KR101352608B1 - A method for extending bandwidth of vocal signal and an apparatus using it - Google Patents

Abstract

In accordance with another aspect of the present invention, there is provided a method of expanding a bandwidth of a voice signal, the method comprising: decoding a received voice signal and converting the received voice signal into a frequency domain; Performing normalization on the converted speech signal; Determining a voiced sound or unvoiced sound interval from the received voice signal; Extracting a first section including a harmonic component of the voiced sound section from the normalized voice signal based on the section determined as the voiced sound; Extracting a second section from the normalized speech signal based on the cross-correlation between the section determined as the unvoiced sound and the normalized speech signal; Generating a high band speech signal based on the first interval and the second interval; And synthesizing the generated high-band speech signal with the converted speech signal and outputting the synthesized high-band speech signal.

Description

A method for extending bandwidth of vocal signal and an apparatus using it}

The present invention relates to a method and device for bandwidth expansion of a voice signal, and more particularly, to a method and device for bandwidth expansion of a voice signal capable of improving performance.

In most voice communication systems, the voice bandwidth is limited to 0.3-3.4 kHz. This voice bandwidth includes voiced and unvoiced sounds, and the bandwidth is lower than the original sound. In order to suppress such sound degradation, a broadband voice receiver has been proposed. Wideband speech with a bandwidth of 50 Hz to 7 kHz can represent all voice bands, including voiced and unvoiced, as well as increasing naturalness and clarity compared to narrowband voice. However, voice codes such as public line switched telephone networks (PSTN), Internet phones (VoIP, VoWiFi), and voice-related applications in smart phones are still being serviced as narrowband voice codecs. There is a big burden in terms of time and money.

On the other hand, according to this problem, a method for receiving a narrowband voice and converting it into a wideband signal at a decoding stage has been proposed. Accordingly, various methods for extending the voice bandwidth have been proposed.

First, there is a method of allocating additional bits for the high band. This is a method of using side information, and is a method of extending a voice band by using encoding information transmitted from an encoding end. The encoder generates and transmits additional information through analysis of highband information of the input signal, and the decoder generates a highband signal based on the transmitted additional information and the lowband signal. For example, G.729.1, a wideband voice codec, can code in 12 different layers from 8 kbit / s to 32 kbit / s. Since the base layer of G.729.1 is a representative narrowband codec, the G.729 codec, it can guarantee narrowband sound quality in 8 kbit / s mode. At this time, the encoder generates a wideband voice using the above-described bandwidth extension technique from the 14 kbit / s mode called Layer 3. The encoder allocates additional bits to use the bandwidth extension scheme used in Layer 3 of G.729.1 to produce high-band signals during decoding. However, this technique of bandwidth extension not only causes network overload by allocating additional bits, but also requires a full modification of the encoder.

In addition, a method of generating a highband signal from a lowband signal at the decoder stage without additional bit allocation has been proposed. For example, estimation methods using pattern recognition techniques such as hidden Markov model (HMM) and Gaussian mixture model (GMM) have been proposed. However, pattern recognition requires a training process, and performance may vary depending on the language. In addition, in the case of a method for prediction or estimation, additional bits are included and the amount of computation is greatly increased, making it difficult to process voice received in real time quickly and effectively. In addition, the bandwidth extension methods proposed without additional bit allocation have a problem of poor output sound quality.

SUMMARY OF THE INVENTION An object of the present invention is to provide a method and an apparatus for extending a bandwidth of a voice signal capable of quickly and effectively extending the bandwidth of a narrowband voice signal.

In addition, the present invention provides a method and apparatus for effectively expanding a bandwidth of a voice signal, which can improve the sound quality of a bandwidth-extended wideband voice signal without additional bits, thereby improving performance.

According to an aspect of the present invention, there is provided a method for extending a bandwidth of a voice signal, the method comprising: receiving a voice signal and extending the bandwidth, decoding the received voice signal and converting the received voice signal into a frequency domain; Performing normalization on the converted speech signal; Determining a voiced sound or unvoiced sound interval from the received voice signal; Extracting a first section including a harmonic component of the voiced sound section from the normalized voice signal based on the section determined as the voiced sound; Extracting a second section from the normalized speech signal based on a correlation between the section determined as the unvoiced sound and the normalized speech signal; Generating a high band speech signal based on the first interval and the second interval; And synthesizing the generated high-band speech signal with the converted speech signal and outputting the synthesized high-band speech signal.

