JPH0766733A - Highly efficirent sound encoding device - Google Patents

Abstract

PURPOSE:To provide a highly efficient sound encoding device capable of reducing the omission of detection or post detection of transient and suppressing the generation of a preecho. CONSTITUTION:A transient detecting part 2 prevents the generation of omission in the detection of a case that transient exists on the center position of a segment by comparing segment power P [i] with P[i-1] and P[i-4] for instance. When the segment power P[i] is compared with P[i-3] and P [i-4], the generation of misrecognition of transient in a long period steady waveform having a case partially reducing the segment power to an extremely small value some times can be prevented. An windowing/orthogonal transformation part 3 orthogonally transforms an audio signal in each sample by a DCT, FET or the like in accordance with reference frame length T or shortened frame length (T/4) based upon a detection flag (trans) detected from the detecting part 2 to divide the signal into plural sub-bands.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a high-efficiency speech coding apparatus for coding an audio signal for each finite length frame.

[0002]

2. Description of the Related Art High-efficiency audio coding in a mini disc (MD), a digital compact cassette (DCC), a karaoke CD, etc. is called music compression because it compresses the data amount of an audio signal. In such an encoding method, an audio signal is divided into a plurality of subbands by a digital filter or orthogonal transformation, and the number of quantization bits for each subband is determined based on psychoacoustic analysis in the frequency domain. In the following description, the term “encode” may be used to mean compression in addition to encoding.

FIGS. 6A to 6D show an example in which a frequency band is divided by orthogonal transformation. FIG. 6A shows that the 16-bit PCM audio signal to be encoded is cut out by 512 samples, and here, the description will be made assuming that the total information amount enclosed by the rectangle in the figure is 16 bits * 512 = 8192 bits. Of course, the number of samples to be cut out and the number of PCM bits are not limited to this value.

FIG. 6 (b) shows the DC signal shown in FIG. 6 (a).
T (discrete cosine transform) and FFT (fast Fourier transform)
Shows a signal frequency-converted by orthogonal transformation such as, and the curve in the figure shows the envelope of the frequency spectrum. Here, assuming that the amount of information is preserved by orthogonal transformation, this total amount of information can also be expressed by the rectangular area in the figure. On the other hand, according to the psychoacoustic model, when a signal shown in FIG. 6B is present, a signal level masked by the signal and inaudible can be defined as a curve, which is generally called a masking effect. .

When a masking curve is drawn from FIG. 6B, it can be expressed as shown in FIG. 6C, where FIG.
Considering requantization of the signal shown in (b), if the quantization noise level generated by the requantization is equal to or lower than the level defined by the masking curve, the noise is inaudible to the human ear. be able to. Therefore,
As shown in FIG. 6 (d), the spectrum is divided into sub-bands for each plurality of data, the maximum signal level for each sub-band is S, and the noise level allowed from FIG. 6 (c) is N. If requantization is performed with the number of bits satisfying S / N, the quantization noise at that time is masked and cannot be heard.

The rectangle in FIG. 6 (d) shows the amount of information required at the time of compression and decompression. In particular, the deformed rectangle in the center of the figure shows the main information, and the elongated rectangle at the bottom of the figure shows the auxiliary information. The auxiliary information is information indicating the maximum value (scale value) and the number of quantization bits of each subband necessary for decoding. Therefore, the total amount of information shown in FIG. 6D is the sum of the amount of main information and the amount of auxiliary information.
It can be seen that it is a fraction of the total amount of information in (b). Therefore, as shown in FIG. 7, the above-described processing (steps S1 to S6) is performed in a section (512 sample sections in this example).
By repeating every time, it is possible to encode with almost no deterioration in sound quality.

FIGS. 8 (a) and 8 (b) respectively show a general voice efficient coding and decoding apparatus. Figure 8
In the encoding device shown in (a), for example, 16-bit PCM
After the audio signal is held by the frame buffering unit 1, 512 samples are cut out by the windowing / orthogonal transform unit 3, and the audio signal of each sample is DCT.
Orthogonal transform is performed by FFT or the like, and divided into a plurality of subbands.

Then, the psychoacoustic analysis unit 4 determines the number of quantization bits of each subband, and the quantization / encoding unit 5
Is quantized and encoded by this quantization bit number, the audio signal of each sub-band divided by the orthogonal transformation unit 2, and the data which is quantized and encoded by the quantization / encoding unit 4 and compressed. The quantized bit number determined by the psychoacoustic analysis unit 3 is multiplexed by the multiplex unit 6 and output.

