JP2001127641A - Audio encoder, audio encoding method and audio encoding signal recording medium - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an audio encoding method that uses window functions to conduct overlap encoding. SOLUTION: In the audio encoding method that uses at least 4 kinds of window functions of a long window, a short window, a start window, an a stop window to conduct overlapping encoding, the long window or the start window of the window function at the start of encoding is used by taking the matching performance of codes into account, the long window or the stop window is used for the window function at the end of encoding to decide the window function, the window function is multiplied, its output is converted from a time axis into a frequency axis, data are encoded so that a virtual buffer value of a frame at the start of encoding and a virtual buffer value of a frame at the end of encoding are the same specified value so as to enhance the matching performance of the codes.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

[0001] 1. Field of the Invention [0002] The present invention relates to an audio (speech) encoding method and apparatus for performing overlap encoding using a plurality of window functions.

[0002]

2. Description of the Related Art As an audio encoding method, a conventional VC is used.
MP, which is an encoding method used in D (video CD), etc.
In the case of the EG1-LayerII system, the linear PCM1152
Encoding is performed in sample units (frames), and 1152
The code amount generated for each sample is constant. In such a case, even if the encoded streams obtained by separately encoding the signals are connected, if the encoding rate is constant as shown in “Stream connection in MPEG1 Layer II” in FIG. Since the code amount (Nbits) generated in units is constant (fixed), overflow and underflow did not occur for the virtual buffer indicating the buffer amount at the time of decoding.

However, M standardized in April 1997
In the case of the PEG2-AAC (Advanced Audio Coding) coding method, coding is performed in units of PCM 1024 (frames), and even if the coding rate is constant, “MPEG” in FIG.
As shown in “2-AAC Stream Connection”, the number of bits per frame is variable and the amount of code generated for each frame is different.

An example of an audio encoding / decoding apparatus of the MPEG2-AAC encoding system will be described with reference to the drawings. FIG. 14 is a block diagram showing an example of the audio encoding / decoding device. This audio encoding / decoding device includes a window function processor 11, a time-frequency converter 12, a quantizer 13,
Auditory model 14, inverse quantizer 15, frequency-time converter 1
6 and a window function processor 17.

First, the input PCM signal is multiplied by a window function in a window function processor 11.
This is necessary for the next time-frequency converter 12 to accurately determine the frequency component. This window function is shown in FIG.
As shown in FIG. 6, a frame is created by providing an overlap area, and a process of multiplying a window function for each frame is performed.

In the time-frequency converter 12, an FFT or MD
The transformation from the time axis to the frequency axis is performed by using a CT (Modified Discrete Cosine Transform) or the like. The input PCM signal is supplied to the auditory model 14, a bit allocation amount for quantization is calculated by calculating a masking level based on psychoacoustic psychology, and the bit allocation amount is supplied to the quantizer 13. The quantizer 13 performs quantization on the basis of the bit allocation amount, and forms and outputs a bit stream.

On the decoding side, an inverse quantizer 15 decomposes the transmitted bit stream and performs inverse quantization. The frequency-time converter 16 has an inverse quantizer 15.
Is supplied and the conversion from the frequency axis to the time axis is performed.

The output of the frequency-to-time converter 16 is supplied to the window function processor 17, and the effect of the window function is removed by further multiplying the frame by the window function used at the time of encoding and adding the preceding and succeeding frames. And the PCM signal is decoded.

In the case of the MPEG2-AAC system,
As shown in “MPEG2 AAC Window Function and Overlap Processing” in FIG. 4, encoding is performed with 1024 samples being overlapped. In this case, 204
After multiplying the window function every 8 samples, the MDCT (ModifiedD
Conversion to the frequency axis is performed by iscrete cosine transform, and encoding is performed.

In order to perform adaptive block length switching in accordance with the block length in which the MDCT operation is performed, a long window, a long window, as shown in FIG. Short window, start
There are four types of functions: windows and stop windows.

FIG. 16 is a diagram showing a sine window and a frame structure with 50% overlap, which are often used when MDCT is used in the time frequency converter 12 of FIG. An audio signal (PCM data) is divided at a certain period, multiplied by a sine window, MDCT is performed, and signal processing is performed (frame 1). The next frame (frame 2) is composed of an audio signal (overlap area) in the latter half of the frame 1 and a new audio signal, and performs the same processing.

