CN101373594A - Method and apparatus for correcting audio signal - Google Patents

Abstract

The invention belongs to the technical field of communication, and discloses a method and a device for correcting audio signals. The method comprises the following steps: intercepting a segment of signal near the boundary of an audio signal data frame; subjecting the intercepted signal to linear treatment to obtain a new signal; and calculating the evaluation index of the new signal and, if the evaluation index is smaller than a preset evaluation index, continuing the linear treatment until the evaluation index of the signal after the linear treatment is larger than or equal to the preset evaluation index. The method has the advantages that the method has less calculation amount, can achieve smooth and periodic time domain waveforms of the audio signal by correcting the audio signal, and can ensure the phase information of the signal and reduce frequency spectrum divergence, thereby achieving smooth frequency spectrum and further eliminating boundary effect.

Description

Revise the method and the device of sound signal

Technical field

The present invention relates to communication technical field, relate in particular to a kind of method and device of revising sound signal.

Background technology

Transform domain coding is one of compress technique of generally adopting of current audio coding standard, belong to Frequency Domain Coding, by reducing the correlativity between each component in the signal, and the coefficient after the conversion quantized and encode, to reach the purpose of Information Compression, make full use of the auditory properties of people's ear on frequency domain,, realize compression sound signal as masking effect and critical band.In actual applications, usually sound signal is divided into some independently Frames and carries out FFT (Fast Fourier Transform, fast fourier transform) or DCT (Discrete Cosine Transform, discrete cosine transform), but can not guarantee that each frame signal is continuous at the edge, can not guarantee that signal can continuous extension become the periodic signal sequence, the saltus step meeting of signal at the data block edge makes the energy spectrum of signal disperse and do not concentrate, thereby produces a large amount of high-frequency signals; In addition, coefficient to FFT or DCT carries out quantization encoding, produces quantization error so inevitably, and the error that this quantification brought is amplified manyfold after the comprehensive window effect when synthetic audio signal, make the serious distortion of synthetic audio signal, promptly produce boundary effect.

Boundary effect is discontinuous causing between Frame by sound signal, and the naturalness of sound signal and intelligibility are seriously influenced, and has influenced the effect of scrambler, and audio quality is seriously descended; And sound signal is sounded have obvious periodic " too " noise, on spectrogram, show as: " the noise vertical line " of property at interval occur significantly.

In the prior art, in order to eliminate boundary effect, usually adopt MDCT (Modified DiscreteCosine, revise discrete cosine transform) as the time-frequency conversion instrument, utilize 50% the overlapping and time domain aliasing elimination bank of filters of sampling point, under the situation that does not reduce the transition coding performance, overcome the boundary effect in FFT, the DCT processing computing.For DCT, MDCT has taked 50% data overlap technique, that is: a back half data of the first half data adjacent data blocks preceding with it of current data block is overlapping, and then a half data is overlapping with the first half data of adjacent data blocks thereafter.

The direct transform of MDCT conversion is defined as follows:

$X_i\left(k\right)=\underset{n=0}{\overset{N-1}{\mathrm{\Σ}}}{w}_a\left(n\right)x_i\left(n\right)\cos \left[\frac{(2k+1)\mathrm{\π}}{N}(n+n_0)\right],k=\mathrm{0,1},\·\·\·\frac{N}{2}-1---\left(1\right)$

Wherein, $n_0=\frac{N/2+1}{2}$ It is the phase variant of MDCT.From the MDCT definition as can be known, the data block length N that carries out conversion is necessary for even number, and MDCT obtains N/2 domain samples with the conversion of N audio frequency time domain samples.

The inverse transformation of MDCT conversion is defined as follows:

$y_i\left(n\right)=w_s\left(n\right)\frac{4}{N}\underset{n=0}{\overset{N/2-1}{\mathrm{\Σ}}}{X}_i\left(k\right)\cos \left[\frac{(2k+1)\mathrm{\π}}{N}(n+n_0)\right],n=\mathrm{0,1},\·\·\·\frac{N}{2}-1---\left(2\right)$

The MDCT inverse transformation is with N time-domain audio sample of N/2 frequency-region signal sample calculation.

Carry out time-frequency conversion and handle after sample of signal is divided into relatively independent Frame, can distort at the edge of data block, solving this effective method is to take the data overlap technique in adjacent data interframe.By as can be known above-mentioned, the data of MDCT employing 50% are overlapping, and utilize analysis, comprehensive window w _a(n), w _s(n) further weakened uncontinuity between Frame.Therefore, MDCT has subdued boundary effect to a certain extent, has improved the intelligibility of coded audio, has improved coding quality.

Yet, the MDCT coefficient produces quantization error inevitably, and this error can influence the continuity of interframe, so MDCT can not eliminate the influence that boundary effect is brought fully, when signal changes more violently, the boundary effect of particularly handling after the stronger sound signal of energy is particularly evident; And handling on the multichannel coding, in decoding end energy is being redistributed and to be made each sound channel discontinuous, even more serious boundary effect occurring.

Summary of the invention

The embodiment of the invention provides a kind of method and device of revising sound signal, can effectively eliminate boundary effect.

A kind of method of revising sound signal comprises:

Near the intercepting one segment signal audio signal data frame boundaries;

Described signal to intercepting carries out linear process, obtains new signal;

Calculate the assessment indicator of described new signal, when described assessment indicator during less than the assessment indicator that presets, proceed linear process, the assessment indicator of the signal after linear process is more than or equal to the described assessment indicator that presets.

A kind of device of revising sound signal comprises:

The intercept signal unit is used near the intercepting one segment signal audio signal data frame boundaries;

The linear process unit is used for the signal that receives is carried out linear process, obtains new signal;

Computing unit is used to calculate the assessment indicator of described new signal;

Comparing unit, be used for receiving assessment indicator from computing unit, the size of more described assessment indicator and the assessment indicator that presets, when described assessment indicator during less than the assessment indicator that presets, described new signal is sent to the linear process unit, up to the assessment indicator that receives more than or equal to the described assessment indicator that presets.