In addition, the apparatus for bandwidth expansion of a voice signal according to an embodiment of the present invention for achieving the above object, the apparatus for bandwidth expansion of a voice signal, receiving unit for receiving a voice signal; A decoder for decoding the received speech signal; A domain converter for converting the decoded speech signal into a frequency domain; A normalizer which normalizes the converted speech signal; A discriminating unit for discriminating voiced or unvoiced sections from the received voice signal; A voiced sound processor extracting a first section including a harmonic component of the voiced sound section from the normalized voice signal based on the section determined as the voiced sound; An unvoiced sound processor extracting a second section from the normalized speech signal based on a correlation between the section determined as the unvoiced sound and the normalized speech signal; A high band generator configured to generate a high band voice signal based on the first and second sections; And an output unit configured to synthesize the generated high-band speech signal and the converted speech signal and output the wideband speech signal.

According to an embodiment of the present invention, it is possible to output a high quality wideband speech signal from a narrowband speech signal without additional bits.

In particular, by determining the voiced sound and the unvoiced sound and performing another operation, the sound quality can be improved while reducing the amount of calculation.

On the other hand, according to an embodiment of the present invention, it is possible to improve to a broadband system without changing the configuration of the decoder of the conventional narrowband voice signal system, it is possible to reduce the cost for broadband voice services.

1 is a diagram schematically illustrating an apparatus for extending a bandwidth of a voice signal according to an embodiment of the present invention.
2 is a block diagram illustrating in more detail an apparatus for extending bandwidth of a voice signal according to an embodiment of the present invention.
3 is a flowchart illustrating a bandwidth extension method of a voice signal according to an embodiment of the present invention.
4 is a diagram illustrating an experimental result using a bandwidth extension method of a voice signal according to an embodiment of the present invention.
5 to 8 are graphs showing respective waveforms of a result of experimenting with a method for expanding a bandwidth of a voice signal according to an embodiment of the present invention illustrated in FIG. 4.

The following merely illustrates the principles of the invention. Thus, those skilled in the art will be able to devise various apparatuses which, although not explicitly described or shown herein, embody the principles of the invention and are included in the concept and scope of the invention. Furthermore, all of the conditional terms and embodiments listed herein are, in principle, intended only for the purpose of enabling understanding of the concepts of the present invention, and are not intended to be limiting in any way to the specifically listed embodiments and conditions .

It is also to be understood that the detailed description, as well as the principles, aspects and embodiments of the invention, as well as specific embodiments thereof, are intended to cover structural and functional equivalents thereof. It is also to be understood that such equivalents include all elements contemplated to perform the same function irrespective of the currently known equivalents as well as the equivalents to be developed in the future, i.e., the structure.

Thus, for example, the block diagrams herein should be understood to represent a conceptual view of example circuitry embodying the principles of the invention. Similarly, all flowcharts, state transition diagrams, pseudo code, and the like are representative of various processes that may be substantially represented on a computer-readable medium and executed by a computer or processor, whether or not the computer or processor is explicitly shown .

The functionality of the various elements shown in the figures, including functional blocks represented by a processor or similar concept, can be provided by the use of dedicated hardware as well as hardware capable of executing software in association with appropriate software. When provided by a processor, the functions may be provided by a single dedicated processor, a single shared processor, or a plurality of individual processors, some of which may be shared.

Also, the explicit use of terms such as processor, control, or similar concepts should not be interpreted exclusively as hardware capable of running software, and may be used without limitation as a digital signal processor (DSP) (ROM), random access memory (RAM), and non-volatile memory. Other hardware may also be included.

In the claims hereof, the elements represented as means for performing the functions described in the detailed description include all types of software including, for example, a combination of circuit elements performing the function or firmware / microcode etc. , And is coupled with appropriate circuitry to execute the software to perform the function. It is to be understood that the invention defined by the appended claims is not to be construed as encompassing any means capable of providing such functionality, as the functions provided by the various listed means are combined and combined with the manner in which the claims require .

BRIEF DESCRIPTION OF THE DRAWINGS The above and other objects, features and advantages of the present invention will become more apparent from the following detailed description of the present invention when taken in conjunction with the accompanying drawings, in which: There will be. In the following description, well-known functions or constructions are not described in detail since they would obscure the invention in unnecessary detail.

Hereinafter, a preferred embodiment of the present invention will be described in detail with reference to the accompanying drawings.

1 is a diagram schematically illustrating an apparatus for extending a bandwidth of a voice signal according to an embodiment of the present invention.