However, in the case of an audio signal containing an attack sound such as a castanet, quantization noise is easily detected in a frame at a fixed interval, so that a frame (short frame) having a shorter interval than a normal frame is used.
, The quantization noise is hardly perceived.
In this case, the processing is performed using a window function corresponding to a short frame (short window function) and a window function for shifting to a short frame (start window function).

FIG. 6 shows four types of window shapes, long, short, start, and stop. start,
The stop window is used for connecting a long window and a short window, and has an asymmetric shape. FIG. 17 shows a state transition diagram of the window function. Normally, the long window is used repeatedly, but when a transient signal such as an attack sound is detected, the next frame transitions to the start window, and the next frame transitions to the short window. .

As it returns to a steady signal, it returns from the short window to the stop window and back to the long window. Further, there is a case where a transition is made from the start window to the stop window without passing through the short window. However, when the coded stream is edited, frames are not always connected in the form following the state transition of FIG. 17, for example, following a long window, such as a long window or a start window.

[0015]

Here, at the start of encoding, as shown in "encoding start frame" in FIG.
Generally, silence data is used for the overlap portion. For this reason, when the bit stream is simply connected at the time of editing using the silence data in the overlap portion, the reproduction sound of the connection portion is as shown in “Example of decoding result of connection portion” in FIG. There is a problem that a normal reproduction sound cannot be obtained. In particular, when a short window or a start window is used at the end of encoding, as shown in each addition result column in FIG. 15, when a short window is used (FIG. 15A), the start window is used. (Fig. 15
(B)) also has a problem that the output sound becomes unpleasant because the level of the reproduced sound rapidly drops in a short time to reach the silence data level.

When the material A and the material B are encoded by the MPEG2-AAC encoding method, when the material A is encoded, A1, A2, A3, and A4 bits are set for each frame.
Is generated, and in the case of the material B, B1,
B2, B3 and B4 bits are generated. For these, code amount control that can be used for each of the materials A and B is performed, and coding is performed so that the virtual buffer is stabilized.
As shown in “stream joining in PEG2AAC”, simply connecting the first half material A and the second half material B, the connected bit streams are A1, A2, B3
Since the code is B4 bits, a problem occurs that the virtual buffer overflows or underflows.

[0017]

In order to solve this problem, the present invention solves this problem by using at least four types of window functions of a long window, a short window, a start window, and a stop window. In the audio coding method of performing overlap coding, the window function at the start of coding is either the long window or the start window in consideration of the joinability of codes, and the window function at the end of coding is the long window function. As a window or a stop window, a window function is determined, and after multiplying by the window function, the output is converted from the time axis to the frequency axis, and the virtual buffer value and the coding of the frame at the start of the coding. The audio encoding method according to claim 2, characterized in that the encoding is performed so that the virtual buffer value of the frame at the end becomes the same prescribed value to improve the joinability of the codes. In an audio coding method for performing overlap coding using at least four types of window functions of a long window, a short window, a start window, and a stop window, a window at the start of coding in consideration of code jointness. The function determines a window function as either the long window or the start window, multiplies the window function, converts the output from the time axis to the frequency axis, and provides a virtual buffer of the frame at the start of encoding. A code string coded so that the value becomes a specified value is defined as a first code string, and a window function at the end of coding determines a window function as one of the long window and the stop window. After multiplying by the function, the output is converted from the time axis to the frequency axis, and a code string coded so that the virtual buffer value of the frame at the end of the coding becomes the specified value is converted into a second code string. age 4. An audio encoding method according to claim 3, wherein a start part of said first code string and an end part of said second code string are connected to improve the joinability. The invention of
In an audio coding apparatus that performs overlap coding using at least four types of window functions such as a long window, a short window, a start window, and a stop window, the window function at the start of coding considers code jointness. Window function processing means 1 and 8 for determining a window function as either the long window or the start window, and the window function at the end of encoding as the long window or the stop window; A time-frequency converter 2 for converting the output of the processing unit from the time axis to the frequency axis, and a virtual buffer value of the frame at the start of encoding and a virtual buffer value of the frame at the end of encoding Are provided with quantizing means 3, 4 and 9 for performing encoding so that the values have the same specified value.
According to a fourth aspect of the present invention, there is provided an audio encoded signal recording medium on which an audio signal subjected to overlap encoding using at least four types of window functions of a long window, a short window, a start window, and a stop window is recorded. The window function at the start of encoding is either the long window or the start window in consideration of the associativity of the code, and the window function at the end of encoding is either the long window or the stop window. Audio encoding characterized in that an encoded audio signal is recorded such that the virtual buffer value of the frame at the start of encoding and the virtual buffer value of the frame at the end of encoding have the same prescribed value. A signal recording medium is provided.

