CN100339886C - Encoder capable of detecting transient position of sound signal and encoding method - Google Patents

Abstract

An encoder includes a polyphase filter bank, a transient detector, and an encoding processing unit. The encoder first performs a sub-band encoding step to generate a plurality of sub-band samples according to an input signal, wherein each sub-band sample comprises a plurality of frequency sub-bands. Then, a selection step is performed to select a plurality of subband samples as reference sampling data, and the block length of the window data is determined according to the energy sum of the frequency subbands of the reference sampling data in a preset frequency range. Finally, transform coding step is carried out, and the multiple frequency sub-bands are transformed by a preset transformation algorithm according to the window data determined in the selection step to generate output signals.

Description

Encoder capable of detecting transient position of sound signal and encoding method

technical field

The invention provides an encoder, especially an encoder capable of detecting the transient position of a sound signal. The encoder of the present invention can further determine the block length of the window data used in the frequency domain encoding.

Background technique

At present, many encoders use special encoding algorithms based on the characteristics of the human auditory system, which can compress digital audio signal data to more than ten times, such as MP3, AAC, WWA and Dolby Digital, etc. These encoders use perceptual encoding, Frequency domain coding, window switching and dynamic bit allocation and other technologies are used to eliminate unnecessary content in the original sound signal data.

Perceptual coding is compression by eliminating sound signal data that is not perceived by the general human auditory system. Generally speaking, the sound frequency that humans can hear is about 20Hz to 20kHz, while the sound in other frequency domains is generally not felt by humans. On the other hand, the human auditory system also produces an auditory mask in some cases, and cannot distinguish quantized noise. The sound will be more difficult to detect, so you don't need to encode all the sound details into it.

Frequency domain coding is a method that can effectively eliminate unnecessary data. It converts time domain data with strong correlation into frequency domain where each element is almost irrelevant to remove unnecessary content in the data. Generally, it can be divided into Transform coding or subband coding. Transform coding has higher spectral resolution, while subband coding has lower resolution but higher efficiency, so the two codes can be combined into a hybrid filter with different resolutions at different frequencies. However, frequency domain coding has a remarkable phenomenon called preechoes. For example, if there is a sudden loud sound after general silence, the quantization error may increase. This phenomenon occurs in both transform coding and subband coding, causing a forward echo of the sound after the data has been converted back to the time domain.

One way to eliminate the forward echo is to limit the error to a smaller time period, separate the other parts of the sound from the forward echo, and make the forward echo occur in the shielded area. Limiting the error to a smaller time period requires the use of smaller blocks for frequency domain transformation. This method is called window switching. When the signal is stable, larger blocks are used for frequency domain coding, and when the signal has When there is a large transient (Transient), smaller blocks are used for frequency domain coding. The disadvantage of window switching is that more bits are required to represent the same data, because more information is required as the amount of encoded data increases.

Whether an encoder has good encoding quality has a lot to do with the allocation of bits between subbands or coefficients. In order to effectively allocate bits, the input signal must be continuously analyzed, and based on the model established by the knowledge of the human auditory system, more bits are allocated to the most effective area of human hearing, and no allocation is required in areas where the human ear is not sensitive Or only allocate few encoding bits. Because the signal is constantly changing, the human auditory system will respond differently to the signal under different conditions. This is the technology of dynamic bit allocation. A good bit allocation scheme requires an accurate psychoacoustic model.