Referring to FIG. 1, the apparatus 100 for expanding a voice signal according to an embodiment of the present invention receives a narrowband voice signal and outputs a wideband voice signal having improved sound quality. Such a bandwidth extension device 100 may be used in a decoding stage of a narrowband speech receiver, and may generate and output a wideband speech signal in which the harmonic component of the narrowband is maintained. The bandwidth extension apparatus 100 may determine the voiced sound or the unvoiced sound by using information obtained in the process of decoding the narrowband speech signal. In addition, in the case of voiced sound, the bandwidth extension device 100 obtains a wideband voice signal in which a harmonic component is maintained by using pitch information, and obtains a wideband voice signal by using a signal having the highest correlation in the case of unvoiced sound. In addition, by adjusting the energy of the obtained wideband voice signal, a wideband voice signal having improved sound quality can be output without adding bits.

2 is a block diagram illustrating in more detail an apparatus for extending bandwidth of a voice signal according to an embodiment of the present invention.

As shown in FIG. 2, the apparatus 100 for expanding bandwidth of a voice signal according to an embodiment of the present invention includes a decoder 110 that receives a voice signal and decodes it into processable data, and decodes the decoded voice signal into a frequency domain. Domain transform unit 120 for converting, normalizer 130 for normalizing the domain-converted speech signal, low band inverse transform unit 140 for inverse transforming the normalized speech signal into a low-band speech signal in the time domain, and domain transformed Determination unit 150 for discriminating voiced or unvoiced sections of the voice signal, voiced sound processor 151 for obtaining a first section including the harmonic section from the section determined as voiced sound, the highest correlation from the section determined as unvoiced sound The unvoiced sound processor 152 for acquiring the second section, the energy scaling for the first or second section obtained by processing the voiced sound or unvoiced sound, Nudge control unit 160, a high-band inverse transformer 170 for inversely converting the energy-adjusted voice section to output a high-band speech signal in the time domain, and a low-band speech signal output and a high-band speech signal output by combining a wideband speech signal It is configured to include a voice signal synthesizer 180 for outputting.

The decoder 110 receives a voice signal and decodes it into processable data. There may be various ways to decode the speech signal. For example, decoder 110 is a well-known narrowband decoding method G.729 [ITU-T Recommendation G.729, Coding of speech at 8 kbit / s using conjugate-structure code-excited linear prediction (CS-ACELP) ] To decode. In addition, the decoder 110 may include a speech decoder of a code exited linear prediction (CELP) type using spectrum analysis.

In one embodiment, the decoder 110 may extract the pitch information or the frequency gradient for the speech signal in the decoding process, and transmit it to the determination unit 150. For example, the decoder 110 may obtain a frequency gradient using a primary reflection coefficient for decoding the received voice signal into G.729 and transmit it to the determination unit 150.

In addition, in an embodiment, the decoder 110 may decode a bitstream according to the speech signal into a narrowband speech signal. For example, the G.729 format speech signal processed by the decoder 110 may be N, which is the number of samples as one frame size.

Meanwhile, the domain converter 120 converts the decoded voice signal into the frequency domain. The domain converter 120 may convert the data into the frequency domain data based on the decoded voice signal.

For example, the domain converter 120 may convert the speech signal into the frequency domain using a modified discrete cosine transform (MDCT). The domain converter 120 receives the decoded voice signal as an input signal in the time domain, converts it to an input signal in the frequency domain, and performs an inter-block overlap operation. In particular, the MDCT conversion scheme does not increase the bit rate even when performing an overlap operation. As described above, when the G.729-format speech signal is N, which is the number of samples of one frame, 80, the domain converter 120 performs 160 frequency band points having 2N from one decoded speech frame and It may be a 2N-point MDCT that outputs coefficients for.

Meanwhile, the normalization unit 130 performs normalization on the domain transformed speech signal. The normalizer 130 may divide the domain-converted speech signal data into a plurality of subbands, and normalize the energy of each subband with respect to frequency band coefficients for each subband. . For example, when 80 frequency band points are partitioned into 8 subbands, each subband may include 10 MDCT coefficients. In this case, the normalization process may be expressed as an equation.

Here, E (b) of Equation 1 may represent the energy of the b-th subband on the frequency band point of the MDCT-converted speech signal. In the present embodiment, since the number of subbands is 16, b may be an integer from 0 to 15.

는 k번째 정규화된 MDCT 계수를 의미할 수 있다.Equation 2 shows a method of normalizing the coefficients for each of the MDCT transformed frequencies using the energy of the subbands thus obtained.

May mean the k th normalized MDCT coefficient.

Meanwhile, the determination unit 150 determines the voiced sound or the unvoiced sound interval based on the normalized voice signal. The determination unit 150 may receive the frequency gradient obtained in the decoding process by the decoder 110 described above, and may determine that it is a voiced sound section when the frequency gradient is greater than or equal to a predetermined value.