By technique scheme as can be known, because the discontinuous adjacent data frames boundary vicinity that occurs in of sound signal, therefore near the audio signal data frame boundaries, intercept a segment signal, described signal to intercepting carries out linear process, replace original saltus step or the discontinuous signal of taking place with the signal after the linear process, and calculating is through the assessment indicator of the new signal after the linear process, when through the assessment indicator of the new signal after the linear process during less than the assessment indicator that presets, proceed linear process, further reduce the uncontinuity of adjacent data frames boundary vicinity signal, the assessment indicator of the signal after linear process is more than or equal to the assessment indicator that presets, make the signal of adjacent data frames boundary vicinity have continuity, realized eliminating the purpose of boundary effect.

Description of drawings

The method flow diagram that Fig. 1 provides for the embodiment of the invention;

The device synoptic diagram that Fig. 2 provides for the embodiment of the invention;

The device synoptic diagram that Fig. 3 provides for the embodiment of the invention one;

The device synoptic diagram that Fig. 4 provides for the embodiment of the invention two;

The device synoptic diagram that Fig. 5 provides for the embodiment of the invention three;

The device synoptic diagram that Fig. 6 provides for the embodiment of the invention four;

The device synoptic diagram that Fig. 7 provides for the embodiment of the invention five;

The device synoptic diagram that Fig. 8 provides for the embodiment of the invention six;

The device synoptic diagram that Fig. 9 provides for the embodiment of the invention seven;

The device synoptic diagram that Figure 10 provides for the embodiment of the invention eight.

Embodiment

The embodiment of the invention provides a kind of method and device of revising sound signal, be used to revise the uncontinuity of sound signal adjacent data interframe, make that the waveform of revising the back sound signal is level and smooth, and then the purpose of boundary effect is eliminated in realization, for the technical scheme that makes the embodiment of the invention clearer, enumerating embodiment in detail, below describes.

At first, the method that the embodiment of the invention is provided is carried out describe, in general terms.

Referring to Fig. 1, the method flow diagram that provides for the embodiment of the invention:

11): near the intercepting one segment signal audio signal data frame boundaries, for example, the X point is the frontier point of signal 1, because the frame length of general sound spectrograph is 256 points, then begin to intercept 128 points respectively forward, intercept 128 points backward, form a frame signal at 256 of intercepting from the X point;

12): the described signal to intercepting carries out linear process, obtains new sound signal;

13): the assessment indicator that calculates new sound signal, as, signal to noise ratio (S/N ratio), distortion spectrum degree, Mean Opinion Score, when judging described assessment indicator less than the assessment indicator that presets, return step 12), proceed linear process, the assessment indicator of the signal after linear process is more than or equal to the described assessment indicator that presets, otherwise, finish linear process.

Wherein, the described assessment indicator that presets is for eliminating signal to noise ratio (S/N ratio), distortion spectrum degree, the Mean Opinion Score of the pairing sound signal of boundary effect.

The method that provides at the embodiment of the invention is enumerated embodiment respectively and is elaborated below:

Embodiment one:

101: near the intercepting one segment signal audio signal data frame boundaries; For example, the X point is the frontier point of signal 1, because the frame length of general sound spectrograph is 256 points, then begins to intercept 128 points respectively forward from the X point, intercepts 128 points backward, forms a frame signal at 256 of intercepting;

102: the signal to intercepting carries out LP (Linear Prediction, linear prediction) analysis, obtains predictive coefficient, utilizes formula then $s\′\left(n\right)=\underset{i=0}{\overset{p}{\mathrm{\Σ}}}{a}_{i}s(n-i)$ Carry out linear prediction, near the hop value the predicted value surrogate data method frame boundaries of gained obtains new sound signal;

Wherein, s ' (n) represents predicted value, and p represents prediction order, a _iRepresent predictive coefficient;

Wherein, it is one of the most effective speech analysis techniques that LP analyzes, and can approach with the linear combination of several sampling values of past by the sample value of a voice signal, and guest (Durbin) Du predication method relatively more commonly used is found the solution linear predictor coefficient.

103: the assessment indicator that calculates new sound signal, as, signal to noise ratio (S/N ratio), distortion spectrum degree, Mean Opinion Score, when described assessment indicator during less than the assessment indicator that presets, return 102 steps, proceed linear prediction, the assessment indicator of the signal after linear process is more than or equal to the described assessment indicator that presets.

Among this embodiment, near the sound signal the data frame boundaries is calculated the data that the bigger data of correlativity replace the less discontinuous or saltus step of original correlativity by LP, make sound signal have continuity.

Embodiment two:

201: near the intercepting one segment signal audio signal data frame boundaries; For example, the X point is the frontier point of signal 1, because the frame length of general sound spectrograph is 256 points, then begins to intercept 128 points respectively forward from the X point, intercepts 128 points backward, forms a frame signal at 256 of intercepting;

202: the signal to intercepting carries out LP (Linear Prediction, linear prediction) analysis, obtains predictive coefficient, utilizes formula then $s\′\left(n\right)=\underset{i=0}{\overset{p}{\mathrm{\Σ}}}{a}_{i}s(n-i)$ Carry out linear prediction, near the hop value the predicted value surrogate data method frame boundaries of gained obtains new sound signal;

Wherein, s ' (n) represents predicted value, and p represents prediction order, a _iRepresent predictive coefficient;

Wherein, it is one of the most effective speech analysis techniques that LP analyzes, and can approach with the linear combination of several sampling values of past by the sample value of a voice signal, and guest (Durbin) Du predication method relatively more commonly used is found the solution linear predictor coefficient.

