JP2001117595A - Audio signal waveform processor - Google Patents

Abstract

PROBLEM TO BE SOLVED: To allow an audio signal waveform processor, which performs a time stretching and pitch shifting by a phase vocoder system, to prevent a noise due to a pre-echo generated by band division from being generated. SOLUTION: The audio signal waveform processor has, as waveform data, the amplitudes and phases of respective band components obtained by dividing an audio signal into the bands, restores an audio signal waveform by reproducing and putting together the band components of the respective bands while performing time compression or expansion according to the waveform data, and performs phase resetting in the beginning of a pre-echo section generated in the audio signal waveform by the band division so that respective restored band components are in phase with the respective original band components shown by the above waveform data.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an audio signal waveform processing apparatus for performing time stretching (time compression / expansion) and pitch shift (pitch conversion) by a phase vocoder method.

In a phase vocoder system, an analysis system divides an audio signal waveform of an original sound into a plurality of frequency bands (bands) using a band filter, analyzes the band components of each band, and outputs the output amplitude and phase. Is extracted and held as a feature parameter, and in the synthesis system, the original band components are reproduced using the output amplitude and phase for each band, and the band components of each band are added and synthesized.
Restore the original audio signal waveform.

FIG. 10 explains the concept of an audio signal waveform processing apparatus using the phase vocoder method. As shown, the audio signal waveform X (n) is input to a plurality of analyzers 60. In this example, the analysis unit 60 is provided for each band obtained by dividing the frequency of the audio signal waveform into 100 bands. As shown in FIG. 11, each analysis unit 60 has bands 0 to 99 each having a center frequency substantially equal to the basic frequency of the audio signal waveform, and has a configuration as shown in FIG. That is, for example, the analysis unit of band k converts the input audio signal waveform X (n) into a complex frequency sin at the center thereof.
(ωk) and cos (ωk) are multiplied (synchronous detection) to obtain the output amplitude value of band k, and the instantaneous frequency information is obtained by differentiating the phase value of the detection output. The instantaneous frequency is an amount of change (differential value) of the phase per unit time at each time point (each position on the time axis of the waveform), and is information indicating a frequency deviation from the center frequency.

In the waveform processing apparatus shown in FIG. 10, waveform data (output amplitude and instantaneous frequency) of each band of the audio signal waveform X (n) is stored in a waveform memory 61. As shown in FIG. 12, the waveform data is stored in the waveform memory 61 at each address on the time axis of the audio signal waveform x (n).
For addr (0) to addr (n), for each band 0 to 99,
The amplitude data A and the instantaneous frequency data f are stored.

The synthesizing system comprises a band component reproducing section 62 provided for each band. Each band component reproducing section 62 includes a time-frequency conversion processing section 31, a cosine oscillator 33, and a multiplier 33.
Consists of The time-frequency conversion processing unit 31 has a configuration as shown in FIG. That is, with respect to the input output amplitude value, the sampling point is skipped / additionally interpolated according to the time stretch ratio (time compression / expansion ratio), and its amplitude envelope (a temporal change of the amplitude value is obtained). And outputs an amplitude value obtained by compressing / expanding the envelope shown in FIG. Also, for the input instantaneous frequency value, the center angular frequency ωk is added to the instantaneous frequency value, and when pitch conversion is performed, the instantaneous frequency value is added to the frequency conversion ratio (depending on the degree of pitch shift). The instantaneous frequency value obtained by compressing / expanding the frequency envelope (envelope indicating the temporal change of the deemed frequency value) by interpolating / additionally interpolating the sample points by the interpolation unit according to the time stretch ratio Is output.

FIG. 15 is a diagram showing how the amplitude value and the instantaneous frequency are interpolated. In case of time extension,
As shown in FIG. 5A, the original amplitude envelope and the frequency envelope are both stretched to generate the amplitude value and the instantaneous frequency with the time axis expanded. In the case of time compression, as shown in FIG. 15B, the original amplitude envelope and the frequency envelope are both contracted to generate an amplitude value and an instantaneous frequency obtained by compressing the time axis. By this interpolation processing, the time axis of the original audio signal waveform can be arbitrarily compressed / expanded.