Please refer to FIG. 1, which is a schematic diagram of known MPEG layer-3 audio signal encoding. First, the input signal 10 of pulse code modulation (pulse code modulation, PCM) is divided into 32 frequency subbands (frequency subbands) of equal width by polyphase filter bank (polyphase filter bank) 12, and polyphase filter bank 12 can be simple Analyze the relationship between frequency and time, but frequency subbands of equal width cannot accurately reflect the auditory characteristics of the human auditory system. In addition, adjacent frequency subbands will have more overlapping parts, so the polyphase filter bank 12 The output needs to be compensated using a modified discrete cosine transform (MDCT)14. The Modified Discrete Cosine Transform 14 further subdivides the frequency subbands to obtain better spectral resolution, and can eliminate some overlaps generated by the polyphase filter bank 12 . The Modified Discrete Cosine Transform 14 contains two window blocks of different lengths, one eighteen-sample long block and one six-sample short block, because consecutive transfer window blocks overlap by 50%, so the length of the block Thirty-six and twelve, respectively. When the sound signal is stable, long blocks have higher frequency resolution and better compression ratio, while short blocks provide better time resolution. Due to the low time resolution of the long block, if a transient phenomenon occurs in the processed block, the quantization noise (Quantization Noise) will spread to the entire block, so that the signal with less energy cannot be detected due to its low masking effect (Mask). Distortions such as forward echoes that mask quantization noise. To avoid forward echo, known MPEG audio signal coding uses a psychoacoustic model 16 to detect the transient (Transient) position of the audio signal, so as to use short blocks to perform Modified Discrete Cosine Transform 14 to avoid forward echo. After the input signal 10 is converted to the frequency domain using the technology of frequency domain encoding, then the quantization program 18 is performed to quantize the data according to the psychoacoustic model 16, and then the encapsulation process 20 is performed to output the data bitstream (bitstream) after encapsulating the data Output signal 22.

It can be seen from the above that, in the frequency domain encoding, in order to avoid the forward echo, window switching is a commonly used technique, and the mechanism for detecting the transient position of the sound signal is very important at this time. It is known that MPEG sound signal coding uses psychoacoustic model 16 to detect the transient position of the sound signal, although it is very accurate, but because the psychoacoustic model 16 is quite complicated, the required cost is also high. It is quite uneconomical to use a high-cost psychoacoustic model 16 for the transient position of the model.

Contents of the invention

It is therefore the main object of the present invention to provide an encoder which detects the transient position of an acoustic signal. On the other hand, the present invention also provides an encoder and an encoding method capable of determining the block length of window data used in frequency domain encoding, so as to solve the above problems.

The present invention provides an encoder for encoding an input signal into an output signal. The encoder includes a polyphase filter bank, which is used to generate a plurality of subband samples according to the input signal, and different subband samples correspond to input signal waveforms of different periods, and each subband sample includes a plurality of frequency subbands; A transient detector, connected to the polyphase filter bank, is used to determine the block length of the window data, the window data contains a plurality of weighted values, and the transient detector includes a subband selector, which is used to select the multiple weighted values sub-band samples as reference sampling data; an energy calculator, connected to the sub-band selector, used to calculate the energy sum of the frequency sub-bands in the reference sampling data; a partitioner, connected to the sub-band selector and the energy calculator between, used to divide the reference sampling data into an array of sub-sampling data, each group of sub-sampling data includes at least one sub-band sample; and a comparator, connected to the energy calculator, used to compare the output value of the energy calculator with The first critical value is compared, and a signal representing the block length of the window data is output according to the comparison result; and an encoding processing unit, connected to the polyphase filter bank and the transient detector, is used for the plurality of frequency subbands Multiple weighted values in the transient window data are multiplied to generate a weighted result, and then a preset conversion algorithm is used to generate the output signal according to the weighted result.

The present invention further provides an encoding method for encoding an input signal into an output signal. The encoding method includes a subband encoding step to generate a plurality of subband samples according to the input signal, different subband samples correspond to input signal waveforms of different periods, and each subband sample includes a plurality of frequency subbands; A selection step is performed to provide window data corresponding to a preset block length, the window data includes a plurality of weighted values; and the selection step includes: among the plurality of sub-band samples, selecting a plurality of sub-band samples as Referring to the sampled data, and determining the block length of the window data according to the energy sum of the frequency subbands of the reference sampled data within the preset frequency range; and performing a transform encoding step, multiplying the plurality of frequency subbands by the selection step A plurality of weighted values of the determined window data are used to generate a weighted result, and a preset conversion algorithm is used to generate the output signal according to the weighted result.