[0018]

BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a block diagram showing an embodiment of an audio encoding / decoding apparatus according to the present invention. FIG. 1 shows a block diagram of an embodiment of an encoding / decoding apparatus when used in the MPEG2-AAC system to which the present invention is applied.

As shown in the block diagram of FIG. 1, an audio encoding / decoding apparatus according to an embodiment of the present invention comprises a window function processor 1, a time-frequency converter 2, a quantizer 3, an auditory model 4, It comprises an inverse quantizer 5, a frequency-time converter 6, a window function processor 7, a window function designator 8, a bit allocation designator 9, and an encoding start / end time detector 10.

First, an embodiment of the audio encoding apparatus according to the present invention will be described. The input PCM data is supplied to a window function processor 1 where it is multiplied by a window function.

The MPEG2-AAC method uses a short window, as shown in "Example of shape of window function" in FIG.
Two types of window functions of a long window and two types of window functions of a start window and a stop window used for switching the window function between the short window and the long window are defined. In the vessel 1, they are multiplied by overlapping. The start window and the stop window serve to connect the long window and the short window, and as shown in FIG. 6, their shapes are asymmetric.

Normally, a long window is repeatedly used. However, when a transition signal such as an attack sound is detected, the next frame is switched to a start window, and the next frame is switched to a short window. I will do it. As it returns to a steady signal, from the short window to the stop window,
Then return to the long window.

FIG. 4 is a view for explaining the "window function and overlap processing" of the long window and the short window in the MPEG2-AAC system. The multiplication of the window function is performed by dividing the audio signal (PCM signal) every 1024 samples (frame), and performing 2048 samples of the encoding target frame (PCM signal) and the PCM signal of the immediately preceding frame. .

In ordinary coding, the window shape actually used for coding one frame before and the input P after one frame
The shape of the window determined by the CM signal determines the shape of the window used for the current multiplication.

At the start of encoding according to the present invention, an encoding start time is detected by an encoding start and end time detector 10 and the detected signal is supplied to a window function designator 8 where it is determined. The window function information is supplied to the window function processor 1 and the window function used in the frame one frame before is set as a long window. Further, encoding is performed using a window function determined by a window shape determined from input data one frame later. That is, if the input data of the frame after one frame is suitable for the long window, the long window is used,
If it is suitable for a short window, encoding is performed using each start window.

As shown in FIG. 8 "when the previous frame is a long window and the next is a short window", for example, the PCM signal of the processed frame after one frame is encoded by the long window one frame before. Is suitable for encoding using a short window, the start window is selected as the window function of the joint.

Also, as shown in FIG. 9 "when the previous frame is a long window and the next is a long window", one frame before is coded by the long window, and the frame after one frame is coded by the long window. If the coding used is suitable, a long window is selected as the window function of the joint.

As shown in "encoding start frame" in FIG. 5, silence data is used for an overlap portion.

Although the window function is usually determined by the window function used for encoding one frame before and the window function determined from the input data one frame after, In the case of the last frame, the window function designator 8 determines the window function using only the window function used for encoding one frame before.

That is, according to the present invention, when the window before the last frame is a long window, as shown in FIG. 8, the last frame is encoded using a long window. Also, the present invention
In the case of the last frame, the window before the last frame is a short window, as shown in “Case before the last frame is short window” in FIG. Encode the last frame.

Further, according to the present invention, in the case of the last frame, if the window one frame before is a start window as shown in FIG. Thus, the last frame is encoded using the stop window.

Further, according to the present invention, in the case of the last frame, if the window one frame before is a stop window as shown in FIG. Thus, the encoding of the last frame is performed using the long window.

