JPS63282794A - Multi-pulse voice encoder - Google Patents

Abstract

(57) [Summary] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

[Detailed Description of the Invention] [Field of Industrial Application] The present invention relates to an encoder that encodes audio by analyzing audio signals frame by frame and extracting characteristic parameters thereof, and particularly relates to an encoder that encodes audio by analyzing audio signals frame by frame and extracting characteristic parameters thereof. The present invention relates to a multi-pulse audio encoder that encodes an audio signal expressed by a combination of pulse trains and an A' parameter of a synthesis filter.

[Conventional technology]

A specific example of a conventional multi-pulse speech encoder is shown in FIG. One frame has N samples, and the number of transmission pulses is M.・About Mars' quantization.

4' It is assumed that the position of the nocellus is quantized by the pit. It is assumed that the opposing decoder performs the decoding for each frame, which is obtained by further dividing one frame into i frames. Reference numeral 1 denotes a feature/grammeter analyzer. If this is an LPG analyzer, for example, the input audio signal is subjected to a Hamming window, cut out into one frame, and subjected to LPC analysis to obtain a PARCOR coefficient. 2 is the first quantizer, feature parameter analyzer 1
The PARCOR coefficients obtained are quantized and sent to a multiplexer (not shown). 3 is an inverse quantizer that inversely quantizes the PARCOR coefficients quantized by the first quantizer 2; 4 is an autocorrelator.

Impulse response of an LPC synthesis filter that converts the PARCOR coefficient obtained by the inverse quantizer 3 into a linear prediction coefficient (hereinafter referred to as α-grammeter) and uses the α-grammeter that has been audibly weighted as a coefficient. Find the normalized autocorrelation rhh. 5 is an inverse filter that inputs the audio signal and outputs the residual. Reference numeral 6 denotes a filter that uses the normalized autocorrelation Fhh obtained by the autocorrelator 4 as an impulse response, and inputs the residual obtained by the inverse filter 5 to perform an aurally weighted input audio signal and weights. Normalized cross-correlation with the impulse response of the filter synthesis filter? Find sh. Reference numeral 7 denotes a pulse searcher which searches for pulses using '5h and '4sh obtained by the autocorrelator 4 and filter 6. Here, the number of pulses to be sought is M per frame, but the number of pulses determined for each subframe is J? Based on the number of pulses, i9 pulses are searched in subframe units to find pulses. 8 is the second quantizer.

The M pulses found by the pulse searcher 7 are input, and the amplitude is quantized into m bits and the position is quantized into bits in units of subframes. First quantizer 2. The output of the second quantizer 8 is multiplexed by a multiplexer and sent out. Such a multinoculus speech encoder is described in two documents [
"Development of 1-step multipulse codec using ADsP" 1 Sato.

Minakami'' (1986 IEICE Communications Division National Conference A
185).

[Problem that the invention seeks to solve]

The multi-IE pulse speech encoder described above uses the Noculus searcher 7 to find M pulses within one frame. However, on the decoder side, one frame is further divided into i subframes, and some information is decoded in units of subframes (J? Even if the pulses are determined, the pulses cannot be directly transmitted to the M number of pulses, and the encoder may have to determine the pulses in subframe units. This causes deterioration in sound quality. Furthermore, since two types of Norse searchers must be provided depending on whether the pulse decoding method used by the decoder is in units of one frame or in units of subframes, there is a drawback that the scale of the encoder becomes large.

[Means for solving problems]

A multi-noise speech encoder according to the present invention comprises: a feature/parameter analyzer and a cross-correlator to which a speech signal is input; a first quantizer to which the output of the feature parameter analyzer is input; an inverse quantizer to which the output of the first quantizer is input; an autocorrelator to which the output of the inverse quantizer is input; and an autocorrelator to which the outputs of the cross-correlator and autocorrelator are input.
It is composed of a base pulse searcher, a second quantizer to which the output of the pulse searcher is input, and a pulse number controller to which the output of the second quantizer is input.

In addition, the pulse number controller specifically receives the output of the second quantizer or the output of the pulse reducer.The output of the base pulse number counter and the /J pulse number counter is input to the pulse number controller. An input pulse interpolator, a pulse number comparator to which the output of the pulse interpolator is input, and a Noculus number reducer and a Noculus position quantizer to which the output of the Noculus number comparator is input, respectively. It consists of

p1 lower days left 〔Example〕 Next, the present invention will be explained with reference to the drawings.

FIG. 1 is a block diagram of one embodiment of the present invention.

One frame has N samples, and the number of transmission pulses is M. I
? For Luss quantization, the Noculus amplitude is set to m bits,
Assume that the positions are quantized in bits.

Furthermore, the decoder opposite to the encoder of this embodiment decodes the information of the parent pulse in units of subframes obtained by further dividing one frame.