In the decoding device shown in FIG. 8 (b), the demultiplexing unit 7 separates the speech code and the number of quantization bits, and the decoding / dequantization unit 8 decodes the speech code to quantize the speech code. Inverse quantization with the number of bits
A 16-bit PCM audio signal is reproduced by the window unit 9 and the frame buffering 10.

Next, the relationship between the nature of the actual audio signal and the result of encoding and decoding the signal will be described. 9A and 9B show signals in the case of being almost stationary in the frame section. In particular, FIG. 9A shows the original waveform, and FIG. 9B shows the waveform after encoding and decoding (hereinafter simply referred to as "process waveform"). It can be said that there is almost no difference in hearing when both signals are processed according to the psychology of hearing as described above.

On the other hand, FIGS. 10A and 10B respectively show an original waveform and a processed waveform of a non-stationary signal whose amplitude (power) rises sharply in a frame section.
As is clear from (b), the noise component N that greatly exceeds the original waveform appears before the power rises. Such noise N is generally called a pre-echo, and noise that goes back about 1 to 3 msec or more from the rising edge is detected as a noise echo incidental to the rising edge of the signal.

The cause of this noise is that the signal power change section is shorter than the frame analysis section. Further, the quantization noise generated by the frame encoding process has a frequency-amplitude characteristic as shown in FIG. 6C, but the frequency-phase characteristic is random, and therefore, the time domain generated on the processed waveform is generated. Since the quantization noise in (1) is uniformly (steadily) distributed in the frame, a large noise N affected by the large amplitude area is originally generated in the area where the signal amplitude is small as shown in FIG. Will appear. Note that noise appears even after the sharp fall of the signal power for the same reason, but in this case, the afterglow of the stimulus generated by the preceding large signal in the auditory mechanism is relatively long (10 to 20 msec). It is persistent and difficult to detect.

As a countermeasure against such a pre-echo, it is most effective to reduce the frame length, that is, to improve the time resolution of processing. For example, when dividing into subbands using orthogonal transformation, the frame length at the time of transformation is shortened from the standard length T to ¼ or the like as shown in FIG.
The time resolution can be quadrupled by performing the conversion once. As a result, the noise due to the unsteady waveform is confined in a shorter section, and the pre-echo is attenuated.

FIG. 12 shows the original waveform shown in FIG.
A waveform processed with a frame length of 4 is shown, and it can be seen from the processed waveform shown in FIG. 10B that the pre-echo is attenuated. When orthogonal transformation is not used for subband division, the same effect can be obtained by quantizing the time domain samples (time waveforms) of each subband for each shorter section.

The point of countermeasures against such pre-echo is how to accurately detect the non-stationarity of the waveform (= steep rise of power). This detection will be referred to as "transient detection" below. To call. here,
Generally, when the frame length is shortened, the amount of auxiliary information in units of frame described in FIG. 6D becomes relatively larger than the amount of main information, so that the amount of information assigned to main information decreases. In the case of using the orthogonal transform, if the frame length is shortened, the frequency resolution deteriorates, and the accuracy of applying psychoacoustic sound deteriorates. Therefore, the longer the frame length is, the better for a steady waveform, and thus the sound quality is generally deteriorated when the transient detection malfunctions.

A conventional transient detection method will be described with reference to FIGS. 13 and 14. Generally, a frame length of about 10 to 20 msec is selected,
For transient detection, approximately 1 to
It is divided into, for example, m segments each having a short length of 3 msec, and the total power P [i] of each segment i is calculated as follows (steps S11 and S12).

[0017]

[Equation 1]

As shown in FIG. 14, the determination of transient is made by determining that a segment i and a segment (i-
The power ratio with 1) is compared with a preset judgment reference value, and for example, for i = 0 ... m-1

[0019]

## EQU00002 ## P [i] / P [i-1]> At (Condition 1) where At is a criterion value P [-1] is P [m-1] of the previous frame

When is satisfied even once, it is determined that there is a transient in the frame, and the detection flag trans is set (steps S13 to S17). Here, the judgment reference value At is generally 15 as shown in FIG.
Approximately 20 dB is selected. In high efficiency coding, average S / N ratio after encoding and decoding is 20 to 30 dB
Since it is often a degree, it is appropriate to set the judgment reference value to a level at which the amplitude of the pre-echo as shown in FIG. 10A cannot be ignored compared with the amplitude of the original waveform in that region. FIG. 10B shows the case where the amplitude of the pre-echo can be ignored. The appearance of the power ratio of the segment in the steady waveform is also referred to.

[0021]

However, the above-mentioned conventional transient detection method has the following two problems. (1) For example, when the transient is located exactly in the center of the segment as shown in FIG. 16, detection omission occurs. The reason for this is that since the segment power changes to a maximum value in just two segments, the amount of change per segment is halved, and comparing the power with adjacent segments, the power ratio P [i] / P [ This is because i-1] may not exceed the determination reference value At, and in this case, the actual waveform and the detection flag trans are different as shown in FIGS.