The instantaneous frequency value (time-stretched as appropriate) processed by the time-frequency conversion processing unit 31 is supplied to a cosine oscillator, whereby the cosine oscillator generates a cosine wave of the frequency of the band, and generates the cosine wave. The wave is added with the amplitude envelope processed by the time-frequency conversion processing unit 31 and output. Thereby, the component signal of the band is reproduced. Further, the original audio signal waveform can be restored by adding and combining the band components of the respective bands 0 to 99.

[0008]

In reproducing the audio signal waveform of a musical tone by the above-described phase vocoder method, the level of the audio signal waveform of a piano or the like greatly changes with time in the attack portion. When such an audio signal waveform is band-divided using a digital filter, a trace, a so-called "pre-echo", appears before each attack in each band.

FIG. 9 illustrates this pre-echo phenomenon. For simplicity, FIG. 9 shows a case where the original audio signal waveform is divided into a high frequency band and a low frequency band. A typical attack waveform (a) is separated into a low-frequency component waveform (b) and a high-frequency component waveform (c), and the appearance of a pre-echo is shown. In the low-frequency component waveform (b) and the high-frequency component waveform (c), it can be seen that a pre-echo occurs in a silent portion preceding the original attack waveform (a). The pre-echo of the low-frequency component waveform (b) and the pre-echo of the high-frequency component waveform (c) are in opposite phases to each other. Therefore, if the low-frequency component waveform (b) and the high-frequency component waveform are finally combined with the same phase, The two pre-echoes cancel each other out, and the silent part is restored. This is the same when the number of bands (bands) to be divided is large. If the component waveforms of each band are added and synthesized after reproduction, the pre-echoes cancel each other out to form a silent portion.

[0010] Such a pre-echo becomes more serious as the number of taps of a digital filter (FIR filter) used increases, and its generation limit becomes the number of taps of the filter itself. Can be estimated.

When time stretching or pitch shifting is performed using the phase vocoder method, the waveform of each band generally does not return to the phase of the original waveform, but changes. For this reason, even if the band components of each band are reproduced and added and synthesized, the pre-echo portions of the waveform of each band also change in phase, so that these pre-echo portions do not cancel each other out. The pre-echo preceding the attack remains in the combined signal.

For this reason, usually, in waveform reproduction using a phase vocoder, noise due to pre-echo occurs immediately before an attack, and the attack sound is not sharp and the so-called attack feeling is significantly impaired. .

The present invention has been made in view of the above problems, and has as its object to prevent the generation of noise due to pre-echo caused by band division while using a phase vocoder method.

[0014]

An audio signal waveform processing apparatus to which the present invention is applied has amplitude and phase of each band component obtained by dividing an audio signal into a plurality of bands as waveform data. The audio signal waveform is restored by reproducing and synthesizing the band components of each band while compressing / expanding them as necessary based on the basis. In this audio signal waveform processing apparatus, in order to solve the above-mentioned problem, according to the present invention, each band to be restored at the beginning of the pre-echo section with respect to the pre-echo section generated in the audio signal waveform by band division. The phase of the component is reset so that it becomes the phase of the original band component corresponding to each of the waveform data. In this pre-echo section, it is desirable not to perform time compression / expansion of the restored waveform. In this way, the phase is reset at the point when the pre-echo occurs, and the phase of the reproduced waveform of each band at that point is set to the same as that of the original sound. The pre-echo is canceled in the audio signal waveform.

In the above-mentioned phase reset, it is more preferable that the phase of the waveform reproduced so far and the value of the phase to be replaced based on the waveform data be replaced by a cross fade. This way,
Since the phase of the reproduced waveform changes continuously at the time of phase reset, it is possible to prevent the generation of noise due to the discontinuous change in phase caused by suddenly changing the phase value.