203: at least two data points before and after the Frame saltus step of new sound signal are averaged, make linearity curve with described data point and described mean value; With mean value is reference point, and perhaps any one point of mean value front is for reference point, and perhaps any one point of mean value back is reference point, carries out linear interpolation, replaces the data of original relevant position with the data of interpolation, further obtains new sound signal;

204: the assessment indicator that calculates new sound signal, as, signal to noise ratio (S/N ratio), distortion spectrum degree, Mean Opinion Score, when described assessment indicator during less than the assessment indicator that presets, return 202 steps, proceed linear prediction, the assessment indicator of the signal after linear process is more than or equal to the described assessment indicator that presets.

Embodiment two compares with embodiment one, and sound signal is being carried out linear prediction, eliminates on the basis of distortion, adopt the method for linear interpolation, replace the point of original relevant position with the value of linear interpolation, make sound signal have continuity, further guaranteed the uncontinuity of sound signal.

Implement three:

301: near the intercepting one segment signal audio signal data frame boundaries; For example, the X point is the frontier point of signal 1, because the frame length of general sound spectrograph is 256 points, then begins to intercept 128 points respectively forward from the X point, intercepts 128 points backward, forms a frame signal at 256 of intercepting;

302: the signal to intercepting carries out LP (Linear Prediction, linear prediction) analysis, obtains predictive coefficient, utilizes formula then $s\′\left(n\right)=\underset{i=0}{\overset{p}{\mathrm{\Σ}}}{a}_{i}s(n-i)$ Carry out linear prediction, near the hop value the predicted value surrogate data method frame boundaries of gained obtains new sound signal;

Wherein, s ' (n) represents predicted value, and p represents prediction order, a _iRepresent predictive coefficient;

Wherein, it is one of the most effective speech analysis techniques that LP analyzes, and can approach with the linear combination of several sampling values of past by the sample value of a voice signal, and guest (Durbin) Du predication method relatively more commonly used is found the solution linear predictor coefficient.

303: described new sound signal is carried out fast fourier transform, time domain is become frequency domain; Intercept the HFS of described frequency domain, carry out forward, perhaps dislocation addition is backward averaged, and replaces described HFS with described mean value; HFS after substituting with mean value is carried out invert fast fourier transformation, obtain new sound signal, replace fast fourier transform signal before with described new signal;

304: the assessment indicator that calculates new sound signal, as, signal to noise ratio (S/N ratio), distortion spectrum degree, Mean Opinion Score, when described assessment indicator during less than the assessment indicator that presets, return 302 steps, proceed linear prediction, the assessment indicator of the signal after linear process is more than or equal to the described assessment indicator that presets.

This embodiment compares with embodiment two, and the method correction frequency spectrum that the method for embodiment two neutral line interpolations is replaced with frequency domain smoothing is dispersed, and makes the spectral smoothing of sound signal, reaches the purpose of eliminating boundary effect.

Embodiment four:

401: near the intercepting one segment signal audio signal data frame boundaries; For example, the X point is the frontier point of signal 1, because the frame length of general sound spectrograph is 256 points, then begins to intercept 128 points respectively forward from the X point, intercepts 128 points backward, forms a frame signal at 256 of intercepting;

402: the signal to intercepting carries out LP (Linear Prediction, linear prediction) analysis, obtains predictive coefficient, utilizes formula then $s\′\left(n\right)=\underset{i=0}{\overset{p}{\mathrm{\Σ}}}{a}_{i}s(n-i)$ Carry out linear prediction, near the hop value the predicted value surrogate data method frame boundaries of gained obtains new sound signal;

Wherein, s ' (n) represents predicted value, and p represents prediction order, a _iRepresent predictive coefficient;

Wherein, it is one of the most effective speech analysis techniques that LP analyzes, and can approach with the linear combination of several sampling values of past by the sample value of a voice signal, and guest (Durbin) Du predication method relatively more commonly used is found the solution linear predictor coefficient.

403: at least two data points before and after the Frame saltus step of new sound signal are averaged, make linearity curve with described data point and described mean value; With mean value is reference point, and perhaps any one point of mean value front is for reference point, and perhaps any one point of mean value back carries out linear interpolation for reference point, replaces the data of original relevant position with the data of interpolation, further obtains new sound signal;

404:, time domain is become frequency domain to carrying out fast fourier transform through the sound signal that obtains after the linear interpolation; Intercept the HFS of described frequency domain, carry out forward, perhaps dislocation addition is backward averaged, and replaces described HFS with described mean value; HFS after substituting with mean value is carried out invert fast fourier transformation, obtain new sound signal, replace fast fourier transform signal before with described new signal;

405: the assessment indicator that calculates new sound signal, as, signal to noise ratio (S/N ratio), distortion spectrum degree, Mean Opinion Score, when described assessment indicator during less than the assessment indicator that presets, return 402 steps, proceed linear prediction, the assessment indicator of the signal after linear process is more than or equal to the described assessment indicator that presets.

This embodiment is carrying out embodiment two and three combinations of embodiment on the basis of linear interpolation to sound signal, further adopt the method correction frequency spectrum of frequency domain smoothing to disperse, and makes the spectral smoothing of sound signal, reaches the purpose of eliminating boundary effect.