That is, the present invention provides a stop window or
By encoding the last frame using a long window, the result of simply connecting the bit stream is decoded and played back, compared to using a short window or a start window for the last frame. In addition, the change in the level of the reproduced sound becomes gentle, and the deterioration of the sound quality can be hardly perceived. That is, as shown in each addition result column of FIG. 7, the level of the reproduced sound is not limited both when the long window is used (FIG. 7A) and when the stop window is used (FIG. 7B). The change becomes gradual, and the deterioration of the sound quality can be hardly perceived.

FIG. 7 also shows a state in which the code processed by the present invention is edited. FIG. 7A shows a case where a code sequence ending with a long window and a code sequence starting with a long window are connected. FIG. 7B shows a case where a code string terminated by a stop window and a code string started by a long window are connected. This shows that even if the start part and the end part of the separately encoded bit stream are simply connected (edited), deterioration of the sound quality at the connection part can be minimized.

Although not shown here other than this,
The same effect can be obtained when the code sequence ended with the long window and the code sequence started with the start window are connected, and when the code sequence ended with the stop window and the code sequence started with the start window are connected. Things.

As described above, the window function processor 1
Multiplies the window function for each 2048 samples by the time-frequency converter 2 and then multiplies the MDCT (Modified Discrete Cosine Transf
orm) to convert to the frequency axis.

In the quantizer 3 to which the PCM signal converted to the frequency axis from the time-frequency converter 2 is supplied, the quantizer 3 is assigned via the bit assignment designator 9 by the auditory model 4 to which the input PCM signal is supplied. The quantization is performed based on the bit amount of the bit allocation.

Normally, the bit assignment designator 9 directly assigns the bit from the auditory model 4 to the quantizer 3 and quantizes the coding start frame and the end frame. , The detection signal from the encoding start / end time detector 10 is supplied to the bit allocation designator 9, and the bit allocation is performed on the quantizer 3 so that the virtual buffer value becomes the same specified value. To output a bit stream as an encoder output.

Next, an embodiment of the audio decoding apparatus according to the present invention will be described below. As shown in FIG. 1, the decoding apparatus comprises an inverse quantizer 5 to which a bit stream is supplied, a frequency-time converter 6, and a window function processor 7 for outputting PCM.

The bit stream from the encoder side is first supplied to an inverse quantizer 5, where an inverse quantization process is performed.

The output of the inverse quantizer 5 is a frequency-time converter 6
And is converted from a frequency axis to a time axis signal. The conversion output of the frequency-time converter 6 is supplied to the window function processor 7
, And is multiplied by the window function in the window function processor 7 to output PCM data as a decoder output.

The window function processor 7 includes a frequency-time converter 6
Is supplied, and the effect of the window function can be removed by multiplying the frame by the window function used at the time of encoding and adding the preceding and succeeding frames, and the PCM signal is decoded.

As described above, by controlling the code amount so that the virtual buffer value at the time of encoding start and the virtual buffer value at the time of encoding end become the same specified value, it is possible to simply combine separately encoded bit streams. Thus, decoding can be performed normally without causing overflow or underflow of the virtual buffer. Through the above processing, it is possible to combine bit streams that have been separately encoded, and it is possible to minimize deterioration in sound quality at the connection part.

[0044]

According to the present invention, an audio coding method for performing overlap coding using at least four types of window functions of a long window, a short window, a start window, and a stop window, in consideration of code jointness. Then, the window function at the start of encoding is any of the long window or the start window, and the window function at the end of encoding is a window function determined as any of the long window or the stop window, Convert the output that has determined the window function from the time axis to the frequency axis,
The encoding is performed such that the virtual buffer value of the frame at the start of encoding and the virtual buffer value of the frame at the end of encoding have the same specified value, and the joinability of the codes can be improved. The integrity of the combined code is assured.

Further, according to the present invention, even if the bit streams separately encoded are simply combined, the deterioration of the sound quality at the connection portion can be suppressed to the minimum.

Further, according to the present invention, by controlling the code amount so that the virtual buffer values at the start of encoding and the end of encoding have the same specified value, the bit streams separately encoded can be simply combined. However, decoding can be performed normally without causing overflow or underflow.