Compared with the known technology, the present invention provides an encoder and an encoding method that can be used to determine the block length of the window data used when performing Modified Discrete Cosine Transformation. The energy value of judging whether the sound signal data has a transient state requires far lower cost than the known psychoacoustic model, which is in line with economic benefits.

Description of drawings

Fig. 1 is the schematic diagram of known MPEG layer-3 sound signal encoding.

Fig. 2 is a schematic diagram of an encoder according to an embodiment of the present invention.

FIG. 3 is a schematic diagram of subband samples in this embodiment.

FIG. 4 is a flow chart of a method for an encoder to detect a transient position of a sound signal in an embodiment of the present invention.

Description of reference symbols

10 Input signal 12 Polyphase filter bank

14 Modified discrete cosine transform 16 Psychoacoustic model

18 Quantization program 20 Encapsulation program

22 Output signal 30 Encoder of the present invention

32 Transient detector 34 Encoding processing unit

36 Subband Selector 38 Energy Calculator

40 Partitioner 42 Comparator

50 reference sampling data

Detailed ways

Please refer to FIG. 2 , which is a schematic diagram of an encoder 30 according to an embodiment of the present invention. The encoder 30 is used to encode the pulse code modulated input signal 10 into an output signal 22 of a bit stream. The encoder 20 includes a polyphase filter bank 12 , a transient detector 32 and an encoding processing unit 34 . The polyphase filter bank 12 generates a plurality of subband samples according to the input signal 10 , and different subband samples correspond to waveforms of the input signal 10 at different periods, and each subband sample includes multiple frequency subbands. The encoding processing unit 34 may perform Modified Discrete Cosine Transformation on the multiple frequency subbands. The transient detector 32 is connected between the polyphase filter bank 12 and the encoding processing unit 34, and can determine the block length of the window data used by the encoding processing unit 34 to perform Modified Discrete Cosine Transformation. Transient detector 32 includes subband selector 36 , energy calculator 38 , partitioner 40 and comparator 42 . The sub-band selector 36 will select some of the sub-band samples in the plurality of sub-band samples in the preset frequency range as reference sample data, and then the energy calculator 38 will calculate the energy value contained in the reference sample data, and then the energy value Handed over to the comparator 42 for comparison with the threshold value. If the total energy of the reference sampling data exceeds the critical value, that is, there may be a transient situation in the reference sampling data, then the partitioner 40 divides the reference sampling data into sub-sampling data with equal widths of arrays, and each group The sub-sampling data contains at least one sub-band sample. At this time, the energy calculator 38 will calculate the energy difference of the frequency sub-bands of the adjacent two groups of sub-sampling data within the preset frequency range, and then send the energy difference to the comparator 42 is compared with a predetermined threshold. If the energy difference is greater than a predetermined critical value, the encoding processing unit 34 may decide to use the window data of the short block to perform Modified Discrete Cosine Transformation, and so on until the partitioner 42 completes all possible combinations of sub-sampled data. If the energy difference between the adjacent two groups of sub-sampled data is still smaller than the predetermined critical value, the encoding processing unit 34 may decide to use the window data of the long block to perform Modified Discrete Cosine Transform.

