KR100736324B1 - Acoustic Codec Using Codebook-based Wavelet Packet Decomposition / Synthesis and Acoustic Signal Restoration Method Using the Same - Google Patents

Abstract

According to the present invention, a codebook method replaces a subband signal of a high frequency band having a low distortion rate due to phase mismatch in a method of compressing and restoring an acoustic signal using wavelet packet decomposition / synthesis. Codebook-based wavelet packet decomposition / synthesis and compression and restoration of codebook based wavelet packet decomposition / synthesis, which can increase the efficiency of saving, transmitting and restoring digital audio signals by reducing the calculation amount by 1/3 while recovering more than 2 times. A sound codec comprising an encoder unit and a decoder unit, wherein the encoder unit and the decoder unit each include a codebook, and the codebook includes a codebook when the subband signal of a high frequency band is replaced with a codebook. Extraction to reduce noise component when it is out of phase with low frequency signal It comprises the step of normalizing the amplitude of the sub-band-specific signals.

Wavelet Packet, Acoustic Signal, Codebook, Compression, Encoding, Reconstruction, Decoding

Description

CODEC using Wavelet Packet Decomposition AND Composition Based On Code Book AND Method of Decompression for Audio Signal Using Thereof}

1 is a schematic diagram showing a general wavelet packet decomposition structure;

2 is a block diagram showing a conventional wavelet encoder;

3 is a block diagram showing a conventional wavelet decoder;

4 is a schematic block diagram illustrating an acoustic codec using codebook based wavelet packet decomposition / synthesis according to the present invention;

5 is a block diagram illustrating an encoder of an acoustic codec using codebook based wavelet packet decomposition according to the present invention;

6 is a flowchart illustrating a method of compressing an audio signal using codebook based wavelet packet decomposition according to the present invention;

7 is a block diagram showing a decoder of an acoustic codec using codebook based wavelet packet synthesis according to the present invention;

8 is a flowchart illustrating a method for recovering an acoustic signal using codebook based wavelet packet synthesis according to the present invention;

9 is a flowchart illustrating a method of extracting a codebook candidate DB included in an acoustic codec according to the present invention;

10 is a waveform diagram showing amplitudes of subband signals before and after normalization;

11 is a flowchart illustrating a training method of a codebook included in an acoustic codec according to the present invention.

* Description of Signs for Main Parts in Drawings *

1: Sound Codec 3: Encoder

5: decoder section 7: codebook

The present invention relates to an acoustic codec and a method for compressing and reconstructing an acoustic signal using the same. Particularly, among subband signals decomposed into a wavelet packet, a codebook method is used for a subband signal of a high frequency band having a low distortion rate due to phase mismatch. By using the codebook-based wavelet packet decomposition / synthesis, which improves the efficiency of storing / transmitting / restoring digital sound signals by increasing the compression rate by more than 2 times compared to the conventional method, and reducing the calculation amount by 1/3. The present invention relates to a codec and a sound signal recovery method using the same.

In general, a method of compressing (encoding) an acoustic signal by wavelet packet decomposition / composition is a wavelet packet having a band pass filter characteristic as shown in FIGS. 1 and 2. This method is to apply a psychoacoustic model to each subband signal (refer to Table 1) generated by decomposing the quantization by assigning an optimal bit for each subband.

Band number Sample Count / Band Lower boundary frequency (Hz) Upper boundary frequency (Hz) One 16 0 250 2 16 250 500 3 16 500 750 4 16 750 1000 5 16 1000 1250 6 16 1250 1500 7 16 1500 1750 8 16 1750 2000 9 32 2000 2500 10 32 2500 3000 11 32 3000 3500 12 32 3500 4000 13 64 4000 5000 14 64 5000 6000 15 128 6000 8000

Table 1 shows characteristics of the signal for each subband (512 samples of 16KHz sampling / analysis frame length) when the decomposition structure of the wavelet packet is as shown in FIG. 1.

In addition, the reconstruction (decoding) method decomposes the input packet as shown in FIG. Next, a wavelet packet synthesis process is performed with an inverse wavelet filter to restore the sound signal.

In the conventional method of compressing and restoring an acoustic signal, the structure of a signal for each subband obtained by splitting a wavelet packet into frequency bands is simplified, so that the characteristics of the signal can be expressed even with low bit allocation. By not assigning or minimizing bits, high efficiency compression / restore is possible while maintaining high quality.

In addition, since the method corresponds to a one-to-one correspondence between the analysis section and the restoration section during compression, high quality compression / restore is possible without additional information of the previous frame for the sound signal of a short (for example, several tens of msec) intervals. Way.