(2) In a steady waveform having a long period as shown in FIG. 18, for example, the segment power may become extremely small in some cases. In this case, the power ratio P [i] / P
Since [i-1] exceeds the judgment reference value At, FIG.
It may be mistaken for a transient as shown in (a) and (b). Therefore, the conventional transient detection method has a problem that detection omission or post-detection may occur depending on the nature of the waveform.

In view of the above conventional problems, it is an object of the present invention to provide a high-efficiency speech coding apparatus capable of suppressing transient detection omission and false detection and suppressing pre-echo.

[0024]

SUMMARY OF THE INVENTION In order to achieve the above object, the present invention seeks the power ratios of a segment, an adjacent segment and a segment separated by one or more when a transient is detected. That is, according to the present invention, in a high-efficiency speech coding apparatus that encodes an audio signal by processing it for each frame of a finite length, the audio signal is divided into segments having a segment length sufficiently shorter than the standard frame length. Calculate the total power of the
A high-efficiency speech coding apparatus comprising: a control unit that controls an audio signal to be processed with a frame length shorter than a standard frame length when the power ratios of the segments separated from each other are equal to or more than a predetermined value. Provided.

[0025]

In the present invention, a transient is detected when the power ratios of the segment, the adjacent segment, and the segments separated by one or more are equal to or more than a predetermined value. Therefore, when the transient is located at the center of the segment, the power ratio with the segment separated by one or more becomes equal to or higher than the reference determination value, and therefore, detection omission can be prevented. Also. In the case where the segment power is extremely small partially in the long-cycle steady waveform, the power ratio with the segments separated by 1 or more becomes equal to or less than the reference determination value, and therefore, it is possible to prevent misidentification as a transient.

[0026]

Embodiments of the present invention will be described below with reference to the drawings. 1 is a block diagram showing an embodiment of a high-efficiency speech coding apparatus according to the present invention, FIG. 2 is an explanatory diagram showing comparison target segments of the transient detection unit of FIG. 1, and FIG. 3 is a diagram of the transient detection unit of FIG. 4 is a flowchart for explaining the transient detection process, FIG. 4 is an explanatory diagram showing a non-stationary waveform and its transient detection flag, and FIG. 5 is an explanatory diagram showing a steady waveform and its transient detection flag.

In FIG. 1, for example, when a 16-bit PCM audio signal is held by the frame buffering unit 1, and in the present embodiment, the transient detecting unit 2 compares the power P [i] of the segments when detecting the transient. In addition to the operation between adjacent segments, the operation is also performed between segments separated by one or more. For example, P [i] and P [i-1] and P [i] and P [i-
By comparing [2], it is possible to prevent detection omission when the transient is located exactly in the center of the segment as shown in FIG.

[0028]

(3) (P [i] / P [i-1]> At) or (P [i] / P [i-2]> At)
(Condition 2)

Further, in this embodiment, for example, as shown by the broken line in FIG. 2, P [i] and P [i-3] and P [i] and P
By comparing [i-4], for example, as shown in FIG. 18, it is possible to prevent the false recognition of a transient in a long-waveform stationary waveform in which the segment power may become extremely small partially.

[0030]

## EQU4 ## (P [i] / P [i-3]> At2) and (P [i] / P [i-4]> At2)
(Condition 3) However, At2 is a judgment reference value that is the same as or different from At.

The windowing / orthogonal transformation unit 3 receives the PCM audio signal from the frame buffering unit 1 by, for example, 51
Cut out 2 samples, then transient detector 2
Based on the detection flag trans from the sampled audio signal of each sample with a standard frame length T as shown in FIG. 11A or a shortened frame length (T / 4) as shown in FIG. 11A. Then, it is orthogonally transformed and divided into a plurality of subbands.

Further, the psychoacoustic analysis unit 4 determines the number of quantization bits of each subband, and the quantization / encoding unit 5 uses the number of quantization bits of each subband of the subbands divided by the orthogonal transform unit 2. Quantize and encode the audio signal. The data quantized and encoded by the quantization / encoding unit 5 and compressed, the flag trans detected by the transient detection unit 2, and the psychoacoustic analysis unit 3
The number of quantized bits determined by is multiplexed by the multiplex unit 6 and output. Further, in a decoder not shown, these data are separated and decoded.