For example, when the decoder 110 is a speech decoder of the CELP type, frequency gradient information may be extracted from information output from the decoder 110 to determine a voiced sound section.

를 다음과 같은 수학식 3을 통하여 획득하고, 판별부(150)로 전달할 수 있다.In addition, in the case of the decoder 110 using G.729, the frequency gradient which is the first order reflection coefficient during the decoding process.

Can be obtained through Equation 3 below, and transmitted to the determination unit 150.

은 수신한 음성 신호의 시간 도메인에서 한 프레임의 n번째 샘플의 값을 의미할 수 있다.here,

May mean the value of the n th sample of one frame in the time domain of the received voice signal.

를 디코더(110)로부터 수신하거나 이를 직접 계산하고, 기 정의된

와 비교하여

가

이상인 경우, 유성음 구간으로 판단할 수 있다.

는 사용자에 의해 미리 설정될 수 있으며, 실험 결과 바람직하게는 0.25가 사용될 수 있다. 또한, 판별부(150)는 유성음 구간으로 판단되지 않은 구간을 무성음 구간으로 판단할 수 있다. 여기서, 판별부(150)는 주파수 경사도를 직접 계산하지 않고, 음성 신호의 디코딩 과정에서 발생한 주파수 경사도 정보를 수신하여 판별함으로써, 연산량을 줄일 수 있게 된다.The determination unit 150 obtains the frequency gradient obtained as described above.

Is received from the decoder 110 or calculated directly, and

In comparison with

end

In this case, it may be determined as a voiced sound section.

May be preset by the user, and preferably 0.25 may be used as a result of the experiment. In addition, the determination unit 150 may determine a section that is not determined as a voiced sound section as an unvoiced sound section. Here, the determination unit 150 may reduce the amount of computation by receiving and determining the frequency gradient information generated in the decoding process of the voice signal without directly calculating the frequency gradient.

The voiced sound processor 151 obtains a first section including a harmonic section from the section determined as voiced sound. The voiced sound processor 151 may extract the harmonic section using the pitch information obtained by the decoder 110 described above from the section determined as the voiced sound of the normalized voice signal. Such a first section may be a harmonic section, and may include a plurality of sections, and the first section may be a plurality of sections.

The pitch information may mean a pitch period of voice, and may include location and spacing information between harmonics in the frequency domain. In the case of voiced sound, the voice signal forms a harmonic according to the period according to the pitch information, so that the voiced sound processor 151 can extract the harmonic section of the voiced sound section using the pitch information. In particular, the decoder 110 may extract the pitch information in the decoding process of the voice signal. When the calculation is performed using this, the computation amount may be reduced, and the calculation may be performed only in the voiced sound section, thereby quickly extracting the harmonic section.

Here, the decoder 110 or the voiced sound processor 151 may extract pitch information, T, by using Equations 4 and 5 shown below.

를 최대화시키는

값을 의미할 수 있다. T는 피치 값으로서,

과

는 각각 20 및 147이 바람직하다.Here, T is through

To maximize

It can mean a value. T is the pitch value

and

Is preferably 20 and 147, respectively.

The voiced sound processor 151 may extract the harmonic section from the voiced sound section based on the pitch information T thus obtained. According to the pitch information T, the harmonic interval in the 2N-point transformed MDCT frequency domain may be calculated by Equations 6 and 7 below.

는 수학식 2를 통하여 정규화부(130)에서 정규화된 음성 신호의 MDCT 계수들을 의미할 수 있다. Mod(x,y)는 x%y의 모듈러 연산을 의미할 수 있으며,

는 x를 초과하지 않는 가장 큰 정수를 의미할 수 있다. k는 샘플 개수에 따라 0에서 N/2-1의 값을 가질 수 있다. 수학식 6과 수학식 7의 연산에 의해, 출력되는

는 판별부(150)에서 판별된 유성음 구간으로부터 하모닉 구간을 추출한 MDCT 계수들을 포함할 수 있다. 따라서, 유성음 처리부(151)는

를 출력함으로써, 유성음 구간에 대한 하모닉 구간의 데이터를 추출하여 출력할 수 있다.Here, T means pitch information, N means the number of samples per frame,

Denotes MDCT coefficients of the speech signal normalized by the normalization unit 130 through Equation 2. Mod (x, y) can mean a modular operation of x% y,

May mean the largest integer that does not exceed x. k may have a value from 0 to N / 2-1 according to the number of samples. Outputted by the operations of Equations 6 and 7

May include MDCT coefficients obtained by extracting a harmonic section from the voiced sound section determined by the determination unit 150. Therefore, the voiced sound processor 151

By outputting the data of the harmonic section for the voiced sound section can be extracted and output.

