JP2820117B2 - Audio coding device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

[0001] 1. Field of the Invention [0002] The present invention relates to a speech coding apparatus, and more particularly to a speech coding apparatus using psychological auditory analysis.

[0002]

2. Description of the Related Art FIG. 3 is a block diagram showing a configuration of a conventional general speech coding apparatus.

[0003] Conventionally, a speech coding method using a psycho-auditory analysis function has performed coding on a frame-by-frame basis.
The input audio signal data of the Nth frame to which 1 is input is divided into frequency domain components. On the other hand, the psychological hearing analysis unit 17
As shown in the explanatory diagram illustrating the configuration of the input audio signal data in FIG. 4, when encoding data in the Nth frame, the input audio in the A region from the (Ni) frame to the Nth frame is encoded. The signal data is input to the spectrum calculator 14 to perform a frequency analysis, and in consideration of the result of the frequency analysis and the masking effect of the human auditory characteristics, a masking curve predictor 15 calculates a masking curve. Based on this, the quantization step width is predicted by the quantization step width predictor 16, and the quantizer 12 quantizes the data output from the frequency division filter bank 11 with the predicted quantization step width.

[0004] Practical examples of use include, for example, ISO / IEC JTC1 / SC29 / W which is an international standardization organization.
Working Group of G11 (MPEG; Moving Pict
MPEG Audio Layer, which is one of the high-efficiency compression schemes for moving images, which is formulated by our Experts Group.
The speech signal input data to the psychological auditory analysis unit used in I / II / III is the data of the Nth frame and the data of the (N-1) th frame when the speech input signal of the Nth frame is encoded. Is entered. Other encoding methods using psychological auditory analysis include ATRAC speech encoding methods used in MDs (mini discs) and PRACs used in DCCs (digital compact cassettes).
ASC audio coding system and the like can be mentioned.

[0005] Here, the masking effect and how to use it will be described. The masking effect includes simultaneous masking and successive masking. The successive masking includes forward and backward masking.

[0006] Simultaneous masking refers to a case in which a masking sound (masker) and a masking sound (masky) are presented simultaneously.

The forward masking of successive masking is the masking of a temporally preceding sound to a succeeding sound, and the backward masking is the masking of a subsequent sound to a preceding sound. Mr. Kuwano of Osaka University reported the amount of each masking in JAS Journal '93.
This is shown in a paper published as "Aural Psychology and Audio" in the June issue, pages 13-25. A part is shown as Table 1 below.

[0008]

[Table 1]

From the masking curve predicted by the masking curve predictor 15, the quantization step width for the frequency component data having a large masking amount in the human auditory characteristics is coarse, and the quantization step width for the frequency component data having a small masking amount is fine. The result of the allocation and the step width is output from the quantization step width predictor 16.

[0010]

The conventional speech coding apparatus described above, when coding the data of the Nth frame,
Since the psychoacoustic analysis is performed using the data before the data of the frame to be coded, the masking curve is determined without using only the forward masking effect among the masking effects. For this reason, since the obtained masking curve does not analyze all data actually masked, it cannot be said that an optimum masking curve is calculated, and there is a problem that coding efficiency is poor.

It is an object of the present invention to provide a speech coding apparatus capable of increasing coding efficiency with the same amount of calculation.

[0012]

SUMMARY OF THE INVENTION A speech encoding apparatus according to the present invention converts input speech signal data, which is divided into a plurality of frames of a fixed length and input via an input terminal, into frequency-divided data for each frame. Frequency dividing means for receiving a plurality of frames of the predetermined length and performing spectrum analysis for each frame, and comparing the latest frame with the latest frame and i (i = 1, 2,... Psychological auditory analysis means for calculating the quantization step width using the result of the spectrum analysis of the frame and the human auditory characteristics including the effect of masking, and the psychological auditory analysis means calculates the frequency-divided data by the frequency dividing means. And a multiplexing means for multiplexing the quantized data quantized by the quantizing means into an encoded bit sequence. Wherein the input audio signal data storage means for temporarily storing the input audio signal data is provided between the input terminal and the frequency division means and the psychological hearing analysis means, wherein the psychological hearing analysis means Are i before and after sandwiching a frame whose quantization step width is to be calculated from the input audio signal data storage means.
Receiving the number of frames, calculating the quantization step width using the result of the spectrum analysis of the frame for which the quantization step width is to be calculated and the human auditory characteristics including the effect of masking, and outputting the result to the quantization means. It is.

In the speech coding apparatus according to the present invention, the frequency division means may be a sub-band division filter bank.

In the speech coding apparatus of the present invention, the frequency dividing means may be a modified discrete cosine transform (MDCT).

The speech coding apparatus according to the present invention is characterized in that a psychoacoustic analysis means receives a plurality of frames of a fixed length from the input speech signal data storage means and performs a frequency analysis for each frame. Consists of a masking curve predictor that calculates the masking curve in consideration of the result and the masking effect that is a human auditory characteristic, and a quantization step width predictor that calculates the quantization step width from the masking curve calculated by the masking curve predictor May be.