Transient detection processing of the transient detector 3 will be described with reference to FIG. First, when a frame length of about 10 to 20 msec is selected, the sample of the frame is divided into, for example, m segments having a short length of about 1 to 3 msec, and then the transient detection flag trans and the segment index i are reset and Total power P of 4 segments
[-1] to P [-4] are loaded (step S21),
Next, the total power P [i] of each segment i is calculated based on the above-mentioned formula (Equation 1) (step S2).
2).

Then, it is judged whether or not the condition (2) is satisfied (step S23).
After 24 and S25, the same processing is repeated from step S22. Then, in step S23, the condition (2)
When the conditions are satisfied, it is determined whether or not the condition (3) is satisfied (step S26), and when the conditions are not satisfied, the step 2 is performed.
After S4 and S25, the same processing is repeated from step S22. When the processing is satisfied, the detection flag trans is set (step S27), and the process proceeds to step S28. If step S27 is not executed for all i, tr
Ans is not set and the process proceeds to step S28.

Therefore, according to the above embodiment, for example, P [i] and P [i-1] and P [i] and P [i-2] are compared as shown by the solid line in FIG. As shown in FIG. 4, detection omission in the case of an unsteady waveform can be prevented, and, for example, as shown by the broken line in FIG. 2, P [i] and P [i-3] and P [i] Since P [i-4] is compared, it is possible to prevent erroneous recognition of transients in a long-period stationary waveform as shown in FIG. If the orthogonal transformation is not used for subband division, the same effect as the use of the shortened frame length can be obtained by quantizing the time domain samples (time waveforms) of each subband for each shorter section. You can

[0036]

As described above, according to the present invention, when the power ratios of the segment, the adjacent segment, and the segments separated by one or more are detected as a predetermined value or more, the transient is detected in the center of the segment. In the case of being located, the power ratio with the segment separated by 1 or more becomes equal to or higher than the reference judgment value, and therefore detection omission can be prevented. Further, in the case where the segment power becomes extremely small partially in the long-cycle steady waveform, the power ratio with the segments separated by 1 or more becomes equal to or less than the reference judgment value, and therefore it is possible to prevent misidentification as a transient. .

[Brief description of drawings]

FIG. 1 is a block diagram showing an embodiment of a high-efficiency speech coding apparatus according to the present invention.

FIG. 2 is an explanatory diagram showing a comparison target segment of the transient detection unit of FIG.

FIG. 3 is a flowchart for explaining a transient detection process of a transient detection unit in FIG.

FIG. 4 is an explanatory diagram showing an unsteady waveform and its transient detection flag.

FIG. 5 is an explanatory diagram showing a steady waveform and its transient detection flag.

FIG. 6 is an explanatory diagram schematically showing a high-efficiency voice encoding method.

FIG. 7 is a flowchart for explaining the high-efficiency speech coding processing of FIG.

FIG. 8 is a block diagram showing a general voice efficient encoding and decoding device.

FIG. 9 is an explanatory diagram showing an original waveform and a waveform after encoding and decoding when the frame section is almost normal.

FIG. 10 is an explanatory diagram showing an original waveform of an unsteady signal whose amplitude (power) sharply rises within a frame section and its encoded and decoded waveforms.

FIG. 11 is an explanatory diagram showing a standard frame length and a shortened frame length.

12 is an explanatory diagram showing a waveform obtained by processing the original waveform shown in FIG. 10A with a frame length of ¼.

FIG. 13 is a flowchart illustrating a conventional transient detection process.

FIG. 14 is an explanatory diagram showing a conventional comparison target segment.

FIG. 15 is an explanatory diagram showing a determination reference value for conventional transient detection.

FIG. 16 is an explanatory diagram showing a case where a transient is located at the center of a segment.

FIG. 17 is an explanatory diagram showing an original waveform and a transient detection flag in the case shown in FIG.

FIG. 18 is an explanatory diagram showing a long-cycle stationary waveform and its segment power.

FIG. 19 is an explanatory diagram showing an original waveform and a transient detection flag in the case shown in FIG. 18.

[Explanation of symbols]

 1 Frame Buffering Section 2 Transient Detection Section (Control Means) 3 Windowing / Orthogonal Transformation Section 4 Auditory Psychological Analysis Section 5 Quantization / Coding Section 6 Multiplex Section

Claims (1)

[Claims] 1. A high-efficiency speech coding apparatus for coding an audio signal by processing each frame of a finite length, dividing the audio signal into segments having a segment length sufficiently shorter than a standard frame length, and totaling each segment. And a control means for calculating the power and controlling so that the audio signal is processed with a frame length shorter than the standard frame length when the power ratio of the segment to the adjacent segment and the segment separated by one or more is a predetermined value or more. A high-efficiency speech coding apparatus characterized by the above.