On the other hand, the unvoiced sound processing unit 152 obtains the second section having the largest correlation from the section determined as unvoiced sound. The unvoiced sound processor 152 may determine the cross-correlation for each frequency section for the section determined as the unvoiced sound in the normalized voice signal, extract the section having the largest cross-correlation, and obtain the above-described second section. The acquired second period may range from 3 kHz to 4 kHz frequency band. Thereafter, the second section may be amplified and shifted to a high band and used as an unvoiced section of the high band voice signal. This can be calculated by the following equation.

는 정규화된 음성 신호중 무성음 구간인

에서 주파수 대역 차수k에 따른 최대 상관도를 만족하는 값 m을 의미할 수 있다. 따라서, m은 0에서 N/4-1 중 어느 하나의 정수일 수 있다. 그리고, 수학식 8에서의 상관도 계산을 보다 자세히 나타내면 하기의 수학식 9와 같다.As in Equation 8,

Is the unvoiced interval of the normalized speech signal.

May mean a value m that satisfies the maximum correlation according to the frequency band order k. Accordingly, m may be an integer of any one of 0 to N / 4-1. In addition, the correlation calculation in Equation 8 is shown in Equation 9 below.

Therefore, the MDCT coefficient corresponding to the frequency section with the highest correlation may be calculated as in Equation 10.

를 고주파 대역의 무성음 구간으로 출력한다. 출력되는 무성음 구간인 제2 구간은, 제1 구간과 같이 복수 개의 구간들을 포함할 수 있으며, 제2 구간이 복수개일 수도 있다.Here, k may mean any one of integers from 0 to N / 2-1. The unvoiced sound processor 15 is calculated based on the degree of correlation

Is output as unvoiced sound section of high frequency band. The second section, which is the output unvoiced section, may include a plurality of sections, like the first section, and may have a plurality of second sections.

를 제1 구간 또는 제2 구간으로 출력하고 있으나, 이와 같은 과정을 이용하면 주파수 구간의 대역폭이 절반으로 줄게 된다. 예를 들어, 구하고자 하는 대역폭이 4kHz인 경우에

는 2kHz의 대역폭을 가질 수 있다. 따라서, 유성음 처리부(151) 또는 무성음 처리부(152)는 하기의 수학식 11과 같은 연산을 수행함으로써, 대역폭을 증폭시킬 수 있다.Meanwhile, the bandwidth amplification process is performed on the output of the first section or the second section obtained by the voiced sound processor 151 or the unvoiced voice processor 152. The voiced sound processing unit 151 or the unvoiced sound processing unit 152 is output according to Equation 7 or Equation 10 described above.

Is output as the first interval or the second interval, but using this process, the bandwidth of the frequency interval is reduced by half. For example, if the bandwidth you want to find is 4 kHz

May have a bandwidth of 2 kHz. Therefore, the voiced sound processor 151 or the unvoiced sound processor 152 may amplify a bandwidth by performing an operation as shown in Equation 11 below.

는 k번째의 차수로 정규화된 주파수 도메인 영역의 MDCT 계수를 의미할 수 있다.here,

Denotes the MDCT coefficient of the frequency domain region normalized to the kth order.

Meanwhile, the energy controller 160 performs energy scaling on the MDCT frequency domain voice signal obtained by processing the voiced or unvoiced sound described above.

The energy control unit 160 serves to reduce a sudden energy change when converting the MDCT speech signal of the first or second section into a high band signal.

Therefore, the energy adjusting unit 160 adjusts the energy of the boundary portion between the low band voice signal and the voice signal when the above-described first or second section transitions to the high band, thereby scaling the abrupt change in energy. Can be adjusted. For example, the energy control unit 160 may adjust the energy scale according to a process as in Equation 12 to Equation 14 below.

는 고대역 구간의 b번째 주파수 대역에 대한 에너지를 의미할 수 있다. b는 0에서 7 사이의 정수일 수 있다. 그리고,

는 수학식 1에서 정의된 바와 같은 저대역 주파수 구간의 b번째 주파수 대역에 대한 에너지를 의미할 수 있다.here,

May mean energy for the b th frequency band of the high band period. b may be an integer between 0 and 7. And,

May mean energy for the b th frequency band of the low band frequency period as defined in Equation 1.

는 하기의 수학식 13과 같이 정해질 수 있다.And a scale factor for energy scaling at the boundary between the low band and the high band.

May be determined as in Equation 13 below.

E (15) means the energy for the subband band of the highest band among the above-described 0 to 15 subband frequency bands in the low band period, E _h (0) is the subband frequency band in the high band period It may mean energy for the first subband frequency band. As such, the energy controller 160 may obtain an energy scaling factor by obtaining a ratio of energy for the aforementioned two frequency bands.

