JPH0562999B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to a pitch conversion method in which only the frequency of an input signal is changed while the time axis progression of the signal as a whole remains constant.

[Prior art] Pitch transform is a basic technique for signal processing.
It has a variety of uses, such as changing only the pitch without changing the rhythm, returning the pitch to the normal pitch when playing a tape in fast forward, and vibrato.
It is used for sound reproduction special effects such as beat generation, and for natural sound simulation such as Doppler effect and chorus effect.

As shown in Figure 2, the conventional pitch conversion method is to reproduce the original signal at a short period T' to increase the pitch, or to reproduce it at a long period T'' to decrease the pitch, relative to the sampling period T of the original signal. As shown in Figure 3, the reproduction period is the same as the sampling period T.
Once every predetermined number of samples according to the pitch conversion ratio,
There are methods such as skimming and pitching up, or reading twice and pitching down (Figure 3 shows examples of 2x and 1/2x). Note that when pitching down, it is also possible to insert these interpolated data between the preceding and succeeding data to further reduce the distortion of the reproduced waveform.

[Problem that the invention seeks to solve]

However, with any of these methods, if continuous processing is continued for a long time, the time difference will increase between the original signal and the pitch-converted signal, and the time axis progress of the signal will not be the same, resulting in rhythm will no longer match. Therefore, in order to absorb these time axis deviations, it becomes necessary to reproduce short and close signal portions within a range that is not noticeable to the audible sense twice when pitching up, or truncating when pitching down. For this reason, the weighting of the time axis of the converted signal with respect to the flow of the original signal is inevitably not uniform, resulting in unnaturalness. Furthermore, the truncated information is completely lost.

Furthermore, since signal continuity is inevitably lost at signal junctions, noise and level fluctuations due to unnatural phase interference occur. In order to solve this problem, it may be possible to perform crossfade processing on this connecting portion, but even with this, unexpected dips will occur if the signals are out of phase during crossfade. Therefore, it may be possible to control the timing phase to match in crossfade processing, but
It is clear that phase detection means would be required and would involve considerable technical difficulty.

This invention solves the problems in the conventional technology, eliminates the unnaturalness caused by double reading and truncation for pitch conversion, and easily eliminates poor continuity at joints. This provides a pitch conversion method that can be used.

[Means for Solving the Problems] The present invention sequentially generates at least three signal waveforms by partially pitch-converting an input signal at different timings, and By sequentially connecting the old waveform and the latest waveform while other waveforms other than these two are being output and without leaving a gap between the oldest waveform and the latest waveform, The present invention is characterized in that a pitch-transformed signal is obtained by continuously synthesizing these signal waveforms while sequentially exchanging the signal waveforms.

[Operation] According to the solution means of the present invention, all parts of the input signal can be used universally in pitch conversion, and it is possible to read only a part of the signal twice or to cut out a part. This makes it possible to make the weighting of the signals more uniform, and since the multiple signal waveforms that are synthesized complement each other in their context, the continuity of the signals is sufficiently maintained, creating a natural-sounding result. You can get pitch changes. In other words, since other pitch-converted waveforms are output at the joint portions of the signal waveforms, the joint portions are masked and become less noticeable audibly. Moreover, since the pitch-converted waveform output at this joint portion is a signal that is temporally close to the joined signals, it does not feel unnatural.

[Embodiment] First, an embodiment of the present invention is shown in FIG. 1 as a block diagram. The input signal (original signal) to be pitch-converted is divided by the distribution control circuit 10 into unit sections that overlap with each other, as shown in the figure. , 14, 16, 18 sequentially and cyclically. As will be described in detail later, each unit interval in the input signal is set to approximately 20 to 50 ms in time, and each start position is offset by approximately the width of the converted interval divided by the number of systems (actually (In order to prevent coloration, which will be explained later, the deviation is not completely uniform, but is consciously randomized to some extent.)