Implement five:

501: near the intercepting one segment signal audio signal data frame boundaries; For example, the X point is the frontier point of signal 1, because the frame length of general sound spectrograph is 256 points, then begins to intercept 128 points respectively forward from the X point, intercepts 128 points backward, forms a frame signal at 256 of intercepting;

502: intercepting a segment signal backward is set to the odd number sign in the past in the signal of intercepting, from after intercept a segment signal forward and be set to the even number sign, the signal of the signal of odd number sign and even number sign is carried out the predicted value that the signal of the predicted value of signal of odd number sign and even number sign is obtained in linear prediction, and the predicted value of the signal of dual numbers sign is carried out back to front again, predicted value to the signal of odd number sign, average with predicted value, with near the hop value of described mean value surrogate data method frame boundaries through the signal of the even number sign of back to front;

503: the signal to intercepting carries out LP (Linear Prediction, linear prediction) analysis, obtains predictive coefficient, utilizes formula then $s\′\left(n\right)=\underset{i=0}{\overset{p}{\mathrm{\Σ}}}{a}_{i}s(n-i)$ Carry out linear prediction, near the hop value the predicted value surrogate data method frame boundaries of gained obtains new sound signal;

Wherein, s ' (n) represents predicted value, and p represents prediction order, a _iRepresent predictive coefficient;

Wherein, it is one of the most effective speech analysis techniques that LP analyzes, and can approach with the linear combination of several sampling values of past by the sample value of a voice signal, and guest (Durbin) Du predication method relatively more commonly used is found the solution linear predictor coefficient.

504: the assessment indicator that calculates new sound signal, as, signal to noise ratio (S/N ratio), distortion spectrum degree, Mean Opinion Score, when described assessment indicator during less than the assessment indicator that presets, return 502 steps, proceed linear prediction, the assessment indicator of the signal after linear process is more than or equal to the described assessment indicator that presets.

This embodiment compares with embodiment one, utilizing the method for near the sound signal the data frame boundaries being carried out having increased in the linear prediction parity flag, guarantees the accuracy of linear prediction, and then better revises sound signal, makes sound signal have continuity.

Wherein, the step 502 among this embodiment constitutes an embodiment before can laying respectively at 202 among the embodiment two; Be arranged in 302 of embodiment three and constitute an embodiment before; Be arranged in 402 of embodiment four and constitute an embodiment before, processing procedure is with embodiment five.

Embodiment six:

601: near the intercepting one segment signal audio signal data frame boundaries; For example, the X point is the frontier point of signal 1, because the frame length of general sound spectrograph is 256 points, then begins to intercept 128 points respectively forward from the X point, intercepts 128 points backward, forms a frame signal at 256 of intercepting;

602: at least two data points before and after the Frame saltus step of sound signal are averaged, make linearity curve with described data point and described mean value; With mean value is reference point, perhaps any one point of mean value front is reference point, and perhaps any one point of mean value back is reference point, interpolative data on described linearity curve, replace the data of original relevant position with the data of interpolation, obtain new sound signal;

603: the assessment indicator that calculates new sound signal, as, signal to noise ratio (S/N ratio), distortion spectrum degree, Mean Opinion Score, when described assessment indicator during less than the assessment indicator that presets, return 602 steps, proceed linear prediction, the assessment indicator of the signal after linear process is more than or equal to the described assessment indicator that presets.

This embodiment compares with embodiment one, with the method for the linear prediction among the embodiment one, replace with the method for linear interpolation, replace the point of original relevant position with the value of linear interpolation, make sound signal have continuity, and on frequency domain, revised because the discontinuous sawtooth wave that causes of data.

Embodiment seven:

701: near the intercepting one segment signal audio signal data frame boundaries; For example, the X point is the frontier point of signal 1, because the frame length of general sound spectrograph is 256 points, then begins to intercept 128 points respectively forward from the X point, intercepts 128 points backward, forms a frame signal at 256 of intercepting;

702: at least two data points before and after the Frame saltus step of sound signal are averaged, make linearity curve with described data point and described mean value; With mean value is reference point, perhaps any one point of mean value front is reference point, and perhaps any one point of mean value back is reference point, interpolative data on described linearity curve, replace the data of original relevant position with the data of interpolation, obtain sound signal through linear interpolation;

703: the sound signal through linear interpolation is carried out fast fourier transform, time domain is become frequency domain; Intercept the HFS of described frequency domain, carry out forward, perhaps dislocation addition is backward averaged, and replaces described HFS with described mean value; HFS after substituting with mean value is carried out invert fast fourier transformation, obtain new sound signal, replace fast fourier transform signal before with described new signal;

704: the assessment indicator that calculates new sound signal, as, signal to noise ratio (S/N ratio), distortion spectrum degree, Mean Opinion Score, when described assessment indicator during less than the assessment indicator that presets, return 702 steps, proceed linear prediction, the assessment indicator of the signal after linear process is more than or equal to the described assessment indicator that presets.

This embodiment compares with embodiment six, sound signal is being carried out on the basis of linear interpolation, further adopts the method correction frequency spectrum of frequency domain smoothing to disperse, and makes the spectral smoothing of sound signal, reaches the purpose of eliminating boundary effect.

Embodiment eight:

801: near the intercepting one segment signal audio signal data frame boundaries; For example, the X point is the frontier point of signal 1, because the frame length of general sound spectrograph is 256 points, then begins to intercept 128 points respectively forward from the X point, intercepts 128 points backward, forms a frame signal at 256 of intercepting;

802: the sound signal that receives is carried out fast fourier transform, time domain is become frequency domain; Intercept the HFS of described frequency domain, carry out forward, perhaps dislocation addition is backward averaged, and replaces described HFS with described mean value; HFS after substituting with mean value is carried out invert fast fourier transformation, obtain new sound signal, replace fast fourier transform signal before with described new signal;

803: the assessment indicator that calculates new sound signal, as, signal to noise ratio (S/N ratio), distortion spectrum degree, Mean Opinion Score, when described assessment indicator during less than the assessment indicator that presets, return 802 steps, proceed linear prediction, the assessment indicator of the signal after linear process is more than or equal to the described assessment indicator that presets.