Please refer to FIG. 3 , which is a schematic diagram of subband samples in this embodiment. The polyphase filter bank 12 outputs eighteen subband samples in a time period t1, and each subband sample contains thirty-two frequency subbands. The encoding processing unit 34 performs Modified Discrete Cosine Transformation on each frequency subband in the overlapping period, that is, thirty-six subband samples. The transient detector 32 detects the position where the transient of the audio signal occurs to determine which window block the encoding processing unit 34 should use to perform Modified Discrete Cosine Transformation. The so-called preset frequency range usually refers to the frequency between the cut-off subband and the coding limit subband, and the subband selector 36 will select a frequency subband within this frequency range as the reference sample data 50 . The cut-off sub-band can be selected from the first sub-band or higher frequency sub-bands based on experience or experimental values. In this embodiment, the frequency of the cutoff sub-band is about 4kHz. The encoding restriction subband must be determined according to the encoding rules. Since bitrate and bandwidth have limitations, the encoder 30 must discard information of part of the high-frequency sub-bands, and the data of the discarded frequency sub-bands is no longer considered. The last subband is the coding-limited subband, assuming no information is discarded. After the reference sampling data 50 is selected, the energy calculator 38 will calculate the energy value contained in the reference sampling data 50, and then the comparator 42 will judge whether to continue to detect the reference sampling data 50, and the partitioner 40 can use the reference sampling data The data 50 is further divided into sub-sampling data of equal width, and then the energy calculator 38 calculates the energy difference between two adjacent groups of sub-sampling data, and the comparator 42 determines the block length of the window data. For example, first the energy calculator 38 calculates the total energy of all frequency subbands in the reference sample data 50 selected by the subband selector 36, if the total energy is greater than -60dB, there may be a transient state in the reference sample data occurs, the sub-band samples in the reference sampling data 50 are divided into six groups of equal-width sub-sampling data by the partitioner 40, and then the energy difference between the adjacent two groups of sub-sampling data is calculated by the energy calculator 38 and handed over to the comparator 42. In comparison, if the energy difference between the two sub-sampled data is not greater than 20dB, it means that there is no transient state between the two sub-sampled data, and the partitioner 40 will re-divide the sub-band samples in the reference sampled data 50 into Three sets of equal-width sub-sampling data, at this time, the energy calculator 38 calculates the energy difference between two adjacent groups of sub-sampling data and sends it to the comparator 42 to determine whether it is greater than 12dB. If it is greater than 12dB, it means that the data contains a transient situation, so it is judged that a short block window should be used; if it is not greater than 12dB, a long block window is used.

Please refer to FIG. 4 . FIG. 4 is a flowchart of a method for the encoder 30 to detect the transient position of the sound signal in an embodiment of the present invention. The encoding method of this embodiment can detect the transient position of the sound signal. The encoding method of this embodiment first performs the sub-band encoding step, and generates a plurality of sub-band samples according to the input signal 10. Different sub-band samples correspond to waveforms of the input signal 10 at different time periods, and each sub-band sample contains a plurality of frequency sub-bands. bring. Then carry out the selection step to determine the block length of the window data to be used in the next step. The window data contains a plurality of weighted values. The method of the selection step is to select a plurality of sub-band samples as a reference in the plurality of sub-band samples. Sampling data, and determining the block length of the window data according to the energy sum of the frequency sub-bands within the preset frequency range of the reference sampling data. Finally, a transform coding step is performed, multiplying the plurality of frequency subbands by the plurality of weighted values of the window data determined in the selection step to generate a weighted result, and using Modified Discrete Cosine Transform to generate an output signal according to the weighted result. The detailed steps of detecting the transient position of the sound signal are as follows:

Step 110: start to detect the transient position of the sound signal;

Step 120: Calculate whether the total energy of the frequency sub-band selected as the reference sampling data is greater than a predetermined critical value, if so, proceed to step 130, if not, proceed to step 170;

Step 130: Divide the reference sampling data into arrays of sub-sampling data of equal width, each group of sub-sampling data contains more than one sub-band sample, and calculate the frequency sub-bands in each group of sub-sampling data in the preset frequency range Energy value, then proceed to step 140;

Step 140: judging whether the energy difference between two adjacent groups of sub-sampling data is greater than a predetermined critical value, if so, proceed to step 160, if not, proceed to step 150;

Step 150: judging whether the reference sampling data can also be divided into different sub-sampling data, if so, then return to step 130, if not, then proceed to step 170;

Step 160: Refer to the transient position contained in the sampling data, output the window data signal using a short block, and proceed to step 180;

Step 170: Referring to the sampling data that does not contain the transient position, output the window data signal using a long block, and proceed to Step 180;

Step 180: output the judgment result, and end detecting the transient position of the sound signal.

