and comparing one of the plurality of synthesis filters with the input audio information. And a coding unit for coding information relating to the synthesis filter and the driving pulse selected by the comparison / selection unit.

[Action]

According to the present invention, speech information (synthesized speech information) from a synthesis filter is compared with input speech information, and information related to a synthesis filter based on the comparison result and a driving pulse corresponding thereto is encoded. Therefore, the synthesized speech obtained from the encoded output is the best synthesized speech.

〔Example〕

Hereinafter, embodiments of the present invention will be described with reference to the drawings.

FIG. 1 is a block diagram showing a schematic configuration of a multi-pulse encoder according to an embodiment of the present invention.

The first diagram of an encoding apparatus, (m is an integer) the long-term prediction synthesis filter 3 ₁ to 3 _m as a plurality of synthesis filters and the short-term prediction synthesis filter 4 ₁ to 4 _m, corresponding to the plurality of synthesis filters a multi-pulse generating circuit 2 ₁ to 2 _m, which is a driving pulse generating means, said multi-pulse generation circuit 2 ₁ to 2 the generated drive pulse by _m (multi-pulse train V (n) ₁ ~
V (n) _m ) to the corresponding synthesis filter, and obtains synthesized voice signals Q (n) _{1 to} Q (n) _m which are voice information obtained.
And the input audio signal S (n), which is the input audio information, to obtain one of the plurality of synthesis filters from which the best synthesized audio signal Q (n) _g (g is any one of 1 to m) is obtained. And a comparison processing unit 11 for selecting a corresponding pair of drive pulses.
And a selection processing unit 12 and a synthesis filter selected by the comparison and selection circuit 10 and a delay tap number M _g and a prediction coefficient β _g of a long-term prediction synthesis filter as information related to the driving pulse. And a coding circuit 6 for coding the multi-pulse train V (n) _g and the prediction coefficients of the short-term prediction synthesis filter.

That is, the present embodiment shown in FIG. 1 device, there is performing voice synthesis using MPC, by using the long-term prediction synthesis filter 3 ₁ to 3 _m, so take a correlation of the speech long-term To improve the sound quality and to reduce the energy of the multi-pulse. Here, each of the long-term prediction synthesis filter 3 ₁ to 3 _m is described above (1) beta prediction coefficients and delay taps M in the formula that are different from each other, each different filter characteristics F _SL (Z) _{1 to} F
_SL (Z) _m . Further, each filter characteristic of the short-term prediction synthesis filter 4 ₁ to 4 _m is respectively a same filter characteristic F _SS (Z). Incidentally, the short-term prediction synthesis filter 4 ₁ to 4 _m each filter characteristic F _SS of (Z) are those that can be represented by the aforementioned formula (2). Furthermore, from the multi-pulse generating circuit 2 ₁ to 2 _m, as FIG. 4 described above, the corresponding long-term prediction synthesis filter 3 ₁ to 3 in the _m optimal multi-pulse train _V (n) 1 _~V ( n) _m is generated. For example, from these multi-pulse generating circuit 2 ₁ to 2 _m, the filter characteristic of the long-term prediction synthesis filter in LPC synthesis filter 123 shown in FIG. 4 described above, the long-term prediction synthesis filter 3 ₁ of this embodiment 3 _m filter characteristics F _SL (Z) _{1 to} F _SL (Z) _m
Using the LPC synthesis filter 123 provided with the long-term prediction synthesis filters designated as F _SL (Z) _{1 to} F _SL (Z) _m , the optimum multi-pulse trains respectively selected in the configuration of FIG. 4 can be obtained. It has become. Thus the multi-pulse train _{V (n) 1 ~V (n} ) m thus obtained is, long-term prediction synthesis filter 3 ₁ to 3 _m and short-term prediction synthesis filter 4 ₁ to 4 _m is a LPC synthesis filter of Figure 1 , Synthetic speech signals Q (n) _{1 to} Q (n) _m can be obtained. These synthesized speech signals Q (n) _{1 to} Q (n) _m are sent to the comparison and selection circuit 10.

In this embodiment apparatus, because the filter characteristics of the long-term prediction synthesis filter 3 ₁ to 3 _m are different from each as described above, in the output or outputs of the synthesis filters in each short-term prediction synthesis filter 4 ₁ to 4 _m A certain synthesized voice signal Q (n) ₁
To Q (n) _m are also different from each other. Among these synthesized speech signals Q (n) _{1 to} Q (n) _m , the best synthesized sound is selected by the comparison and selection circuit 10.