This embodiment compares with embodiment one, will eliminate the method correction frequency spectrum that the method for distortion replaces with frequency domain smoothing by linear prediction and disperse, and makes the spectral smoothing of sound signal, reaches the purpose of eliminating boundary effect.

Wherein, there is no strict sequential between the step in the various embodiments described above and close preface, each label is the process of the representative realization embodiment of the invention just.

With reference to the accompanying drawings, the device that the embodiment of the invention is provided is elaborated:

Referring to Fig. 2, the device synoptic diagram for this law embodiment comprises:

Intercept signal unit 201 is used near the intercepting one segment signal audio signal data frame boundaries;

Linear process unit 202 is used for the signal that receives is carried out linear process, obtains new signal;

Computing unit 203 is used to calculate the assessment indicator of described new signal, as, signal to noise ratio (S/N ratio), distortion spectrum degree, Mean Opinion Score;

Comparing unit 204, be used for receiving assessment indicator from computing unit 203, the size of more described assessment indicator and the assessment indicator that presets, when described assessment indicator during less than the assessment indicator that presets, described new signal is sent to linear process unit 202, up to the assessment indicator that receives more than or equal to the described assessment indicator that presets.

Above-mentioned is the describe, in general terms of device synoptic diagram that the embodiment of the invention is provided, enumerates embodiment below respectively and is described in detail:

Referring to Fig. 3, the device synoptic diagram for the embodiment of the invention one provides comprises:

Intercept signal unit 201 is used near the intercepting one segment signal audio signal data frame boundaries;

Linear prediction unit 301 is used for the signal that receives is carried out linear prediction, obtains predicted value;

Replace unit 302, near the hop value the predicted value surrogate data method frame boundaries that is used for receiving obtains new signal;

Computing unit 203 is used to calculate the assessment indicator of described new signal, as, signal to noise ratio (S/N ratio), distortion spectrum degree, Mean Opinion Score;

Comparing unit 204, be used for receiving assessment indicator from computing unit 203, the size of more described assessment indicator and the assessment indicator that presets, when described assessment indicator during less than the assessment indicator that presets, described new signal is sent to linear prediction unit 301, up to the assessment indicator that receives more than or equal to the described assessment indicator that presets.

Wherein, linear prediction unit 301 is replaced unit 302, places linear process unit 202.

Referring to Fig. 4, the device synoptic diagram for the embodiment of the invention two provides comprises:

Intercept signal unit 201 is used near the intercepting one segment signal audio signal data frame boundaries;

Linear prediction unit 301 is used for the signal that receives is carried out linear prediction, obtains predicted value;

Replace unit 302, near the hop value the predicted value surrogate data method frame boundaries that is used for receiving obtains new signal.

Drawing unit 401 is used at least two data points before and after the Frame saltus step are averaged, and makes linearity curve with described data point and described mean value;

Linear interpolation unit 402, being used for described mean value is reference point, perhaps any one point of mean value front is reference point, perhaps any one point of mean value back is reference point, interpolative data on described linearity curve, replace the data of original relevant position with the data of interpolation, obtain new signal;

Computing unit 203 is used to calculate the assessment indicator of described new signal, as, signal to noise ratio (S/N ratio), distortion spectrum degree, Mean Opinion Score;

Comparing unit 204, be used for receiving assessment indicator from computing unit 203, the size of more described assessment indicator and the assessment indicator that presets, when described assessment indicator during less than the assessment indicator that presets, described new signal is sent to linear linear prediction unit 301, up to the assessment indicator that receives more than or equal to the described assessment indicator that presets.

Wherein, linear prediction unit 301, replacement unit 302, drawing unit 401, linear interpolation unit 402 place linear process unit 202.

The synoptic diagram that embodiment two provides is compared with the synoptic diagram that embodiment one provides, and has increased drawing unit 401 and linear interpolation unit 402, is used for sound signal is carried out the basis that distortion is eliminated in linear prediction, further revises the uncontinuity of sound signal.

Referring to Fig. 5, the device synoptic diagram for the embodiment of the invention three provides comprises:

Intercept signal unit 201 is used near the intercepting one segment signal audio signal data frame boundaries;

Linear prediction unit 301 is used for the signal that receives is carried out linear prediction, obtains predicted value;

Replace unit 302, near the hop value the predicted value surrogate data method frame boundaries that is used for receiving obtains new signal;

Fourier transformation unit 501 is used for described new signal is carried out fast fourier transform, and time domain is become frequency domain;

Frequency domain smoothing unit 502 is used to intercept the HFS of described frequency domain, carries out forward, and perhaps dislocation addition is backward averaged, and replaces described HFS with described mean value;

Inverse fourier transform unit 503 is used for the HFS after substituting with mean value is carried out invert fast fourier transformation, obtains new signal, replaces fast fourier transform signal before with the new signal behind the Fourier transform;

Computing unit 203 is used to calculate the assessment indicator of described new signal, as, signal to noise ratio (S/N ratio), distortion spectrum degree, Mean Opinion Score;

Comparing unit 204, be used for receiving assessment indicator from computing unit 203, the size of more described assessment indicator and the assessment indicator that presets, when described assessment indicator during less than the assessment indicator that presets, described new signal is sent to linear prediction unit 301, up to the assessment indicator that receives more than or equal to the described assessment indicator that presets.

Wherein, linear prediction unit 301, replacement unit 302, Fourier transformation unit 501, frequency domain smoothing unit 502, inverse fourier transform unit 503 place linear process unit 202.

The device synoptic diagram that this embodiment provides is compared with the device synoptic diagram that embodiment two provides, drawing unit among the embodiment two 401 and linear interpolation unit 402 are replaced with Fourier transformation unit 501, frequency domain smoothing unit 502, inverse fourier transform unit 503, being used to revise frequency spectrum disperses, make the spectral smoothing of sound signal, reach the purpose of eliminating boundary effect.