Compared with the known technology, the present invention provides an encoder and an encoding method that can be used to determine the block length of the window data used when performing Modified Discrete Cosine Transformation. The energy value of judging whether the sound signal data has a transient state requires far lower cost than the known psychoacoustic model, which is in line with economic benefits.

The above descriptions are only preferred embodiments of the present invention, and all similar changes and improvements made according to the claims of the present invention shall fall within the scope of the patent of the present invention.

Claims (18)

1. A coding method, which is used to encode an input signal into an output signal, the method comprising: Performing a subband encoding step to generate a plurality of subband samples according to the input signal, different subband samples correspond to input signal waveforms of different periods, and each subband sample includes a plurality of frequency subbands; performing a selection step to provide window data corresponding to a preset block length, the window data including a plurality of weighted values; And the selection step includes: Among the plurality of subband samples, a plurality of frequency subbands are selected according to a preset frequency range as reference sample data, and the energy sum of the frequency subbands of the reference sample data within the preset frequency range is combined with a first A threshold value is compared to determine the block length of the window data; and performing a transform encoding step, multiplying the plurality of frequency subbands in the reference sampling data by the plurality of weighted values of the window data determined in the selection step to generate a weighted result, and using a preset conversion algorithm to generate according to the weighted result the output signal. 2. The coding method as claimed in claim 1, wherein when performing the selecting step, if the energy sum of the frequency subbands in the reference sampling data is greater than the first critical value, then additionally performing a comparing step, which includes: Dividing the reference sampled data into arrays of sub-sampled data, each set of sub-sampled data comprising at least one frequency subband; and Calculating the energy magnitude difference of the frequency subbands in two adjacent groups of sub-sampling data, if the difference is greater than a second critical value, then in the transform encoding step, using window data with a short block length; and If the energy magnitude difference of the frequency sub-bands in the adjacent two groups of sub-sampling data is less than or equal to the second critical value, then perform another comparison step, and make the frequency sub-bands contained in the sub-sampling data in this comparison step Differs from the subsampled data in the previous comparison. 3. The encoding method as claimed in claim 2, wherein if the energy sum of the frequency subbands in the reference sample data is less than the first critical value, then in the transform encoding step, a long block-length window data is used . 4. The encoding method as claimed in claim 1, wherein the input signal is a pulse code modulated signal. 5. The encoding method as claimed in claim 1, wherein the output signal is an encoded bit stream. 6. The encoding method as claimed in claim 1, wherein the preset transformation algorithm is Modified Discrete Cosine Transform. 7. An encoder for encoding an input signal into an output signal, comprising: A polyphase filter bank is used to generate a plurality of subband samples according to the input signal, and different subband samples correspond to input signal waveforms of different periods, and each subband sample includes multiple frequency subbands; A transient detector, connected to the polyphase filter bank, is used to determine the block length of the window data, and the window data contains a plurality of weighted values, and the transient detector includes: a subband selector, used for according to a The preset frequency range selects a plurality of frequency subbands from the plurality of subband samples as reference sampling data; an energy calculator, connected to the subband selector, is used to calculate the energy sum of the frequency subbands in the reference sampling data; partition A device, connected between the subband selector and the energy calculator, is used to divide the reference sampled data into groups of subsampled data, each group of subsampled data includes at least one frequency subband; and a comparator connected to the an energy calculator for comparing an output value of the energy calculator with a first critical value, and outputting a signal representing a block length of the window data according to the comparison result; and an encoding processing unit, connected to the polyphase filter bank and the transient detector, for multiplying a plurality of frequency subbands in the reference sampling data by a plurality of weighted values in the transient window data to generate a weighted result , and then use a preset conversion algorithm to generate the output signal according to the weighted result. 