Here, in the comparison processing section 11 of the comparison and selection circuit 10, the synthesized voice signal Q which is closest to the input voice signal S (n) is output.
(N) Perform a process to select _g . Specifically, for example, an error signal (an error signal with respect to the input audio signal S (n)) between each synthesized audio signal Q (n) _{1 to} Q (n) _m and the input audio signal S (n) via the terminal 5. ), And a synthesized speech signal Q (n) _g that minimizes the square error is selected based on these error signals. In other words, the smallest square error indicates that the synthesized speech signal Q (n) _g is closest to the input speech signal S (n). The comparison result in the comparison processing unit 11 is sent to the selection processing unit 12.

Moreover, the selection processing unit 11 of the comparison selection circuit 10, the multi-pulse train V from the multi-pulse generating circuit 2 ₁ to 2 _m
(N) 1 ~V and _{(n) m,} the long-term prediction synthesis filter 3 ₁ -
Information on the delay tap numbers M _{1 to} M _{m of} 3 _m and the prediction coefficients β _{1 to} β _m is also supplied. In the selection processing unit 12, according to the comparison result of the comparison processing unit 11, the optimal synthesized speech signal Q
(N) The number of delay taps _Mg and the prediction coefficient β _{g of the} long-term prediction synthesis filter from which _g can be obtained and the information on the corresponding multi-pulse train V (n) _g of the multi-pulse generation circuit are selected and output. .

After the information of the delay tap number M _g , the prediction coefficient β _g and the multi-pulse train V (n) _g , and the information of the prediction coefficient of the short-term prediction synthesis filter from the terminal 9 are sent to the encoding circuit 6 and encoded. , From the output terminal 7 as an output signal C (n).

From the above, in the apparatus of this embodiment, when determining a long-term prediction synthesis filter used in multi-pulse encoding (determining filter characteristics), a plurality of long-term prediction synthesis filters having different filter characteristics are prepared. In advance, a signal from which an optimum synthesized sound can be obtained is selected from synthesized speech signals obtained using these long-term prediction synthesis filters, and the selected synthesized speech signal Q (n) _g is selected. obtaining information of the corresponding long-term prediction delay tap number M _g and prediction coefficients of the synthesis filter beta _g, the output of the multi-pulse generating circuit corresponding to the long-life prediction synthesis filter (multi-pulse train V _(n) g), of The information is encoded and output together with the prediction coefficients of the short-term prediction synthesis filter. Therefore, if speech is later synthesized based on the encoded output, a good synthesized speech can be obtained. As described above, even when encoding is performed at the same bit rate as in the conventional example, the sound quality of the synthesized sound based on the encoded output of the apparatus of the present embodiment is improved. Furthermore, since the filter characteristics of the long-term synthesis filter 3 ₁ to 3 _m even the best one has to be selected, the quality of the synthesized speech is improved.

〔The invention's effect〕

In the multi-pulse encoding device of the present invention, a plurality of synthesis filters for performing long-term prediction synthesis and short-term prediction synthesis,
A plurality of audio information obtained by supplying drive pulses from the drive pulse generating means corresponding to the plurality of synthesis filters to the corresponding synthesis filter is compared with input audio information,
By selecting one synthesis filter and a corresponding drive pulse pair, it is possible to select the optimum characteristics of the synthesis filter and to encode information relating to the selected pair of synthesis filter and drive pulse. By,
An audio signal can be compression-coded at a low bit rate, and a high-quality synthesized sound can be easily obtained by voice synthesis even at a low bit rate.

[Brief description of the drawings]

FIG. 1 is a block circuit diagram showing a schematic configuration of an apparatus according to an embodiment of the present invention, FIG. 2 is a block circuit diagram showing a conventional synthesis circuit, and FIG. 3 is a configuration for determining filter characteristics. FIG. 4 is a block circuit diagram showing a configuration for searching for a multi-pulse train. 6 ...... encoding circuit 2 ₁ to 2 _m ...... multi-pulse generating circuit 3 ₁ to 3 _m ...... long-term prediction synthesis filter 4 ₁ to 4 _m ...... short-term prediction synthesis filter 10 ...... Comparison selection circuit 11 ...... comparison Part 12 …… Selection processing part

────────────────────────────────────────────────── ─── Continued from the front page (58) Fields surveyed (Int. Cl. ⁷ , DB name) G10L 11/00-13/08 G10L 19/00-21/06 INSPEC (DIALOG) JICST file (JOIS) WPI ( DIALOG)

Claims (1)

(57) [Claims] 1. A plurality of synthesis filters for performing long-term prediction synthesis and short-term prediction synthesis, driving pulse generating means corresponding to the plurality of synthesis filters, and a synthesis filter corresponding to the driving pulse generated by the driving pulse generating means. Comparing the audio information obtained by supplying the input audio information with the input audio information, and selecting one of the plurality of synthesis filters and a pair of the corresponding drive pulse. A multi-pulse encoding device comprising: a synthesis filter; and encoding means for encoding information related to the drive pulse.