When the waveform conversion is accompanied by pitch conversion, it is desirable that the degree of the phase change based on the pitch conversion and the step of the read address for reading the waveform data be matched. As a result, the phase of the pitch-converted waveform becomes the same as the original waveform, the original phase is reproduced in the reproduced waveform of the band component, and the pre-echo can be canceled by combining the band components.

[0017]

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 shows an audio signal waveform processing apparatus as one embodiment of the present invention. In this embodiment, the present invention is applied to a keyboard-type electronic musical instrument. In the drawing, reference numeral 2 denotes a CPU (central processing unit) for controlling the entire apparatus, 3 denotes a ROM (read only memory) for storing various table data and control programs for the CPU, and 4 denotes a plurality of audio signal waveforms. RAM for storing waveform data and used as working memory for CPU
(Random access memory). Reference numeral 5 denotes an operator group composed of various operators, such as a selection button for selecting a reproduction waveform from various waveforms, a flexibility to be described later, and instructions for starting and ending waveform reproduction. There are various controls. Reference numeral 6 denotes a keyboard device having a lever 61 on its panel for adjusting the reproduction speed. Reference numeral 1 denotes a waveform synthesizing unit, which comprises a waveform memory 10 and band component reproducing units 11-0 to 11-99 provided corresponding to the respective bands, and based on waveform data selected for reproduction, an original audio signal. Restore the waveform and output.

The waveform memory 10 of the waveform synthesizing section 1 has an area for storing waveform data of an audio signal waveform and an area for storing reset address information.

The waveform data of the audio signal waveform includes an output amplitude value and a phase value. The output amplitude value is the same as that described in the prior art. The phase value replaces the instantaneous frequency value of the prior art described above, and is an original characteristic parameter of the phase vocoder. In the present embodiment, the waveform data (amplitude value and phase value) of a plurality of types of audio signal waveforms are determined in advance by analysis and stored in the RAM 4. Then, the waveform data of the audio signal waveform selected for reproduction by the selection button for waveform selection is stored in the waveform data area of the waveform memory 10.

FIG. 2 shows a conceptual configuration of an analyzer for analyzing the audio signal waveform and extracting the waveform data of the output amplitude value and the phase value of the waveform of each band. As shown in the figure, the audio signal waveform X (n) is multiplied by complex frequencies cos (ωk, n) and sin (ωk, n) at the center of the band, respectively, and passed through an analysis filter of an impulse response W (n). The output is squared and then added to calculate the square root to obtain an amplitude value. Further, the outputs Xcos and Xsin of the analysis filter are passed through a calculation unit to obtain a phase value. The arithmetic unit calculates Arctan (Xsin / Xcos) when Xcos> 0 and calculates Arctan (Xsin / Xcos) + π when Xcos <0, according to the outputs Xcos and Xsin of the analysis filter, and outputs the result as a phase value. .

The reset address information area of the selected audio signal waveform (phase reset information) is stored in the reset address information area of the waveform memory 10. In this reset address information, an address at a position before the address indicating the attack of the audio signal waveform by an amount corresponding to the occurrence of the pre-echo is set, and is set for each attack waveform existing in the audio signal waveform in time series. .. Are sequentially assigned as RstAdr (0), RstAdr (1), RstAdr (2),... And are arranged in order of their size. It should be noted that how much the pre-echo precedes the attack can be estimated from the delay amount of the filter that divides the band.

In this embodiment, the reset address information is analyzed and obtained in advance for a plurality of types of audio signal waveforms and stored in the RAM 4. And
The reset address information of the audio signal waveform selected for reproduction by the selection button for waveform selection is stored in the waveform memory 10.
Is stored in the reset address information area.

The amplitude and phase values from the waveform memory 10 and the reset address are input to the band component reproducing units 11-0 to 11-99, and the time position information (read address DataAdr) and the pitch are input from the CPU 2 side. Information is input, and the band component of the corresponding band is reproduced and output based on these data. Each band component reproducing unit 11-0 ~
The band components output from 11-99 are added and synthesized with each other, restored as an audio signal waveform, and output.