In addition, even when packet loss occurs in transmitting the compressed data, the restoration result is superior to the general method (CELP), which requires information on a previous frame. Here, CELP refers to an ITU standard of a voice speech codec optimized for a modem. For example, the LD-CELP method is used, and provides long-distance call quality audio at 5.3 or 6.4 Kbps. To get higher speeds, the processing power is lowered, and it also refers to providing a mode that one person speaks at a time and using additional bandwidth in a conversation group.

However, in comparison with the above advantages, there was a problem that it takes a lot of calculation to decompose into a wavelet packet during compression and reconstruct when restoring, and a compression method having a compression ratio of about 8 times (20 to 30 times). There was a disadvantage that is not high compared to).

Accordingly, the present invention has been made in view of the problems of the prior art, and an object thereof is to provide a high frequency band subband signal having a low distortion rate due to phase mismatch among subband signals obtained by splitting a wavelet packet for each frequency band. By replacing the codebook method, codebook-based wavelet packet decomposition / synthesis is possible to increase the efficiency of saving, transmitting and restoring digital audio signals by reducing the calculation amount by 1/3 while recovering the compression rate more than twice as compared to the conventional method. The present invention provides a sound codec and a method of compressing and restoring a sound signal using the same.

In order to achieve the above object, the present invention is an acoustic codec having an encoder unit and a decoder unit, wherein the encoder unit and the decoder unit include a codebook, wherein the acoustic codec using wavelet packet decomposition / synthesis, respectively. To provide.

In order to reduce noise components when the subband signal does not coincide with the phase of the low frequency band subband signal when the subband signal of the wavelet packet-decomposed high frequency band is replaced with the codebook, the codebook is the amplitude of the extracted subband signal. This is done by normalizing the training and conducting a training process to include the various features.

Another aspect of the invention, the step of decomposing the wavelet packet into a subband signal for each frequency band using a wavelet filter the digital sound signal input to the encoder section of the acoustic codec; Applying a psychoacoustic model to the decomposed low frequency subband signal and quantizing an optimal bit for each subband; Normalizing the subband signal amplitude of the high frequency band to be replaced by the codebook: searching the closest code number by comparing acoustic characteristics with the codebook for the subband signals of each normalized high frequency band and converting the signal of the corresponding subband into the codebook Searching for a replaceable codebook; And formatting the bitstreaming, which is a convenient form for storing, transmitting, and restoring the result generated by the above process, and provides a method of compressing an acoustic signal using codebook based wavelet packet decomposition.

Another feature of the present invention includes the steps of: classifying header information and sample information by decomposing a packet of a compressed digital sound signal input to a decoder of an acoustic codec; Restoring the subband signal using bit allocation information and sample information on the subbands in the low frequency band; The subband signal of the high frequency band replaced by the codebook restores the codebook signal corresponding to the code number of each subband to the signal of the corresponding subband, and performs the wavelet packet synthesis using the inverse wavelet filter to perform the final sound signal. The present invention provides a method for recovering an acoustic signal using codebook based wavelet packet decomposition, characterized in that the step of recovering.

Therefore, in the present invention, a signal book for each subband of a high frequency band obtained by decomposing a wavelet packet is replaced with a codebook, which significantly reduces the amount of computation during restoration and greatly improves the compression ratio, as well as maintaining high quality sound.

Moreover, the present invention can be applied to acoustic signals, voice signals, and the like, and in particular, high performance can be obtained in the field of speech synthesis.

Hereinafter, with reference to the accompanying drawings showing a preferred embodiment of the present invention described above in more detail.

4 is a schematic block diagram illustrating an acoustic codec using codebook based wavelet packet decomposition / synthesis according to the present invention, and FIGS. 5 and 6 illustrate an encoder of an acoustic codec using codebook based wavelet packet decomposition according to the present invention. 7 and 8 are block diagrams illustrating a decoder unit of an audio codec using codebook based wavelet packet synthesis according to the present invention, and a flowchart illustrating a method of recovering an audio signal.

First, the acoustic codec 1 using codebook-based wavelet packet decomposition / synthesis according to the present invention includes an encoder unit 3 and a decoder unit 5 each having a code book 7 as shown in FIG. do.

5 and 6, when the digital sound signal is input, the encoder unit 3 is a wavelet filter having a band pass filter characteristic and decomposes the wavelet packet into subbands for each frequency band. For the low frequency band subband signal, the psychoacoustic model is applied to allocate the optimal bits for each subband and quantize them. The codebook 7 is searched for the NC (number of codebook applied band) subband signals of the high frequency band to obtain the optimal code number. Find and replace, then format it as bitstreaming. Then, if the signal is not at the end, the above process is repeated to compress the continuous sound signal.

In this way, by substituting the high frequency band subband signal having a low distortion rate due to phase mismatch among the subband signals in which the wavelet packet is decomposed by the codebook 7, the compression rate can be increased by more than twice as compared with the conventional method. Can be.