Referring to Fig. 6, the device synoptic diagram for the embodiment of the invention four provides comprises:

Intercept signal unit 201 is used near the intercepting one segment signal audio signal data frame boundaries;

Linear prediction unit 301 is used for the signal that receives is carried out linear prediction, obtains predicted value;

Replace unit 302, near the hop value the predicted value surrogate data method frame boundaries that is used for receiving obtains new signal;

Drawing unit 401 is used at least two data points before and after the Frame saltus step are averaged, and makes linearity curve with described data point and described mean value;

Linear interpolation unit 402, being used for described mean value is reference point, perhaps any one point of mean value front is reference point, perhaps any one point of mean value back is reference point, interpolative data on described linearity curve, replace the data of original relevant position with the data of interpolation, obtain new signal;

Fourier transformation unit 501 is used for described new signal is carried out fast fourier transform, and time domain is become frequency domain;

Frequency domain smoothing unit 502 is used to intercept the HFS of described frequency domain, carries out forward, and perhaps dislocation addition is backward averaged, and replaces described HFS with described mean value;

Inverse fourier transform unit 503 is used for the HFS after substituting with mean value is carried out invert fast fourier transformation, obtains new signal, replaces fast fourier transform signal before with the new signal behind the Fourier transform;

Computing unit 203 is used to calculate the assessment indicator of described new signal, as, signal to noise ratio (S/N ratio), distortion spectrum degree, Mean Opinion Score;

Comparing unit 204, be used for receiving assessment indicator from computing unit 203, the size of more described assessment indicator and the assessment indicator that presets, when described assessment indicator during less than the assessment indicator that presets, described new signal is sent to linear prediction unit 301, up to the assessment indicator that receives more than or equal to the described assessment indicator that presets.

Wherein, linear prediction unit 301, replacement unit 302, drawing unit 401, linear interpolation unit 402, Fourier transformation unit 501, frequency domain smoothing unit 502, inverse fourier transform unit 503 place linear process unit 202.

The device synoptic diagram that this embodiment provides is compared the device synoptic diagram that embodiment one provides, drawing unit 401, linear interpolation unit 402, Fourier transformation unit 501, frequency domain smoothing unit 502 and inverse fourier transform unit 503 have been increased, be used on the basis of sound signal being carried out linear interpolation, further adopt the method correction frequency spectrum of frequency domain smoothing to disperse, make the spectral smoothing of sound signal, reach the purpose of eliminating boundary effect.

Referring to Fig. 7, the device synoptic diagram for the embodiment of the invention five provides comprises:

Intercept signal unit 201 is used near the intercepting one segment signal audio signal data frame boundaries;

Linear prediction unit 301 is used for the signal that receives is carried out linear prediction, obtains predicted value;

Parity flag unit 701 is used for from intercept signal unit 201 received signals, and intercepting a segment signal backward is set to the odd number sign in the past to described signal, and sends to described linear prediction unit 301, obtains the predicted value of the signal of odd number sign; From after intercept a segment signal forward and be set to the even number sign;

Back to front unit 702 is used for carrying out forward direction and being inverted when described linear prediction unit 301 receives the predicted value of signal of even number sign;

Mean value unit 703, be used to receive the predicted value of the signal of described odd number sign, and the predicted value of the signal of the even number sign of process back to front, and average to the predicted value of the signal of described odd number sign with through the predicted value of the signal of the even number sign of back to front, obtain the predicted value of intercept signal unit intercept signal.

Replace unit 302, near the hop value the predicted value surrogate data method frame boundaries that is used for receiving obtains new signal;

Computing unit 203 is used to calculate the assessment indicator of described new signal, as, signal to noise ratio (S/N ratio), distortion spectrum degree, Mean Opinion Score;

Comparing unit 204, be used for receiving assessment indicator from computing unit, the size of more described assessment indicator and the assessment indicator that presets, when described assessment indicator during less than the assessment indicator that presets, described new signal is sent to linear prediction unit 301, up to the assessment indicator that receives more than or equal to the described assessment indicator that presets.

Wherein, linear prediction unit 301, replacement unit 302, drawing unit 401, linear interpolation unit 402, parity flag unit 701, back to front unit 702, mean value unit 703 place linear process unit 202.

Wherein, the device that parity flag unit 701 in the device synoptic diagram that provides among this embodiment, back to front unit 702, mean value unit 703 can be respectively provide with embodiment two, execute the device that example three provides, the device that embodiment four provides, further constitute new device.

The device synoptic diagram that this embodiment provides is compared with the device synoptic diagram that embodiment one provides, parity flag unit 701, back to front unit 702, mean value unit 703 have been increased, be used to guarantee the accuracy of linear prediction, and then better revise sound signal, make sound signal have continuity.

Referring to Fig. 8, the device synoptic diagram for the embodiment of the invention six provides comprises:

Intercept signal unit 201 is used near the intercepting one segment signal audio signal data frame boundaries;

Drawing unit 401 is used at least two data points before and after the Frame saltus step are averaged, and makes linearity curve with described data point and described mean value;

Linear interpolation unit 402, being used for described mean value is reference point, perhaps any one point of mean value front is reference point, perhaps any one point of mean value back is reference point, interpolative data on described linearity curve, replace the data of original relevant position with the data of interpolation, obtain new signal;

Computing unit 203 is used to calculate the assessment indicator of described new signal, as, signal to noise ratio (S/N ratio), distortion spectrum degree, Mean Opinion Score;

Comparing unit 204, be used for receiving assessment indicator from computing unit 203, the size of more described assessment indicator and the assessment indicator that presets, when described assessment indicator during less than the assessment indicator that presets, described new signal is sent to drawing unit 401, up to the assessment indicator that receives more than or equal to the described assessment indicator that presets.

