JPS62278598A - Band division type vocoder - Google Patents

Abstract

(57) [Summary] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

[Detailed description of the invention]

3. Detailed Description of the Invention [Field of Industrial Application] The present invention relates to a band division type vocoder, and particularly to a band division type vocoder of a linear predictive analysis type in which zero points are used as band division points and which improves decision accuracy. [Prior art] Linear predictive analysis
Fundamental drawbacks of LPC vocoders that utilize the Coding (hereinafter abbreviated as LPC analysis) technique, namely, the possible underestimation of the format bandwidth, especially for the first format, and the 37th allmant in comparison to the 17th allmant. In order to improve the disadvantage of poor approximation, band-splitting vocoders are well known that perform LPC analysis by dividing the input audio signal into multiple subbands in order to eliminate other concentrations in a specific frequency region. , its effects and problems have also been clarified. One of the problems is spectral discontinuity at the band division point, and as a mitigation measure, a method of matching the band division point with the zero point has been proposed, and this is described in Japanese Patent Laid-Open No. 59-5
297. This is explained in detail in the band division type vocoder and the like. [Problems to be Solved by the Invention] The above-mentioned conventional band-splitting vocoder of this type is influenced to some extent by the bi and chis vector structure in its zero point estimation, and accurate estimation is not always possible. Not there,
Therefore, the effect of band division through the painstakingly introduced zero point is not fully exhibited. Another drawback is that the specific method of zero point estimation is not clarified. SUMMARY OF THE INVENTION An object of the present invention is to eliminate the above-mentioned drawbacks, remove the bi, tis vector structure from an input audio signal, and provide a means for accurately estimating the zero point of the spectrum so that the specified zero point becomes a band division point. The object of the present invention is to provide a band-splitting vocoder in which the band-splitting effect is significantly improved. cepstrum analysis means for separating and extracting cepstrum information regarding an input speech signal, pitch search means, and pix vector removal for determining the spectral envelope of the input speech signal based on a low-que frency component obtained by removing the bi, chis vector structure from the cepstrum. means for estimating zero points of the input audio signal based on two types of polar information detected by linear predictive analysis of the spectrum envelope and its inverse spectrum, respectively; and analysis and synthesis means for analyzing and synthesizing. [Example] Next, the present invention will be described in detail with reference to the drawings. FIG. 1 is a block diagram showing one embodiment of the present invention. The block diagram shown in FIG. 1 is composed of an analysis side 1 and a synthesis 4A3 that transmits and receives via a transmission line 2. The analysis side 1 includes an A/D converter 11. Cepstral analyzer 12. 7
Ta13. Pitch extractor 14, encoder/multiplexer]5. Spectrum calculator 16, zero estimator]7. Pole estimation unit 18. Autocorrelation calculator 191 division frequency determiner 20. Band division L
It is configured with a PC analyzer 21 and the like. Furthermore, the combining side 3 includes a decoding/demultiplexing unit 31 . t scum vector calculator 32
．． Predicted residual power output device 33. Pi, Chi pulse train generator 34
．． Switcher 35. Autocorrelation nose device 36, LPC analyzer 37
．． Noise generator 38. Variable gain amplifier 39. LPC synthesis filter 40. When the 0-person-powered audio signal, which is configured with a D-1 converter 41 and the like, is input to the A/D converter 11, it is filtered by an LPF (Law Pa5s Filter) with 3.4kHz as the high-frequency S-fraction wave number. Later, the data is converted into an original at a sampling frequency of 8 kHz, and further converted into a doji using a predetermined number of bits. This Doji-formed audio signal is stored in the internal memory ljK in 30m5EC, 240 samples each as a window time, and is successively subjected to window processing, which is multiplied by a predetermined window function and a Hamming function, and cut out at a cycle of 10m5EC. The output of is supplied to the cepstral analyzer 12. The execution period of this window processing, 10 m5EC, is the analysis frame period.The cepstrum analyzer I2 analyzes and calculates the cepstrum of the filtered audio signal for each input analysis frame as follows.The cepstrum is a short waveform Time amplitude spectrum lX[n
It is well known that this method is defined as the inverse 7-Lier transform of the logarithm of l, and has the characteristic of being able to approximately separate and extract the spectral envelope of the input audio signal and the fine structure, that is, the sound source information. where ω indicates the angular frequency 0. Let x(t) be the input audio signal (voiced sound), and g[t]
If we let h(t) be the sound source and h(t) be the impulse response of the μ-way filter, the following equation (1) holds true.0 Here, t indicates time.0 In equation (1), X(ω), G((b), and H(ω)
are the Fourier transforms of x(t), gft) and h(t), respectively. If Og(t) has periodicity, then When viewed on the frequency axis, 0 is expressed as equally spaced sharp peaks.If you take the logarithm of -11, it becomes the following equation (2).log lXfω)l=log IQ(nl+1og
IH ('41... (2) The cepstrum is obtained by inverse Fourier transform of equation (2) with ω as a variable, and the cepstrum obtained by inverse Fourier transform of IoglX(ml is log l G(
→The sum of 1 and log l Hf-1 converted to inverse Fourier