The energy value of the scaled high band is expressed by Equation 14 below.

Since the high-band speech signal data obtained in Equation 14 requires bandwidth extension as described in Equation 11, the energy control unit 160 performs the same operation as in Equation 15 so that the bandwidth of the high-band speech signal is obtained. Can be increased.

In addition, the energy adjusting unit 160 may output a voice signal having a high band frequency using Equation 11 and Equation 15 as shown in Equation 16 below.

를 출력할 수 있다.As described above, the energy controller 160 performs energy control on the first section or the second section to be converted into the high-band speech signal based on the energy value of the normalized speech signal.

You can output

The voice signal synthesizing unit 180 may generate a wideband voice signal by converting the energy-adjusted high band voice signal and the voice signal output from the normalization unit 130, and convert the frequency signal into a time band. To this end, the speech signal synthesizing unit 180 may output a wideband speech signal by performing an operation as shown in Equation 17 below, converting the output data into a time band, and outputting the same.

Meanwhile, according to another exemplary embodiment of the present disclosure, the apparatus 100 for expanding a voice signal may further include a low band inverse transform unit 140 and a high band inverse transform unit 170 as shown in FIG. 2. .

에 대하여 시간 도메인의 저대역 음성 신호로 역변환하여 시간 도메인의

으로 출력할 수 있다.The low band inverse transform unit 140 is a normalized speech signal

Inversely transforms the time-domain low-band speech signal into

You can output

에 대하여 시간 도메인의 고대역 음성 신호로 역변환하고, 시간 도메인의

로 출력할 수 있다.And, the high band inverse transform unit 170 is an energy-controlled speech signal

Inverse to the time-domain high-band speech signal,

Can be printed as

The voice signal synthesizing unit 180 may synthesize the low band voice signal and the high band voice signal output in the time domain and output the synthesized voice signal as a filtered voice signal. To this end, the speech signal synthesis unit 180 may perform speech synthesis using a quadrature mirror filterbanks (QMF) method. QMF can be a 64-band complex QMF.

3 is a diagram for describing a bandwidth extension method of a voice signal according to an embodiment of the present invention.

Referring to FIG. 3, first, the decoder 110 receives a narrowband speech signal (S100). As described above, the method for decoding the speech signal is G.729 [ITU-T Recommendation G.729, Coding of speech at 8 kbit / s using conjugate-structure code-excited linear prediction (CS-ACELP)] ] May be used to decode. In addition, the decoder 110 may perform decoding by using a speech decoder of a code exit linear prediction (CELP) type using spectrum analysis.

The domain converter 120 converts the decoded speech signal into the frequency domain (S110). As described above, the domain converter 120 may convert the voice signal into the frequency domain by using a modified discrete cosine transform (MDCT).

As described above, the domain converter 120 may receive the decoded voice signal as an input signal in the time domain, convert it to an input signal in the frequency domain, and perform an inter-block overlap operation. In addition, when using the MDCT conversion method, there is an advantage that the bit rate does not increase.

In operation S120, the normalization unit 130 normalizes the converted speech signal. As described above, the normalizer 130 divides the domain-converted speech signal data into a plurality of subbands, obtains energy for each subband with respect to frequency band coefficients for each subband, Normalization can be performed. For example, when 80 frequency band points are divided into 16 subbands, each subband may include 5 MDCT coefficients.

를 상술한 수학식 3을 통하여 획득하여 판별할 수 있다.Thereafter, the determination unit 150 determines the voiced sound or unvoiced sound interval from the normalized voice signal (S130). As described above, the determination unit 150 may receive the frequency gradient obtained during the decoding process in the decoder 110 and may determine that the voiced sound interval is when the frequency gradient is greater than or equal to a predetermined value. For example, when the decoder 110 is a CELP type speech decoder, the determination unit 150 may extract the frequency gradient information from the information output from the decoder 110 to determine the voiced sound section. In the case of the decoder 110 using the 729, during the decoding process, the frequency gradient which is the first order reflection coefficient

It can be obtained by determining through Equation 3 described above.

The voiced sound processor 151 extracts a first section including a harmonic section calculated based on the pitch information in the case of the voiced sound section (S140), and the unvoiced sound processor 152 correlates the unvoiced sound section. Based on the extracted normalized speech signal, the section having the highest correlation with the second section (S135), each processing unit 151, 152 amplifies the bandwidth for each extracted section and transitions to a high band ( S150).