FIG. 3 shows an example of the configuration of the band component reproducing unit 11. As shown in the figure, the phase value read from the waveform memory 10 is branched into two, one of which is directly connected to the cosine oscillator 3.
2 and the other is input to the differentiator 30. The phase value input to the differentiating unit 30 is converted into an instantaneous frequency by being differentiated, and input to the time-frequency conversion processing unit 31. The amplitude value read from the waveform memory 10 is input to the time-frequency conversion processing unit 31. The time-frequency conversion processing unit 31 has the same configuration as that described in the section of the prior art, and, like the conventional configuration, multiplies the amplitude value by the multiplier 3.
3, the value of the instantaneous frequency is sent to the cosine oscillator 32, respectively.

The reset address information read from the waveform memory 10 is input to the phase reset signal generator 34. A read address DataAdr (time position information) for reading waveform data from the waveform memory 10 is input to the phase reset signal generator 34. The phase reset signal generator 34 includes a reset address RstAdr read from the waveform memory 10 and a read address DataAdr.
When the read address DataAdr exceeds the reset address RstAdr, a phase reset signal is generated and sent to the cosine oscillator 32.

FIG. 4 shows the phase reset signal generator 34.
2 shows a flowchart of a phase reset signal transmission procedure. As shown, when the read address DataAdr is updated with the reproduction of the waveform (step S1), it is determined whether the read address has exceeded the reset address RstAdr (k), that is, whether RstAdr (k) <DataAdr ( Step S2) If not exceeded, terminate the processing and wait until the next read address is updated. If exceeded, update the parameter k indicating the order of the reset address, that is, k = k + 1, and execute the phase reset signal. Is generated and transmitted (step S3).

FIG. 5 shows an example of the configuration of the cosine oscillator 32. The phase value data from the waveform memory 10 is input to the adder 42, where the rotation (angular frequency) ωk of the center frequency of the block is added, and the resultant value is added to the adder 40.
Is input to An instantaneous frequency value from the time-frequency conversion processing unit 31 is input to the adder 40, and the sum is converted into an instantaneous phase value by summing and supplied to the cosine calculator 41. Then, a cosine wave having a frequency regarded as the frequency of the component signal of the block is generated and output. A phase reset signal is input to the adder, and when the phase reset signal arrives, the phase of the instantaneous frequency is replaced with the absolute phase from the adder 42.

Next, in this embodiment, in order to perform time stretching (time compression / expansion) of the audio signal waveform, the time compression / expansion rate and flexibility are used.
These will be described. The compression / expansion rate indicates the amount of time compression / expansion with respect to the audio signal waveform by the ratio of the time length before and after compression / expansion. On the other hand, the flexibility indicates the degree of correction of how to correct the compression / decompression rate in each section of the audio signal waveform.

[Flexibility] For example, as shown in FIG. 6, a musical tone waveform is temporally divided into a plurality of sections (the total number of sections is m).
And the flexibility E (i) is set for each of these sections. This section is desirably set to coincide with, for example, a pre-echo section, an attack section, a decay section, a sustain section, a release section, a silent section, and the like.

When the flexibility E (i) is "0", the reproduction speed of the section is maintained at the original waveform regardless of the setting state of the lever 61 and the value of the compression / expansion rate. Will be. When the flexibility is “1”, the reproduction speed is set by the lever 61. When the flexibility is smaller than “1”, the playback speed is lower than the playback speed set by the lever 61. When the flexibility is larger than “1”, the playback speed is higher than the playback speed set by the lever 61. Become.

The flexibility of each section is set such that the sum of the product of the length of each section of the musical tone waveform and the flexibility in that section becomes the length of all sections. This is represented by a mathematical formula: ΣE (i) × L (i) = L (i) (1) where Σ is the addition from i = 0 to (m−1). The length E (i) is the degree of flexibility m in each section m is the total number of sections i is a section counter indicating the section being reproduced among the sections from 0 to m. Therefore, even if the compression / expansion rate of each section is corrected by the flexibility, the compression or decompression time of the entire waveform data is determined by multiplying the compression / expansion rate by the time of the original waveform data, that is, the entire waveform data. This is simply the time when compression / expansion is performed based on the compression / expansion rate.