[expressed. Thus, the cepstrum obtained through the inverse transformation of the logarithm of the spectrum, Is/, is called Keffle/C in time domain data. In the example shown in FIG. 1, these Fourier transforms and inverse Fourier f transforms are performed based on discrete data processing. In Equation (2), the first term on the right side represents the fine structure of the spectrum, and the second term represents the spectral envelope. Therefore, there is a big difference between the inverse Fourier transforms of these two, with the former being a peak that is concentrated in a part of the high quefrency, and the latter being concentrated in a part of the low quefrinous area. By analyzing this, the fundamental period of the sound source (gft) can be found, and by Fourier transformation of a portion of the low frequency spectrum, a logarithmic spectral envelope can be obtained, and further, by exponential transformation of this, a spectral envelope can be obtained. The process is carried out using a lid (l1fter) to separate only a portion of the high-cephalic cepstrum from the cepstrum and take out the low-cephalic. In the embodiment of FIG. 1, the cepstrum analyzer 12 calculates the cepstrum of the quantized audio signal for each analysis frame and supplies this to a lifter 13 and a pitch extractor 4. The lifter 13 removes pitch components using a lifter that can be set based on pitch period information supplied from a pitch extractor 14, which will be described later, and supplies the removed pitch component to the spectrum calculator 16. The spectrum calculator 16 performs a discrete Fourier function on the input and then performs exponential transformation to obtain the spectrum envelope of the input audio signal, and supplies this to the zero estimator 17, the pole estimator 18, and the autocorrelation calculator 19. do. The spectral envelope thus obtained from the spectrum calculator 6 is a spectral envelope component obtained by removing the pitch spectral structure from the input audio signal. The pitch extractor 14 extracts pitch information and V (voiced) based on the cepstrum received from the cepstrum analyzer 12.
Information regarding Uv (unvoiced) K is searched using known techniques, and the information is supplied to the encoder/multiplexer 15. Furthermore, the pitch extractor 15 supplies the pitch period information to the input audio signal as described above.Now, the spectrum envelope output from the spectrum calculator 16 eliminates the pitch influence on the input audio signal, and It's as I said before. Therefore, it has a different number of bandwidth poles and zeros for each analysis frame. In retrospect, the problem with band-splitting vocoders that use zeros as band-splitting points to alleviate spectral discontinuities at band-splitting points is also the problem with accurate processing of zero-point estimation. Get 1. Therefore, in this embodiment, a fundamental solution to this problem is provided by the zero estimation section]7. This is implemented by the pole estimator 18 and the division frequency determiner 20 as follows. In this embodiment, the number of divided bands is 2 channels, but this number is usually considered to be the most efficient based on the trade-off between the effect of one division and the increase in processing capacity, as well as actual results. Of course, this number may be changed as appropriate in consideration of the purpose of use of the vocoder. The zero estimator 17 includes an inverse spectrum calculator 171. LPC analyzer 172. It is configured with a pole calculator 173 and a pole saver 174, and calculates seven poles for the inverse spectrum of the input spectrum envelope. Perform the test. In other words, by inverting the spectral envelope, what is originally a zero point is transformed into an upper pole,
A well-known pole calculation method is applied to this to perform qualitative zero estimation. The inverse spectrum calculator 171 calculates the inverse spectrum of the spectral envelope of each analysis frame manually, and converts this into L
It is supplied to the PC9vr device 172. [LPC analyzer] 72 performs KLI'C analysis of the input inverse spectrum for each analysis frame to obtain a predetermined order. The P-order α parameter is extracted and supplied to the pole calculator 173. The pole calculator 73 calculates the pole frequency and bandwidth using the following known method based on the input P-order α parameter. In other words, the following (3) with the P-th order α parameter as a coefficient
The polar frequency is calculated as a conjugate complex solution obtained as the root of the higher-order equation shown in the equation to zero. 1 + aI Z-'+a2 Z-2+ ------
-+apZ-p=0...(3) (3) Subtracting equation K, we get Z=eJ'', and the left side is the number of transmission rll for gjmofz, H(Z-')= ] /A
It is also well known that it represents the denominator of the right-hand side of p(Z'). In general, from equation (3), a conjugate complex root of P/2 or less is obtained corresponding to an even or complete number of P, and this provides the polar circumference V number, other than the conjugate complex root. All roots of 0 are real roots. Furthermore, the bandwidth of the pole frequency obtained in this way can be expressed by the following equation (4). B i = -logγi/l・ΔT ・・・・・・
...... f4) In equation (4), ΔT is the sampling period,
γi represents the root of equation (3) in polar coordinates, and Zi−=rieJ”
λi is the wavelength corresponding to the polar frequency. The other tester 174 receives the calculation result of the other calculator 173, makes a final determination of the pole based on the test standard set in advance based on audio materials, etc., and supplies the information to the dividing tube e and the number determiner 20. However, as described above, both the pole frequency and the bandwidth verified in this way actually designate the zero point. The zero pole estimator 18 that performs zero estimation in this way includes an LPC analyzer 18], a pole calculator 182. It is configured with a polar tester 183, estimates the polar frequency and bandwidth using the zero estimator 17 and the four-dimensional method for the spectrum itself of each input analysis frame, and supplies the data to the divided frequency 20.0 divided frequency data 20 receives the thus supplied test data regarding the zero estimation and other estimation for each analysis frame,
The zero point is comprehensively and accurately estimated by comparison with preset judgment criteria based on sound F data, operational results, etc., and the dividing frequency to be the two dividing points is determined, and the data is supplied to the autocorrelation calculator 9. At the same time, it is supplied to the encoder/multiplexer 15. The autocorrelation calculator i:lli unit 19 is the spectrum calculator 16