7 and 8, when the compressed digital sound signal is input, the decoder unit 5 decomposes the input packet to classify header information (frame length, bit allocation, etc.) and sample information. The subband signal is restored by using bit allocation coefficients and sample information for each subband of the low frequency band, and the subband signal of the high frequency band is obtained by restoring a codebook signal corresponding to a code number of each band into a subband signal. A wavelet packet is synthesized using an inverse wavelet filter to restore an acoustic signal. Subsequently, if the end of the bitstream is not repeated, the above process is repeated to restore a continuous sound signal.

As described above, when the digital sound signal compressed by the decoder unit 5 is restored, the wavelet packet synthesis process using the inverse wavelet filter for the high frequency band can be simplified to reduce the amount of calculation to 1/3.

As a result, through the compression and reconstruction method through the encoder unit 3 and the decoder unit 5, it is possible to maximize the efficiency in the storage / transmission / restoration of the digital sound signal.

9 is a flowchart illustrating a method of extracting a codebook candidate DB included in an acoustic codec according to the present invention, and FIG. 10 is a waveform diagram illustrating amplitudes of signals before and after normalization of subband signals.

On the other hand, when the subband signal of the high frequency band is replaced by the codebook 7, when the signal of the codebook 7 does not coincide with the phase of the low frequency band subband signal, it appears as a noise component. As shown in FIG. 9, a noise component may be minimized by performing a process of normalizing the amplitude of the extracted subband signal.

That is, the extraction method of the codebook candidate DB is the amplitude of the NC (codebook applied band) subband signal before wavelet filtering (wavelet packet decomposition) and quantizing the acoustic signal input from the large-capacity acoustic signal DB into N (total number of bands) bands. Normalize

This normalization process is a process of adjusting the magnitude of the amplitude to be similar to the average peak with respect to the signal exceeding the average peak as shown in FIG. In other words, when the difference between the average peak and the amplitude is large, it occurs mainly at the pitch starting point of the voiced sound where strong energy is excited, and when these signals do not match the pitch, they appear as noise components and the normalization process mainly suppresses these signals. .

On the other hand, the amplitude-normalized subband signal is quantized and stored in the codebook candidate DB, and if not the end of the sound signal, the above process is repeated to extract the codebook candidate DB.

11 is a flowchart illustrating a training method of a codebook included in an acoustic codec according to the present invention.

Meanwhile, the codebook 7 may be extracted from subband signals extracted from various signals, exclude similar signals, and be trained and written to include different characteristics to minimize degradation of sound quality.

For example, in order to minimize the degradation of sound quality when the number of final code words is N, a general codebook training process by VQ (vector quantization) is performed as shown in FIG. 11. Is detected. The initial code word is zero.

Subsequently, the coder divides the data into two codewords and extracts the center points of the candidates for each codeword. Then, for each candidate, each code word center and similarity are measured, and each candidate is reassigned to the code word corresponding to the most similar center point.

If the reassigned code words are optimally allocated and the code words match the total number N, the center value of each code word is stored and the codebook training process is terminated.

A sound signal compressed by the method of the present invention can maintain a compression quality of 16 times or more while maintaining the sound quality of MOS 4.0 or more at the time of restoration, and can also reduce the amount of calculation to 1/3 during restoration.

Moreover, the present invention can be applied to acoustic signals, voice signals, and the like, and in particular, high performance can be obtained in the field of speech synthesis. In addition, the present invention can be applied to both the acoustic signal compression / restoration apparatus using the filter bank method.

As described above, in the present invention, the subband signal of the wavelet packet-decomposed subband signal is replaced with a codebook, thereby significantly reducing the amount of computation at the time of restoration and greatly improving the compression ratio, as well as maintaining high quality sound quality. There is this.

Claims (4)

In the acoustic codec having an encoder unit and a decoder unit, The codec based wavelet packet decomposition / synthesis, characterized in that the encoder unit and the decoder unit comprises a codebook, respectively. The amplitude of the extracted subband signal according to claim 1, wherein the codebook reduces the noise component when the subband signal of the high frequency band does not match the phase of the low frequency band when the subband signal of the high frequency band is replaced with the codebook. A codebook-based wavelet packet decomposition / synthesis, comprising: a process for normalizing the audio codec. (delete) In the method of restoring an acoustic signal using code-based wavelet packet synthesis, characterized in that the step of performing inverse wavelet filtering, Classifying header information and sample information by decomposing a packet of a compressed digital sound signal input to a decoder of an acoustic codec; Reconstructing the subband signal of the low frequency band using bit allocation information and sample information, and reconstructing the codebook signal corresponding to the code number of the corresponding subband into the corresponding subband signal for each subband signal of the high frequency band: And restoring a final sound signal by performing wavelet packet synthesis using an inverse wavelet filter on each subband signal for each band.

Images