The voiced sound processor 151 may extract the harmonic section using the pitch information obtained by the decoder 110 from the section determined as the voiced sound of the normalized voice signal as described above. Such a first section may be a harmonic section, and may include a plurality of sections, and the first section may be a plurality of sections. In addition, the unvoiced sound processor 152 determines the cross-correlation of each frequency section for the section determined as unvoiced sound in the normalized speech signal, and extracts a section having the largest cross-correlation using the above-described equation. The second section described above may be obtained. Each of the processing units 151 and 152 performs bandwidth amplification for the bandwidth of the acquired section and decreases to half of the bandwidth to be expanded, and then transitions to the high band.

Thereafter, the energy controller 160 adjusts the energy scale of the output first or second section (S160). As described above, the energy adjusting unit 160 may serve to reduce a sudden energy change when converting the MDCT speech signal of the first or second section into a high band signal. Therefore, the energy adjusting unit 160 adjusts the energy of the boundary portion between the low band voice signal and the voice signal when the above-described first or second section transitions to the high band, thereby scaling the abrupt change in energy. Can be adjusted.

The voice signal synthesizer 180 synthesizes the scaled high band voice signal and the normalized voice signal, that is, the low band voice signal to obtain a wideband signal (S170), and converts the wideband voice signal into a wideband voice signal. (S180). The speech singer synthesis unit 180 may perform inverse MDCT for speech synthesis and conversion, and perform speech synthesis using the above-described QMF method for wideband speech signal synthesis.

4 is a result graph obtained by experimenting with the performance of the apparatus 100 for expanding a bandwidth of a voice signal according to an embodiment of the present invention.

MUSHRA test (ITU / ITU-R BS 1534, Method for Subjective Assessment of Intermediate Quality Level of Coding Systems, 2001.) was performed for the experiment of the present invention, and the performance of sound quality was measured by performing a spectral comparison according to the test. . Three male and three female files were used for the MUSHRA test from the speech quality assessment material (SQAM) database (EBU, Sound Quality Assessment Material Recording for Subjective Tests, 1988.).

In particular, since the SQAM voice files are sampled at stereo and 44.1 kHz, they were downsampled to 8 kHz and 16 kHz, respectively, and regenerated into mono signals for performance measurements. This is to generate a signal processed by the prior art (G.729) and the embodiment of the present invention for a sound source downsampled at 8 kHz, and the sound source downsampled at 16 kHz is a general broadband transmission technology (G.729.1). This is to obtain a signal processed by. The experimenter participated in the test and seven test subjects who did not have an auditory disease participated in the test. For each test file, the six files described above were assigned a score from 0 to 100 for each sound quality.

 As a result of the MUSHRA test, the present invention recorded about 75.5 based on 100 points of the original sound as shown in FIG. As expected, this is higher than 66 points in G.729, a conventional narrowband processing and output technique, and 87 points in G.729.1 (Layer 3), which is a method of allocating additional bits for wideband signal generation. It can be seen that it is somewhat low. However, even without the use of additional bits, it can be seen that the sound quality is improved by about 43% compared to the general narrowband transmission technology G.729, and that the degradation of sound quality is not much higher than that of the broadband transmission method using additional bits. .

5 to 8 are graphs for comparing the results of an experiment on a method of expanding a bandwidth of a voice signal according to an embodiment of the present invention shown in FIG. 4 with respective spectral waveforms of other techniques.

5 shows the spectrum of the original sound before transmission. The high band portion of FIG. 5 is not transmitted in narrowband transmission.

6 shows the spectrum of the signal reconstructed by the conventional narrowband output technique (G.729). As shown in Fig. 6, it can be seen that the high-band signal portion of the prior art shows that the voice data is not restored and thus the sound quality is degraded.

And, Figure 7 shows the spectrum of the signal reconstructed by the general wideband transmission technique (G.729.1) using additional bits. As shown in FIG. 7, it can be seen that even when using the broadband transmission technology, the data of the high-band portion is not completely recovered. When using this scheme, there is an increase in the amount of calculation due to additional bits, which requires equipment replacement. do.

8 illustrates a spectrum of a signal that receives a narrowband signal (for example, a signal coded by G.729) and restores the wideband signal according to an embodiment of the present invention. As shown in FIG. 8, although the signal of the high band part is slightly different from the original sound, it can be seen that the signal is improved compared to the conventional method of FIG. 6. In addition, it can be seen that the results of the broadband transmission using additional bits are not significantly different from those of FIG. 8.

Therefore, according to an embodiment of the present invention, it is possible to improve the sound quality due to post processing at the decoder stage without additional bit allocation. In addition, according to an embodiment of the present invention, it is possible to secure communication bandwidth between terminals while maintaining high sound quality, and it is possible to reduce time and cost due to the installation of broadband facilities because the existing network replacement and maintenance are unnecessary.