[0016]

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

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

The speech encoding apparatus 10 of the present invention comprises: an input data memory 8 for temporarily storing input speech signal data which is divided into a plurality of frames of a fixed length and inputted via an input terminal 9; A frequency-division filter bank 1 that is data that is frequency-divided for each frame, and i frames before and after a frame for which a quantization step width is to be calculated are received from an input data memory 8, and a spectrum analysis of the corresponding frame is performed. Psychological auditory analysis unit 7 that calculates the quantization step width using the result and the human auditory characteristics including the effect of masking, and quantization that psychological auditory analysis unit 7 calculates the frequency-divided data by frequency division filter bank 1. A quantizer 2 for quantizing with a step width, and a multiplexer 3 for multiplexing the quantized data quantized by the quantizer 2 into an encoded bit sequence. It is equipped with a. Further, the psychological auditory analysis unit 7 receives a plurality of frames of a fixed length from the input data memory 8 and performs a frequency analysis for each frame, a calculation result of the spectrum calculator 4, and human auditory characteristics. It includes a masking curve estimator 5 for obtaining a masking curve in consideration of a masking effect, and a quantization step width estimator 6 for obtaining a quantization step width from the masking curve obtained by the masking curve estimator 5.

Next, the relationship between the data input to the spectrum calculator 4 of the psychological auditory analyzer 7 in FIG. 1 and the input audio signal data actually encoded will be described with reference to the configuration of the input audio signal data in FIG. This will be described with reference to an explanatory diagram. The names and reference numerals shown in FIG. 1 are used.

The input audio signal data input to the spectrum calculator 4 is as follows when encoding the Nth frame.
From the (N-j) frame that contains the Nth frame inside, (N +
k) The input audio signal data in the B region up to the frame is used. The input audio signal data in this case is (i + 1) × n
Consists of samples. Here, i = j + k.

The (i + 1) × n input audio signals represented by the B region are input to the spectrum calculator 4 to perform frequency analysis, and the analysis result is used as a masking curve predictor 5
And calculate the masking curve. Subsequently, the quantization step width predictor 6 predicts the quantization step width based on the masking curve information, so that it is possible to use both the forward masking and the backward masking effects of the successive masking. Becomes

When a masking curve is obtained by using the same number of (i + 1) × n samples, it is possible to obtain a masking curve having the largest amount of masking by changing the ratio between forward masking and reverse masking. Yes, the quantization step width is predicted using the masking curve having the largest amount of masking, and the quantized data quantized with the predicted quantization step width is obtained when the data having the same data amount is encoded. Among them, the most efficient encoding is performed.

Although the above description has been made with reference to an example in which the frequency division means uses the frequency division filter bank 1 as the frequency division means, the operation is the same even if the frequency division means is a sub-band division filter bank device, and the most efficient operation is achieved. Can perform good coding.

Even if the frequency dividing means is a modified discrete cosine transform method (MDCT), the most efficient coding can be performed similarly.

[0025]

As described above, according to the present invention, when predicting the quantization step width, the conventional forward masking effect is obtained by using both the forward masking effect and the backward masking effect in combination. It is possible to obtain a larger amount of masking as compared with the case where a masking curve is calculated using only the effect of (1), and there is an effect that efficient coding can be performed.

[Brief description of the drawings]

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

FIG. 2 is an explanatory diagram illustrating a configuration of input audio signal data according to the first embodiment of this invention.

FIG. 3 is a block diagram showing a configuration of a conventional general speech encoding device.

FIG. 4 is an explanatory diagram illustrating a configuration of input speech signal data of a conventional general speech encoding device.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Frequency division filter bank 2 Quantizer 3 Multiplexer 4 Spectrum calculator 5 Masking curve predictor 6 Quantization step width predictor 7 Psychological auditory analysis part 8 Input data memory 9 Input terminal 10 Voice encoder

────────────────────────────────────────────────── ─── Continuation of front page (56) References JP-A-8-123490 (JP, A) JP-A-8-51366 (JP, A) JP-A-8-204575 (JP, A) JP-A-8-204 204574 (JP, A) JP-A-7-183818 (JP, A) JP-A-3-139923 (JP, A) JP-A-6-242797 (JP, A) JP-A-6-232825 (JP, A) JP-A-6-201744 (JP, A) JP-A-8-211899 (JP, A) (58) Fields investigated (Int. Cl. ⁶ , DB name) H03M 7/30 G10L 7/04 G10L 9/18

Claims (4)

(57) [Claims] 1. A frequency dividing means for dividing input audio signal data, which is divided into a plurality of frames of a fixed length and input through an input terminal, into frequency-divided data for each frame; The spectrum analysis is performed for each of the received frames, and the latest frame and i (i = 1, 1) before this latest frame are compared with the latest frame.
2,... N) psychological auditory analysis means for calculating a quantization step width using a result of spectrum analysis of frames and human auditory characteristics including a masking effect, and data obtained by frequency division by the frequency division means. A speech encoding apparatus comprising: a quantizing means for quantizing with the quantization step width calculated by the psychological auditory analysis means; and a multiplexing means for multiplexing the quantized data quantized by the quantizing means into an encoded bit string. In, provided input voice signal data storage means for temporarily storing the input voice signal data between the input terminal and the frequency division means and the psychological auditory analysis means,
The psychological auditory analysis means receives i frames before and after the frame for which the quantization step width is to be calculated from the input voice signal data storage means, and performs spectrum analysis of the frame for which the quantization step width is to be calculated. A speech coding apparatus comprising: calculating a quantization step width using a result and a human auditory characteristic including a masking effect and outputting the calculated quantization step width to the quantization unit. 2. The speech coding apparatus according to claim 1, wherein the frequency division means is a sub-band division filter bank unit. 3. The speech coding apparatus according to claim 1, wherein the frequency dividing means is a modified discrete cosine transform (MDCT). 4. A spectrum calculator which receives a plurality of frames of a fixed length from the input voice signal data storage means and performs frequency analysis for each frame, and a result of the spectrum calculator and human auditory characteristics. A masking curve estimator for obtaining a masking curve in consideration of a certain masking effect, and a quantization step width estimator for obtaining a quantization step width from the masking curve obtained by the masking curve estimator. 2. The speech encoding device according to claim 1.