Wherein, drawing unit 401, linear interpolation unit 402 place linear process unit 202.

The synoptic diagram that this embodiment provides is compared with the synoptic diagram that embodiment one provides, and with linear prediction unit 301, replaces unit 302, replaces with drawing unit 401 and linear interpolation unit 402, is used to revise the uncontinuity of sound signal.

Referring to Fig. 9, the device synoptic diagram for the embodiment of the invention seven provides comprises:

Intercept signal unit 201 is used near the intercepting one segment signal audio signal data frame boundaries;

Drawing unit 401 is used at least two data points before and after the Frame saltus step are averaged, and makes linearity curve with described data point and described mean value;

Linear interpolation unit 402, being used for described mean value is reference point, perhaps any one point of mean value front is reference point, perhaps any one point of mean value back is reference point, interpolative data on described linearity curve, replace the data of original relevant position with the data of interpolation, obtain new signal;

Fourier transformation unit 501 is used for described new signal is carried out fast fourier transform, and time domain is become frequency domain;

Frequency domain smoothing unit 502 is used to intercept the HFS of described frequency domain, carries out forward, and perhaps dislocation addition is backward averaged, and replaces described HFS with described mean value;

Inverse fourier transform unit 503 is used for the HFS after substituting with mean value is carried out invert fast fourier transformation, obtains new signal, replaces fast fourier transform signal before with the new signal behind the Fourier transform;

Computing unit 203 is used to calculate the assessment indicator of described new signal, as, signal to noise ratio (S/N ratio), distortion spectrum degree, Mean Opinion Score;

Comparing unit 204, be used for receiving assessment indicator from computing unit 203, the size of more described assessment indicator and the assessment indicator that presets, when described assessment indicator during less than the assessment indicator that presets, described new signal is sent to drawing unit 401, up to the assessment indicator that receives more than or equal to the described assessment indicator that presets.

Wherein, drawing unit 401, linear interpolation unit 402, Fourier transformation unit 501, frequency domain smoothing unit 502 and inverse fourier transform unit 503 place linear process unit 202.

The device synoptic diagram that this embodiment provides is compared the device synoptic diagram that embodiment one provides, with linear prediction unit 301, replace unit 302, replace with drawing unit 401, linear interpolation unit 402, Fourier transformation unit 501, frequency domain smoothing unit 502 and inverse fourier transform unit 503, be used on the basis of sound signal being carried out linear interpolation, further adopt the method correction frequency spectrum of frequency domain smoothing to disperse, make the spectral smoothing of sound signal, reach the purpose of eliminating boundary effect.

Referring to Figure 10, the device synoptic diagram for the embodiment of the invention eight provides comprises:

Intercept signal unit 201 is used near the intercepting one segment signal audio signal data frame boundaries;

Fourier transformation unit 501 is used for described new signal is carried out fast fourier transform, and time domain is become frequency domain;

Frequency domain smoothing unit 502 is used to intercept the HFS of described frequency domain, carries out forward, and perhaps dislocation addition is backward averaged, and replaces described HFS with described mean value;

Inverse fourier transform unit 503 is used for the HFS after substituting with mean value is carried out invert fast fourier transformation, obtains new signal, replaces fast fourier transform signal before with the new signal behind the Fourier transform;

Computing unit 203 is used to calculate the assessment indicator of described new signal, as, signal to noise ratio (S/N ratio), distortion spectrum degree, Mean Opinion Score;

Comparing unit 204, be used for receiving assessment indicator from computing unit 203, the size of more described assessment indicator and the assessment indicator that presets, when described assessment indicator during less than the assessment indicator that presets, described new signal is sent to Fourier transformation unit 501, up to the assessment indicator that receives more than or equal to the described assessment indicator that presets.

Wherein, Fourier transformation unit 501, frequency domain smoothing unit 502 and inverse fourier transform unit 503 place linear process unit 202.

The device synoptic diagram that this embodiment provides is compared the device synoptic diagram that embodiment one provides, with linear prediction unit 301, replace unit 302, replace with Fourier transformation unit 501, frequency domain smoothing unit 502 and inverse fourier transform unit 503, being used to revise frequency spectrum disperses, make the spectral smoothing of sound signal, reach the purpose of eliminating boundary effect.

Above embodiment as can be seen, because the discontinuous boundary vicinity that occurs in adjacent data frames of sound signal, therefore near the audio signal data frame boundaries, intercept a segment signal, described signal to intercepting carries out linear process, replace original signal with the signal after the linear process, and calculating is through the assessment indicator of the new signal after the linear process, when through the assessment indicator of the new signal after the linear process during less than the assessment indicator that presets, proceed linear process, near further revising the boundary sound signal, the assessment indicator of the signal after linear process is more than or equal to the assessment indicator that presets, make the signal of adjacent data frames boundary vicinity have continuity, realized eliminating the purpose of boundary effect.

One of ordinary skill in the art will appreciate that all or part of step that realizes in the foregoing description method is to instruct relevant hardware to finish by program, described program can be stored in a kind of computer-readable recording medium.

The above-mentioned storage medium of mentioning can be a ROM (read-only memory), disk or CD etc.

More than a kind of method and device of revising sound signal provided by the present invention is described in detail, for one of ordinary skill in the art, thought according to the embodiment of the invention, part in specific embodiments and applications all can change, in sum, this description should not be construed as limitation of the present invention.

Claims (14)

1. a method of revising sound signal is characterized in that, comprising:

Near the intercepting one segment signal audio signal data frame boundaries;

Described signal to intercepting carries out linear process, obtains new signal;

Calculate the assessment indicator of described new signal, when described assessment indicator during less than the assessment indicator that presets, proceed linear process, the assessment indicator of the signal after linear process is more than or equal to the described assessment indicator that presets.