As will be described later, in the present invention, the processing according to the present invention is made possible by setting the degree of flexibility E (i) in the pre-echo section to "0".

[Compression / Expansion Rate] The compression / expansion rate is a coefficient for compressing or expanding the time length of the waveform to be reproduced as compared with the time length of the original waveform. That is, in this embodiment, the waveform reproduction speed can be varied by operating the lever 61 provided on the keyboard 6. The playback speed of the lever 61 is the same as that of the original waveform at the midpoint position, which is the automatic return point, and the playback speed can be increased or decreased by swinging back and forth. This reproduction speed has a reciprocal relationship to the compression / expansion rate. For example, if a certain waveform is played back at twice the playback speed, the time required for the playback will be 1/2 times.
That is, the compression / decompression rate is 1/2. When this waveform is reproduced at a reproduction speed of l / 2 times, the time required for reproduction is doubled. Similarly, this also indicates that the compression / expansion rate is 2.

The above-mentioned reproduction speed is determined by the step amount (tcomp, tcomp ',
tcomp "). The step amount is determined by a time position (read address DataAd) for reproducing the audio signal waveform.
r) is the amount by which step is advanced, in other words, the speed of waveform reproduction, that is, the reciprocal of the time expansion / compression ratio. This step amount is set by operating the lever 61.

The step amount includes a basic step amount tcomp which is a basic value determined by the setting position of the lever 61, and a flexibility-corrected step amount obtained by correcting the basic step amount tcomp in consideration of the flexibility.
There is a pitch correction step amount tcomp "obtained by further correcting the flexibility correction step amount tcomp 'in consideration of the pitch conversion.

First, the basic step amount tcomp is specifically a value that changes when the lever 61 is at the midpoint position and changes between, for example, “+2” and “0” when the lever 61 is tilted back and forth. is there.
When the read address DataAdr is updated with the basic step amount tcomp, if the basic step amount tcomp is “1”, the audio signal waveform is reproduced with the original time length, and the basic step amount tc
If omp is "2", the audio signal waveform is reproduced in half of the original time length, and if the basic step amount tcomp is "0", the audio signal waveform is repeatedly reproduced at the same position.

The flexibility correction step amount tcomp 'is obtained by correcting the basic step amount tcomp by the following equation based on the flexibility E (i) of the corresponding section. tcomp ′ = 1 / [E (i) × ｛(1 / tcomp) −1｝ +1] Here, assuming that the flexibility E (i) is 1, the flexibility correction step amount tc
omp 'becomes the basic step amount tcomp, and becomes the reproduction speed set by the lever 61. When the flexibility E (i) is made smaller than 1, the value of the flexibility correction step amount tcomp ′ becomes the basic step amount tcom.
p and becomes lower than the reproduction speed set by the lever 61. When the flexibility E (i) is larger than 1, the value of the flexibility correction step amount tcomp ′ becomes the basic step amount.
It becomes larger than tcomp and becomes higher than the reproduction speed set by the lever 61. When the flexibility E (i) is set to "0", the flexibility correction step amount tcomp 'becomes "1", which is the reproduction distance of the original waveform. Therefore, the section M (i) can be reproduced at the reproduction speed of the original waveform even though the compression / expansion is performed by setting the flexibility E (i) to "0". In this case, even if the value of the basic step amount tcomp set by the lever 61 is changed during playback, the flexibility-corrected step amount tcomp 'maintains "1". Thus, the CPU 2 corrects the basic step amount tcomp.

Although the pitch correction step amount tcomp "will be described later in detail, the flexibility correction step amount tcomp 'is further corrected, and is obtained by tcomp" = W × tcomp'. Here, W is the degree of correction, and W = (L / tcomp−L0) / (L / tcomp−L0 / P). Here, “L0” is a section of flexibility 0 starting from the reset address, and “L” is a length from the reset address of interest to the next reset address.