The spectral envelope received from the spectral envelope is divided into low band (L) and high band () () by the division frequency, and the autocorrelation coefficient in each band is calculated over a predetermined time delay range, and this is calculated by the pre-region division LPC. At the same time, the autocorrelation coefficient at zero delay time is calculated for each band, and this is supplied to the encoder/multiplexer 5 as the average power. [Band division L'PC2] inputs or uses the autocorrelation coefficient of the H2 region, extracts parameters as LPC coefficients for each of the two bands by a known method, and sends them to the encoder/multiplexer. 15. The encoder/multiplexer 15 encodes each data regarding the pitch, V/UV, divided IkJ wave number, average power, and 'LPC coefficient supplied through each analysis process described above in a predetermined format. These are appropriately combined and multiplexed, and transmitted to the combining side 3 via the transmission line 2. On the combining side 3, a decoding/multiplexer 31 performs multiplexing, demultiplexing and decoding of the received signal, LPC constant data is sent to an electric waste vector calculator 32, and average power data is sent to an electric waste vector calculator 32 for prediction. The divided frequency data is supplied to the residual power output device 33, the divided frequency data is supplied to the electric scum vector calculator 32, the pitch data is supplied to the pitch pulse generator 34, and the V/UV data is supplied to the switch 35. The electric waste vector calculator 32 calculates the normalized predicted residual power U for both the low and high bands based on the following equation (5) using the parameters input as the LPC coefficients. (5) In equation (5), P represents a parameter for the analysis order. The power calculator uses the normalized predicted residual power and the average power to obtain predicted residual power Pa for both the low and high bands. The electric waste vector P(→(S) can be obtained from the following equation (6) using the predicted residual power and the α parameter that can be easily derived from the K parameter of the LPC coefficient. This two-band electric cassette vector is then provided to an autocorrelation coefficient 36, subjected to an inverse Fourier transform, etc., and then supplied to an LPG analyzer 37 as an autocorrelation coefficient sequence corresponding to all frequency bands.LPC analysis The device 37 extracts LPC coefficients of a predetermined order using the input autocorrelation coefficient sequence and outputs them as filter coefficients of an LPC synthesis filter 40 configured as an all-pole digital filter. 37 supplies the normalized predicted residual power obtained along with the LPC coefficient extraction to the predicted residual power calculator 33. The predicted residual power calculator 33 is also supplied with power data for each analysis frame, The predicted residual power calculator 33 outputs the predicted residual power to the variable gain amplifier 39 for each analysis frame based on the power of these two people, and changes the gain corresponding to the predicted residual power.The pitch data is generated by pitch pulse train generation. is supplied to the vessel 34,
A pitch pulse train having a period corresponding to the pitch data is generated. Further, the after-sound generator 38 generates white noise, and the divided frequency is provided as boundary frequency information when calculating predicted residual power for both the low and high bands, and is used to maintain continuity of both bands. used. The electric flux vectors in both the low and high bands are supplied to the autocorrelation calculator 36 and subjected to inverse Fourier KIs, and autocorrelation coefficients corresponding to all frequency bands are extracted and provided to the LPC analyzer 37. . The LPC analyzer 37 uses the autocorrelation coefficients of all frequency bands to extract parameters of a predetermined order as LPC coefficients for each analysis frame, and constructs the LPC synthesis filter 40 as an all-pole digital filter. The 0LPC analyzer 37 also uses (5
) is used to calculate the normalized prediction residual and supplies it to the prediction residual power calculator 33. The predicted residual power calculator 33 is also supplied with the average power of the low frequency band and the high frequency band. This average power is added and converted to the average power of the entire frequency band f, and by combining this with the normalized predicted residual power, the predicted residual power is calculated for each analysis frame and provided to the variable gain amplifier 39, and its gain is is changed corresponding to the predicted residual power for each analysis frame, that is, the sound m, and the amplitude. Now, the pitch data is provided to the pitch pulse train generator 34, which generates a pitch pulse train corresponding to the pitch data. This pitch pulse train is supplied to the switch 35. The switch 35 is also supplied with white noise output from the frequency generator 38 . The switch 35 further inputs V/UV data, and this
In other words, when specifying voicing, the pitch pulse train generator 3
The output of the noise generator 38 is outputted to the variable gain amplifier 39, and when specifying Uv, that is, no μ, the output of the noise generator 38 is switched to be outputted to the variable gain amplifier 39. The variable gain amplifier 39 generates the modeled sound source information, which is then used to synthesize the LPG synthesis filter 54. The LPC synthesis filter 40 driven in this way reproduces the input audio signal in digital format, and then converts it to the D/A converter. At step 41, the signal is converted into analog data, and then sent out as an output audio signal with unnecessary high frequencies removed through an LPF. The band division type vocoder which uses the zero point obtained based on accurate estimation as the division point enables analysis and synthesis that essentially eliminates spectral discontinuity at the band division point. [Effects of the Invention] As explained above, according to the present invention, in addition to removing the pi, chis vector structure from the input audio signal, a means for accurately estimating the zero point of the spectrum is provided, and the specified zero point is set as the band division point. By doing so, it is possible to realize a band division type vocoder which can significantly improve the band division effect.