Features/Meter Analyzer 1 is 9 For example, in the case of an LPC analyzer, the input audio signal is cut out into one frame by applying a Hamming window, and LPC analysis is performed to obtain the PARCOR coefficient. The first quantizer 2 quantizes the PARCOR coefficients obtained by the feature parameter analyzer 1.

to a multiplexer (not shown). The dequantizer 3 dequantizes the PARCOR coefficients quantized by the first quantizer 2. The autocorrelator 4 uses P obtained by the inverse quantizer 3.
ARCOR coefficients are converted into linear prediction coefficients (hereinafter referred to as α parameters), and αA? is audibly weighted. The normalized autocorrelation≠hh of the impulse response of the PC synthesis filter is calculated using the parameter I as a coefficient.

Reference numeral 5 denotes a cross-correlator, which is composed of an inverse filter 6 that inputs an audio signal and outputs a residual, and a filter 7 that makes ψhh obtained by the autocorrelator 4 an innogulus response. The normalized cross-correlation ψ, h between the weighted audio signal and the impulse response of the weighted synthesis filter is determined.

The pulse searcher 8 performs a Noculus search using ψhh tψ8h obtained by the autocorrelator 4 and cross-correlator 5. The search for nof pulses is performed in one frame, and the number of pulses to be found is M. The second quantizer 9 inputs the M number of pulses determined by the pulse searcher 8, and calculates the amplitude of the pulse by m.
Quantize into bits. 10 is a Luss number controller, the second
The output Noculus of the quantizer 9 is inputted, and a pulse matching the pulse decoding of the opposing decoder is determined and sent to the multiplexer.

- The cell number controller 10 will be explained using figures. FIGS. 2 and 3 show a block diagram of an example of the bus pulse number controller of the present invention and pulse trains of each part. It is assumed that one frame is divided into two, for example, four subframes. It is assumed that the number of transmission pulses is determined to be two, for example three, in each frame. FIG. 3a is the output of the second quantizer. The output a is input to the pulse number counter 11 and is divided into 7'' frames as shown in FIG. 3, and the number of noculus in each subframe is counted. If 3. and 4, then the Nogle number is 1 to 3.+2 to 4.3 to 2.4 to 3.

The pulse train in Figure 3 is i? is input to the Lux interpolator 12,
If the amplitude is not O or there are two or more pulses (hereinafter referred to as effective pulses) within a subframe, the previous pulse is
A pulse with an amplitude of 0 (hereinafter referred to as a dummy pulse) is inserted at 2 from the mother pulse. This output is C (third
In Figure C), dummy A is indicated by Δ at +3? Luz is inserted. The output C is input to a pulse number comparator 13 and compared to see if each subframe is equal to a predetermined base pulse number.

+1. 3. +4 is equal, but +2 has one more pulse. If the determined number of base pulses is greater than the predetermined number of base pulses, it is input to the pulse reducer 15, and the base pulses are deleted in order from the smallest base pulse amplitude. This output d is shown in Figure 3.
As shown in , one pulse at the position indicated by the x mark is removed at +2. The output d is the base pulse number counter 11.
is input, and the above process is repeated.・The Norse number determined by the Norse number comparator 13 is determined by /? When the number of pulses becomes equal in all subframes, the obtained pulse train is input to the pulse position quantizer 14, and the pulse position is quantized in bits to determine the pulse amplitude.

Send the position. The pulse train sent out is as shown in FIG. 3e.

〔Effect of the invention〕

As explained above, the present invention provides: - a pulse number counter.

By having a pulse number controller consisting of a pulse interpolator, a Cruz number comparator, a pulse position quantizer, and a pulse reducer, the Cruz search process can be performed regardless of the Noculus decoding method in the decoder. This can be done frame by frame, thereby preventing deterioration in sound quality. Furthermore, the same encoder can be used for the pulse decoding process of the decoder on a frame-by-frame basis and on a sub-frame-by-subframe basis, which has the effect of providing compatibility.

[Brief explanation of drawings]

FIG. 1 is a block diagram of one embodiment of the present invention, and FIG. 2 is a block diagram of an embodiment of the present invention.
3 is a block diagram of the pulse number controller shown in FIG. 3, FIG. 3 is a diagram showing the base pulse train of each part in FIG. 2, and FIG. 4 is a block diagram of a conventional example. 1 is a feature/matrix meter analyzer, 2 is a first quantizer, 3
is an inverse quantizer, 4 is an autocorrelator, 5 is a cross-correlator, 6 is an inverse filter, 7 is a filter, 8 is a Hano Prus searcher, 9 is a second quantizer, 10 is /? Lux number controller. Figure 1 Figure 2 Figure 3

Claims (1)

[Claims] 1. A feature parameter analyzer and a cross-correlator to which the audio signal is input, a first quantizer to which the output of the feature parameter analyzer is input, and an output of the first quantizer to be input. an inverse quantizer, an autocorrelator to which the output of the inverse quantizer is input, a pulse searcher to which the outputs of the cross-correlator and the autocorrelator are input, and an output of the pulse searcher to which the output is input. A multi-pulse speech encoder comprising: a second quantizer; and a pulse number controller to which the output of the second quantizer is input.