[Generation of Read Address DataAdr] When time compression / expansion and pitch shift are not performed, the read address DataAdr is generated by being advanced in the address order of the audio signal waveform. The sampling speed of the audio signal waveform data is 44. If it is 1 kHz, it becomes 44.1 kHz. When the reproduction speed is changed by the lever 61, the read address DataAdr is incremented by the basic increment tcomp determined by the lever 61. For example, if the basic step amount tcomp is “2”, the read address DataAdr skips every other sampling point (address Addr) of the original audio signal waveform.
Is generated (ie, time compressed). For example, if the basic step amount tcomp is "1/2", the read address DataAdr is generated (that is, time-expanded) even at an intermediate point between the sampling points of the original audio signal waveform.

The operation of this embodiment will now be described.
As described above, for a plurality of audio signal waveforms,
The head of each pre-echo portion in each audio signal waveform is preset as a reset address, and these reset addresses are stored in the RAM 4 in advance together with the waveform data of the corresponding audio signal waveform.
For these audio signal waveforms, information on the flexibility E (i) is also associated with each audio signal waveform.
It is stored in M4. In this embodiment, “0” is set as the flexibility E (i) for each pre-echo section.

When an arbitrary audio signal waveform is selected using the waveform selection button in the operator group 5, the waveform data of the audio signal waveform and the reset address information are transferred to and stored in the waveform memory 10 of the waveform synthesizing section 1. You. The flexibility information is referred to by the CPU 2 in order to create the read address DataAdr.

When the start of waveform reproduction is instructed, the CPU 2 refers to the setting state of the lever 61 at that time, the flexibility information of the waveform reproduction position, the pitch information, and the like, and reads out the read address Data.
Adr is generated, and this is used as a band component reproducing unit 11-0 to 11-99.
As time position information.

In the band component reproducing units 11-0 to 11-99,
The waveform data of each band is read from the waveform memory 10 according to the read address DataAdr, and the band components are reproduced by the band component reproducing units 11-0 to 11-99. The processing in the time-frequency conversion processing unit 31 in the band component reproduction unit 11 converts the phase value read from the waveform memory 10 into an instantaneous frequency by differentiating the phase value in the differentiation unit 30 and sends the instantaneous frequency to the time-frequency conversion processing unit 31. Except for this point, the processing is similar to that described in the related art.

The reset address RstAdr (n) is read out from the waveform memory 10 and input to the phase reset signal generator 34. The phase reset signal generator 34 reads the reset address DataAdr from the reset address Rs.
It is compared with tAdr (n), and if they match, a phase reset signal is generated and sent to the cosine oscillator 32. Upon receiving the phase reset signal, the cosine oscillator 32 resets the phase of the generated cosine wave to the phase value of the waveform data input from the waveform memory 10 at that time (that is, the phase of the original sound waveform).

As a result, the reset address RstAdr (n)
At the time position, that is, at the head position of the pre-echo section, the phase of each band component signal reproduced by each band component reproducing unit 11-0 to 11-99 is determined by If these are added and synthesized, the pre-echo waveforms of the respective band component signals cancel each other out and disappear from the restored audio signal waveform. Since the flexibility is set to “0” in the pre-echo section, time compression /
Since no decompression is performed, no phase shift due to the time compression / decompression processing occurs in the reproduced signal. Therefore, the state in which the pre-echo waveforms of the reproduced band component signals cancel each other continues over the entire pre-echo section.

The operation in the cosine oscillator 32 is such that when the phase reset signal is input, the held phase is reset in the adder 40 and the phase value directly sent from the waveform memory 10 is output. And the rotational frequency ωkn of the center frequency added. Then, the value of the instantaneous frequency obtained by replacing the phase is given to the cosine calculator 41, and the cosine wave of the phase and frequency is generated and output by calculation. As described above, at least the interval from the reset address RstAdr (n) to the start address of the attack section is set so that the flexibility is 0. Therefore, in the section until the attack is started, the time position is set. The information (read address) has a value such that the original sound is faithfully reproduced. In this section, the pre-echo waveform which is initially divided into bands is reproduced as it is, so that the pre-echo is canceled at the time of synthesis.