The above-described method for extending the bandwidth of an audio signal according to the present invention may be stored in a computer-readable recording medium that is produced as a program for execution in a computer, and examples of the computer-readable recording medium include ROM, RAM, CD. ROMs, magnetic tapes, floppy disks, optical data storage devices, and the like, as well as those implemented in the form of carrier waves (eg, transmission over the Internet).

The computer readable recording medium may be distributed over a networked computer system so that computer readable code can be stored and executed in a distributed manner. And, functional programs, codes and code segments for implementing the above method can be easily inferred by programmers of the technical field to which the present invention belongs.

In addition, although the preferred embodiment of the present invention has been shown and described above, the present invention is not limited to the specific embodiments described above, but the technical field to which the invention belongs without departing from the spirit of the invention claimed in the claims. Of course, various modifications can be made by those skilled in the art, and these modifications should not be individually understood from the technical spirit or prospect of the present invention.

Claims (12)

A method of receiving a voice signal to expand the bandwidth,
Decoding the received speech signal into a frequency domain;
Performing normalization on the converted speech signal;
Determining a voiced sound or unvoiced sound interval from the received voice signal;
Extracting a first section including a harmonic component of the voiced sound section from the normalized voice signal based on the section determined as the voiced sound;
Extracting a second section from the normalized speech signal based on a correlation between the section determined as the unvoiced sound and the normalized speech signal;
Generating a high band speech signal based on the first interval and the second interval; And
And synthesizing the generated high band speech signal and the converted speech signal into a wideband speech signal. The method of claim 1,
The determining of the voiced sound or unvoiced sound interval may include:
Extracting a frequency gradient from the received voice signal; And
And determining the voiced sound interval when the extracted frequency gradient is greater than a preset value. The method of claim 1,
Extracting the first section may include:
Extracting pitch information from the received voice signal;
Obtaining a harmonic section of the section determined as the voiced sound based on the extracted pitch information; And
And extracting the harmonic section into the first section. The method of claim 1,
Extracting the second section may include:
And extracting, as the second section, a section having the greatest correlation with the normalized speech signal from the section determined as the unvoiced sound. The method of claim 1,
Generating the high band speech signal,
Transitioning a bandwidth of at least one of the first section or the second section to a high band frequency band; And
And generating the high-band speech signal by performing energy compensation of the transitioned section. The method of claim 5,
Performing the energy compensation is
Dividing the normalized speech signal into a plurality of first subbands according to a frequency band;
Dividing the voice signal of the transitioned section into a plurality of second subbands;
Obtaining a scaling factor based on the first subband and the second subband; And
And performing energy compensation of the transitioned period by using the scaling factor. A device for extending the bandwidth of a voice signal,
Receiving unit for receiving a voice signal;
A decoder for decoding the received speech signal;
A domain converter for converting the decoded speech signal into a frequency domain;
A normalizer which normalizes the converted speech signal;
A discriminating unit for discriminating voiced or unvoiced sections from the received voice signal;
A voiced sound processor extracting a first section including a harmonic component of the voiced sound section from the normalized voice signal based on the section determined as the voiced sound;
An unvoiced sound processor extracting a second section from the normalized speech signal based on a correlation between the section determined as the unvoiced sound and the normalized speech signal;
A high band generator configured to generate a high band voice signal based on the first and second sections; And
And an output unit configured to synthesize the generated high-band speech signal and the converted speech signal and output a wideband speech signal. The method of claim 7, wherein
Wherein,
And extracting a frequency gradient from the received voice signal, and determining a voiced sound interval when the extracted frequency gradient is greater than a preset value. The method of claim 7, wherein
The voiced sound processor,
Bandwidth extension apparatus for extracting pitch information from the received speech signal, acquiring a harmonic section of the section determined as the voiced sound based on the extracted pitch information, and outputting the harmonic section to the first section . The method of claim 7, wherein
The unvoiced sound processing unit,
And a band having the greatest correlation with the normalized voice signal from the section determined as the unvoiced sound to the second section. The method of claim 7, wherein
The high band generation unit,
And a bandwidth of the first section or the second section transitions to a high band frequency band, and performs energy compensation of the section transitioned to the high band frequency band to generate the high band speech signal. 12. The method of claim 11,
The high band generator
The transitioned interval using the normalized speech signal divided into a plurality of first subbands according to a frequency band and a scaling factor obtained based on a speech signal of the transitioned interval divided into a plurality of second subbands Bandwidth extension device for speech signal to perform energy compensation.

Images