2. method according to claim 1 is characterized in that, described described signal to intercepting carries out linear process, comprising:

Signal to intercepting carries out linear prediction, near the hop value the predicted value surrogate data method frame boundaries of gained.

3. method according to claim 1 is characterized in that, described described signal to intercepting carries out linear process, comprising:

At least two data points before and after the Frame saltus step are averaged, make linearity curve with described data point and described mean value;

With mean value is reference point, and perhaps any one point of mean value front is for reference point, and perhaps any one point of mean value back is reference point, and interpolative data on described linearity curve replaces the data of original relevant position with the data of interpolation.

4. method according to claim 1 is characterized in that, described described signal to intercepting carries out linear process, comprising:

Described signal is carried out fast fourier transform, time domain is become frequency domain;

Intercept the HFS of described frequency domain, carry out forward, perhaps dislocation addition is backward averaged, and replaces described HFS with described mean value;

HFS after substituting with mean value is carried out invert fast fourier transformation, obtain new signal, replace fast fourier transform signal before with described new signal.

5. method according to claim 2 is characterized in that, described described signal to intercepting carries out linear process, comprising:

In described signal, in the past intercept a segment signal backward and be set to the odd number sign, from after intercept a segment signal forward and be set to the even number sign;

The signal of odd number sign and the signal of even number sign are carried out linear prediction, obtain the predicted value of the signal of the predicted value of signal of odd number sign and even number sign, and the predicted value of the signal of dual numbers sign is carried out back to front again;

Average to the predicted value of the signal of described odd number sign with through the predicted value of the signal of the even number sign of back to front, with near the hop value of described mean value surrogate data method frame boundaries.

6. method according to claim 1 is characterized in that, described assessment indicator comprises:

The signal to noise ratio (S/N ratio) of sound signal, distortion spectrum degree, Mean Opinion Score.

7. a device of revising sound signal is characterized in that, comprising:

The intercept signal unit is used near the intercepting one segment signal audio signal data frame boundaries;

The linear process unit is used for the signal that receives is carried out linear process, obtains new signal;

Computing unit is used to calculate the assessment indicator of described new signal;

Comparing unit, be used for receiving assessment indicator from computing unit, the size of more described assessment indicator and the assessment indicator that presets, when described assessment indicator during less than the assessment indicator that presets, described new signal is sent to the linear process unit, up to the assessment indicator that receives more than or equal to the described assessment indicator that presets.

8. device according to claim 7 is characterized in that, described linear prediction unit comprises:

Linear prediction unit is used for the signal that receives is carried out linear prediction, obtains predicted value;

Replace the unit, near the hop value the predicted value surrogate data method frame boundaries that is used for receiving obtains new signal.

9. device according to claim 8 is characterized in that, described device further comprises:

The parity flag unit is used for from intercept signal unit received signal, and intercepting a segment signal backward is set to the odd number sign in the past to described signal, and sends to described linear prediction unit, obtains the predicted value of the signal of odd number sign; From after intercept a segment signal forward and be set to the even number sign;

The back to front unit is used for carrying out back to front when described linear prediction unit receives the predicted value of signal of even number sign;

The mean value unit, be used to receive the predicted value of the signal of described odd number sign, and the predicted value of the signal of even number sign, and average to the predicted value of the signal of described odd number sign with through the predicted value of the signal of the even number sign of back to front, obtain the predicted value of intercept signal unit intercept signal.

10. device according to claim 8 is characterized in that, described device further comprises:

Drawing unit is used at least two data points before and after the Frame saltus step are averaged, and makes linearity curve with described data point and described mean value;

The linear interpolation unit, being used for described mean value is reference point, and perhaps any one point of mean value front is reference point, and perhaps any one point of mean value back is reference point, interpolative data on described linearity curve replaces the data of original relevant position with the data of interpolation.

11. device according to claim 8 is characterized in that, described device further comprises:

Fourier transformation unit is used for described signal is carried out fast fourier transform, and time domain is become frequency domain;

The frequency domain smoothing unit is used to intercept the HFS of described frequency domain, carries out forward, and perhaps dislocation addition is backward averaged, and replaces described HFS with described mean value;

The inverse fourier transform unit is used for the HFS after substituting with mean value is carried out invert fast fourier transformation, obtains new signal, replaces fast fourier transform signal before with described new signal.

12. device according to claim 7 is characterized in that, described linear process unit comprises:

Drawing unit is used at least two data points before and after the Frame saltus step are averaged, and makes linearity curve with described data point and described mean value;

The linear interpolation unit, being used for described mean value is reference point, and perhaps any one point of mean value front is reference point, and perhaps any one point of mean value back is reference point, interpolative data on described linearity curve replaces the data of original relevant position with the data of interpolation.

13. device according to claim 12 is characterized in that, described device further comprises:

Fourier transformation unit is used for described signal is carried out fast fourier transform, and time domain is become frequency domain;

The frequency domain smoothing unit is used to intercept the HFS of described frequency domain, carries out forward, and perhaps dislocation addition is backward averaged, and replaces described HFS with described mean value;

The inverse fourier transform unit is used for the HFS after substituting with mean value is carried out invert fast fourier transformation, obtains new signal, replaces fast fourier transform signal before with described new signal.

14. device according to claim 7 is characterized in that, described linear process unit comprises:

Fourier transformation unit is used for described signal is carried out fast fourier transform, and time domain is become frequency domain;

The frequency domain smoothing unit is used to intercept the HFS of described frequency domain, carries out forward, and perhaps dislocation addition is backward averaged, and replaces described HFS with described mean value;

The inverse fourier transform unit is used for the HFS after substituting with mean value is carried out invert fast fourier transformation, obtains new signal, replaces fast fourier transform signal before with described new signal.

Images