The above operation is performed for pitch conversion (pitch shift).
The present invention is most effectively used in the case where such a pitch conversion is not performed. The present invention can be applied by performing the processing.

First, a section having a degree of flexibility of 0 following the reset address is temporarily referred to as a rigid section. For convenience, the length of the rigid section is “L0”, and the length from the reset address of interest to the next reset address is “L”. Further correction is added to the flexibility correction step amount tcomp ′ described above.
That is, in the section of the degree of flexibility "0" following the reset address, tcomp "= P. Here, P is pitch information P (information indicating the amount of pitch shift) supplied to the band component reproducing unit.

Further, in each section up to the next reset address, a re-corrected pitch correction step amount tcomp ", tcomp" = W × tcomp 'obtained by further correcting the flexibility correction step amount tcomp' is used as the step amount. Used. Here, W is the degree of correction, and W = (L / tcomp−L0) / (L / tcomp−L0 / P).

In the rigid section, the phase is advanced by P times because the pitch is converted. Therefore, if the basic step amount tcomp is set to the same value, the phase advance becomes the same as the original waveform, and the original phase is reproduced even in the pitch-converted waveform. Further, since the length of the rigid section becomes 1 / P accordingly, the amount of advance of the address of the remaining section until the next reset address comes is adjusted by the correction degree W, so that the next reset address is reset. Correct the length to the address to the desired time length.
FIG. 7 shows this state, in which a black portion is a rigid section. Time position information is calculated from the pitch correction step amount tcomp ".

In carrying out the present invention, various modifications are possible. For example, in the above-described embodiment, when the phase is reset by the cosine oscillator, the phase generally becomes discontinuous, which may cause noise. In general,
Since the phase reset occurs shortly before the attack, and the amplitude is usually very small, even if the phase of this part is discontinuous, no noise is generated or it is small enough to cause no problem if it occurs. However, improvements can be made to further reduce the possibility of this small noise generation.

That is, in the cosine oscillator, instead of suddenly resetting the phase at the timing of the arrival of the phase reset signal, the sum of the frequency values and the absolute phase (the waveform data of the waveform The phase is continuously changed to the absolute phase by so-called cross-fade, which is performed while gradually changing the ratio of the phase to the absolute phase. It is sufficient that the cross-fade time is shorter than the length of the pre-echo from the reset address, but the reset address can be set in advance so as to be earlier by that amount.

FIG. 8 shows an example of the configuration of a cosine oscillator when performing this cross-fade. In this configuration example, a cross-fade calculator 43 is newly provided, and the phase reset signal, the absolute phase, and the output of the adder 40 (instantaneous phase value)
Is input and a crossfade end notification is sent to the adder 40. The cross-fade calculator 43 normally outputs the data from the adder 40 as it is, and starts the cross-fade when the phase reset signal arrives. After the phase reset, the output signal of the crossfade calculator 43 is
The ratio of the absolute phase value is gradually increased with the passage of time from the instantaneous phase value of the adder 40. At the end of the crossfade, the crossfade end notification is sent to the adder 40.
And rewrites the phase of the adder 40 to an absolute phase value, and switches so that the output signal from the adder 40 is output as an instantaneous phase value.

In the above-described embodiment, the pre-echo can be removed by generating the phase reset signal at the start position of the pre-echo section. However, for example, when the audio signal waveform to be reproduced faithfully reproduces the original sound (that is, when time compression / expansion is not performed), the phase reset signal is used when the waveform reproduction starts or during reproduction. May occur at any time. By doing so, at the time of the phase reset, each phase of the component signal of each band of the audio signal waveform to be restored matches that of the original sound, so that the reproduced waveform becomes more faithful to the original sound.

Further, in the above-described embodiment, the pre-echo can be removed by setting the section having the degree of flexibility “0” as the section of the pre-echo. Can also be set to If the flexibility of the attack section is set to "0" in this manner, even when the audio signal waveform is time-compressed / expanded, the attack portion is reproduced with a waveform similar to the original sound (that is, a waveform that has not been time-compressed / expanded). Will be done. In general, the attack part best describes the characteristics of the musical sound,
If this attack portion is time-compressed / expanded, it sounds like a tone different from the original sound. However, by setting the attack portion to the degree of flexibility “0” as described above, the entire waveform is time-compressed / expanded. Even when decompressing, the characteristics of the musical tone expressed by the attack portion can be well preserved.