[Brief explanation of drawings]

Figure 1 is a diagram of one embodiment of the present invention. 1...jf-analysis, 2...transmission line, 3.
・・・Synthesis side,】】・・・A/D converter, 1
2...Cefstrum minute v'r device, ]3...
・S7ri, 14... Pitch extractor, ]5...
...encoder/multiplexer,]6...spectrum! Output device, ]7...Zero estimation section, ]8...
Other estimators, ]9... Autocorrelation calculator, 20...
... Division frequency determiner, 2] ...Band division L
PG analyzer, 31...Decoding/demultiplexer, 3
2... Electric waste vector calculator, 33...
- Prediction residual power calculator, 34...Pitch pulse train generator, 35...Switcher, 36...
Autocorrelation calculator, 37... LPC analyzer, 38
...Noise generator, 39...Variable gain amplifier, 40...LPC synthesis filter, 41...
...D/A conoverter, 171 ... Inverse spectrum calculator, 172 ... LPC analyzer, 17
3...Pole calculator, ]74...Other species determiner, 181...LPC analyzer, ]82...
・Koku Calculator, 183...Koku Verifier 0 Agent Patent Attorney Susumu Uchihara -1 Procedural Amendment (Spontaneous) 62.8.10 Showa Year/Month Date Chief Palace of the Licensing Agency 2, Name of the invention Band-splitting vocoder 3, relationship with the amended person case Applicant: 5-33-1 Shiba, Minato-ku, Tokyo (423) NEC Corporation Representative: Tadahiro Sekimoto 4, agent & of the specification subject to amendment "Detailed Description of the Invention" Column & Contents of the Amendment In page 9, line 13, K "excluded and is an envelope" is corrected to "excluded spectral envelope."

Claims (1)

[Claims] A band division vocoder based on a linear predictive analysis format, comprising:
) a cepstrum analysis means for analyzing and extracting information; a pitch search means; a pitch spectrum removal means for determining the spectral envelope of the input audio signal based on the low frequency component obtained by removing the pitch spectral structure from the cepstrum; Zero point estimating means for estimating the zero point of the input audio signal based on two types of polar information detected by linear predictive analysis of the inverse spectra, and analysis and synthesis for analyzing and synthesizing the input audio signal for each band divided by the zero points. A band division type vocoder comprising: means.