8. The encoder as claimed in claim 7, wherein the energy calculator calculates the energy magnitude difference of the frequency sub-bands in two adjacent sets of sub-sampling data, and then sends the result to the comparator to be used as the second critical value. Compare. 9. The encoder according to claim 8, wherein the partitioner further divides the reference sampling data into arrays of sub-sampling data according to the comparison result of the comparator, and the frequency sub-bands contained in each group of sub-sampling data A frequency subband different from that contained in the previous subsampled data. 10. The encoder as claimed in claim 7, wherein the input signal is a pulse code modulated signal. 11. The encoder of claim 7, wherein the output signal is an encoded bit stream. 12. The encoder as claimed in claim 7, wherein the predetermined transformation algorithm is Modified Discrete Cosine Transform. 13. A method of detecting the transient state of a sound signal during sound signal encoding, the method comprising: (a) generating a plurality of sub-band samples according to the sound signal, different sub-band samples correspond to sound signal waveforms of different periods, and each sub-band sample includes a plurality of frequency sub-bands; (b) Among the plurality of sub-band samples, select a plurality of frequency sub-bands as reference sampling data according to a preset frequency range, and calculate the energy sum of the frequency sub-bands within the preset frequency range according to the reference sampling data ; (c) If the energy sum of the frequency sub-bands in the reference sampling data is greater than the first critical value, divide the reference sampling data into an array of sub-sampling data, each group of sub-sampling data includes at least one frequency sub-band, and perform step ( d); On the contrary, if the energy sum of the frequency subbands in the reference sampling data is not greater than the first critical value, it is judged that there is no sound signal transient in the reference sampling data, and the processing ends; (d) Calculating the energy difference of the frequency sub-bands in the adjacent two sets of sub-sampling data, and judging the audio transient position in the sound signal according to the difference. 14. The method as claimed in claim 13, wherein when performing step (d) and judging the transient state of the sound signal according to the difference, if the difference is greater than a second critical value, then judging that the two groups of sub-sampled data The corresponding sound signal waveform is a transient waveform, and if the difference is less than the second critical value, the reference sampling data is divided into an array of sub-sampling data different from step (c), and step (d) is performed again ). 15. A transient detector arranged in a sound signal encoder, used to detect whether the sound signal input to the encoder contains a transient state, and the sound signal encoder includes a polyphase filter bank, which is used according to the input signal A plurality of sub-band samples are generated, and different sub-band samples correspond to input signal waveforms of different periods, and each sub-band sample includes a plurality of frequency sub-bands, the transient detector is connected to the polyphase filter bank, and includes : a subband selector, configured to select a plurality of frequency subbands from the plurality of subband samples according to a preset frequency range as reference sampling data; An energy calculator, connected to the subband selector, used to calculate the energy sum of the frequency subbands in the reference sample data; a partitioner, connected between the subband selector and the energy calculator, for dividing the reference sample data into an array of sub-sampling data, each group of sub-sampling data comprising at least one frequency sub-band; and A comparator, connected to the energy calculator, is used to compare the output value of the energy calculator with the first critical value, and judge whether the sound signal input to the encoder contains a transient state according to the comparison result. 16. The transient detector as claimed in claim 15, wherein the energy calculator calculates the energy magnitude difference of the frequency sub-bands in the adjacent two sets of sub-sampling data, and then transmits the result to the comparator and the second threshold value for comparison. 17. The transient detector as claimed in claim 16, wherein the partitioner further divides the reference sampling data into arrays of sub-sampling data according to the comparison result of the comparator, and the frequency contained in each group of sub-sampling data The subbands are different from the frequency subbands contained in the previous subsampled data. 18. The transient detector as claimed in claim 15, wherein the sound signal is a pulse code modulated signal.

Images