[0056]

As described above, according to the present invention, a phase reset signal is generated at the time when a pre-echo occurs, and the phase at that time is set to the same as the original sound. This makes it possible to prevent the occurrence of noise, thereby improving the crispness of the attack sound and preventing the attack feeling from being impaired.

[Brief description of the drawings]

FIG. 1 is a diagram showing an audio signal waveform processing device as one embodiment according to the present invention.

FIG. 2 is a diagram for explaining the concept of an analysis unit for extracting a phase value and an amplitude value from an audio signal waveform in the apparatus according to the embodiment.

FIG. 3 is a diagram illustrating a configuration example of a band component reproducing unit in the apparatus according to the embodiment.

FIG. 4 is a flowchart illustrating a processing procedure of transmitting a phase reset signal by a phase reset signal generating unit in the band component reproducing unit of the embodiment device.

FIG. 5 is a diagram illustrating a configuration example of a cosine oscillator in a band component reproducing unit of the embodiment device.

FIG. 6 is a diagram for explaining the concept of flexibility in the embodiment device.

FIG. 7 is a diagram illustrating a waveform shape when pitch conversion is involved in the embodiment device.

FIG. 8 is a diagram illustrating another configuration example of the cosine oscillator in the band component reproducing unit in the device of the embodiment.

FIG. 9 is a diagram for explaining a relationship between an attack waveform and a pre-echo in a band-divided audio signal waveform.

FIG. 10 is a diagram for explaining the general configuration concept of a conventional audio signal waveform processing device using a phase vocoder method.

FIG. 11 is a diagram illustrating the relationship between components of an audio signal waveform of an original sound and each band obtained by band division.

FIG. 12 is a diagram illustrating waveform data of an audio signal waveform of an original sound.

FIG. 13 is a diagram for explaining a concept of extracting an amplitude value and a phase value by analyzing an audio signal waveform in the apparatus according to the embodiment.

FIG. 14 is a diagram illustrating a configuration example of a time-frequency conversion processing unit in a conventional audio signal waveform processing device.

FIG. 15 is a diagram illustrating a processing example of a signal waveform in a time-frequency conversion processing unit of a conventional audio signal waveform processing device.

[Explanation of symbols]

 Reference Signs List 1 waveform synthesizer 2 CPU (central processing unit) 3 ROM (read only memory) 4 RAM (random access memory) 5 operator group 6 keyboard device 61 lever for speed adjustment 10 waveform memory 11 band component reproducing unit REFERENCE SIGNS LIST 30 differentiator 31 time-frequency conversion processor 32 cosine oscillator 33 multiplier 34 phase reset signal generator 40 sum divider 41 cosine calculator 42 adder 43 cross-fade calculator

Claims (4)

[Claims] An audio signal is divided into a plurality of bands and has amplitude and phase of each band component as waveform data.
An audio signal waveform processing apparatus for restoring an audio signal waveform by reproducing and synthesizing band components of each band while performing time compression / expansion as necessary based on the waveform data. At the beginning of the pre-echo section, the phase of each band component to be reproduced is reset so as to be the phase of the original band component corresponding to each of the sections indicated by the waveform data. Audio signal waveform processing device configured to perform 2. The audio signal waveform processing apparatus according to claim 1, wherein in the section of said pre-echo, time compression / expansion of a restored waveform is not performed. 3. The audio signal waveform according to claim 1, wherein in the phase reset, the phase of the waveform reproduced so far and the value of the phase to be replaced based on the waveform data are replaced by crossfading. Processing equipment. 4. The method according to claim 1, wherein, when the waveform conversion is accompanied by pitch conversion, the degree of phase change based on the pitch conversion and the degree of increment of a read address for reading waveform data are matched. An audio signal waveform processing device according to any one of the above.