[Brief description of the drawings]

FIG. 1 is a block diagram showing an embodiment of an audio encoding / decoding apparatus according to the present invention.

FIG. 2 is a diagram for describing stream joining in MPEG1.

FIG. 3 is a diagram for explaining stream joining in MPEG2-AAC.

FIG. 4 is a diagram illustrating “window function and overlap processing” of a long window and a short window in the MPEG2-AAC method.

FIG. 5 is a diagram illustrating an encoding start frame.

FIG. 6 is a diagram for explaining four types of window functions used in MPEG2: a short window, a long window, a start window, and a stop window.

FIG. 7 is a diagram showing an example of a decoding result of a connection part in the case of the present invention.

FIG. 8 is a diagram for explaining a case in which a frame before the present invention is a long window and a next is a short window.

FIG. 9 is a diagram for explaining a case where a frame before the present invention is a long window and a next is a long window.

FIG. 10 is a diagram for explaining a case where a long window precedes a last frame according to the present invention.

FIG. 11 is a diagram illustrating a case where a short window is provided before the last frame according to the present invention.

FIG. 12 is a diagram for explaining a case where a start window is before a last frame according to the present invention.

FIG. 13 is a diagram for explaining a case where a stop window is provided before the last frame according to the present invention.

FIG. 14 is a block diagram showing an example of a conventional audio encoding / decoding device.

FIG. 15 is a diagram illustrating an example of a decoding result of a connection part in a conventional case.

FIG. 16 is a diagram showing a frame configuration using MDCT.

FIG. 17 is a state transition diagram showing a normal state transition.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Window function processor (window function processing means) 2 Time frequency converter (time frequency conversion means) 3 Quantizer (quantization means) 4 Auditory model 5 Inverse quantizer 6 Frequency-time converter 7 Window function processor 8 Window function designator 9 Bit allocation designator 10 Coding start / end time detector

Claims (4)

[Claims] 1. An audio encoding method for performing overlap encoding using at least four types of window functions: a long window, a short window, a start window, and a stop window. The window function at the start of the coding is either the long window or the start window, and the window function at the end of the coding is a window function determination as either the long window or the stop window. After multiplying
The output is converted from the time axis to the frequency axis, and the encoding is performed so that the virtual buffer value of the frame at the start of encoding and the virtual buffer value of the frame at the end of encoding have the same specified value. An audio encoding method characterized by improving the joinability of audio data. 2. An audio coding method for performing overlap coding by using at least four types of window functions of a long window, a short window, a start window, and a stop window. The window function at the start of coding determines the window function as either the long window or the start window, multiplies the window function, converts the output from the time axis to the frequency axis, and starts the coding at the start of the coding. A code sequence coded so that the virtual buffer value of the frame becomes a specified value is defined as a first code sequence, and a window function at the end of coding is determined as either the long window or the stop window. After multiplying the window function, the output is converted from the time axis to the frequency axis, and the encoded code sequence is encoded so that the virtual buffer value of the frame at the end of the encoding becomes the specified value. Second As a code sequence, and it connects the termination portion of said the start of the first code string second code string, audio encoding method is characterized in that so as to improve the bonding properties. 3. An audio encoding apparatus that performs overlap encoding using at least four types of window functions, a long window, a short window, a start window, and a stop window,
The window function at the start of encoding is either the long window or the start window in consideration of the joinability of codes, and the window function at the end of encoding is the window function of either the long window or the stop window. Window function processing means for determining, a time-frequency conversion means for converting the output of the window function processing means from a time axis to a frequency axis, and a virtual buffer value of the frame at the start of the encoding, And a quantizing means for performing encoding so that the virtual buffer value of the frame at the end of encoding becomes the same prescribed value. 4. An audio coded signal recording medium on which an audio signal subjected to overlap coding using at least four window functions of a long window, a short window, a start window, and a stop window is recorded. The window function at the start of coding is either the long window or the start window in consideration of the associativity of the code, and the window function at the end of coding is any of the long window or the stop window. An audio encoded signal recording characterized in that an encoded audio signal is recorded such that a virtual buffer value of a frame at the start and a virtual buffer value of a frame at the end of encoding have the same specified value. Medium.