Each of the pitch conversion circuits 12, 14, 16, and 18 has a memory, and after once storing the distributed input signal waveform in these memories, the pitch conversion circuits 12, 14, 16, and 18 can be used in the various methods described above, for example, in the case of pitch down. You can either lengthen the data read cycle or read the address twice once every few samples depending on the conversion ratio without changing the cycle, or in the case of pitch-up, you can shorten the data read cycle or read the address twice without changing the cycle. By skipping reading one piece of data every few samples depending on the conversion ratio,
Pitch conversion is performed at a predetermined conversion ratio. As a result, a plurality of signal waveforms obtained by partially pitch-converting the input signal at different timings are sequentially obtained at the output of each pitch conversion circuit. These outputs are given coefficients a1, a2, a3, and a4 as necessary by coefficient multipliers 20, 22, 24, and 26, respectively, and are then fully added in an adder 28 to produce a final pitch conversion output signal. are synthesized.

In this synthesis process, as mentioned above, the signal distribution to each pitch conversion circuit is performed cyclically, and each distribution section is set as shown in the figure. Approximately at the same time as the pitch conversion of the input signal is completed, a new input signal is distributed, and this input signal is pitch-converted again, so that the conversion process is continuously performed. Therefore, at the same time that the addition of the oldest converted waveform ends and it is removed from the addition content, the latest waveform whose conversion has just started will be added to the addition content. This means that the current waveform and the latest waveform are continuously added and synthesized while being sequentially replaced.

In addition, during this replacement, the coefficients of the coefficient multiplier are complementarily increased or decreased for each of the waveforms to be replaced, that is, the coefficients of the waveforms that are excluded from addition are changed, for example, from 1 → 0.9 → ... → 0.1 → 0. , change the coefficient of the waveform newly added to the addition from 0 → 0.1 →
...→0.9→1 and perform crossfade processing, the synthesized power average will always be approximately three times the power of one waveform, and even in the swapped part, there will be no peak dip in power, and it will be smooth. It can be changed. Specifically, in each pitch conversion circuit, the above-mentioned crossfade processing is performed at the beginning of conversion (in this case, about the first 1/4) and at the end of conversion (also about the last 1/4).

According to such a pitch conversion method, the input signals distributed to each of the pitch conversion circuits 12, 14, 16, and 18 are calculated by taking into account the time required to convert the distribution on the time axis. Even when pitch-down, that is, extending the waveform in the time axis direction, each conversion circuit can utilize and convert the entire input signal without cutting off the rear part of the supplied input signal, and even when pitch-up,
The output of each pitch conversion circuit is shortened in the time axis direction, but the next distribution is supplied from the input signal waveform at that point almost immediately after the conversion ends, and this conversion starts at the same time. In either case, the output waveform has almost no blank space, and the final addition/synthesis output is completely seamless and continuous.

Also, from the perspective of continuity of signal content, for any waveform to be added, there is always another waveform generated from the previous and previous time portions of the input signal; They complement each other, and as a result, the converted output does not lose its smoothness in terms of content.

Furthermore, the addition state always changes continuously, and in conjunction with the crossfade processing mentioned above, the combined power becomes the same in terms of time, making it possible to achieve extremely natural-looking pitch conversion.

In addition, if we consider pitch conversion sound by synthesizing multiple waveforms of this kind from an auditory standpoint, we will find that synthesis of multiple waveforms like this is similar to the synthesis of early reflected sounds in architectural acoustics, etc., and the frequency If you look at the area, the sound will be as if it has passed through a multi-resonant system, but in the case of pitch conversion as described above, if the number of systems is n, then 1/n reflected sound components will be
The difference is that the sound appears and disappears within a short period of 50ms, which is completely unnoticeable to the auditory sense. Therefore, it can be said that the larger the n, the less the influence of the multiple resonance etc., and
Even in the case of reflected sound, as mentioned above, the distance from the direct sound is 50 m.
It is known that the sound that comes within 20ms has the role of reinforcing the direct sound, that is, it cannot be heard as an echo, so even in this case, the above 20~50ms If a plurality of waveforms are generated and synthesized in a time range, a sufficiently natural impression can be given to the auditory sense without causing any discomfort.

The pitch conversion operation using the pitch conversion configuration shown in FIG. 1 will be explained in a more general form with reference to FIGS. 4 and 5.

(A) Pitch down Figure 4 is an example of pitch down by 1/N. Here, N is any number greater than or equal to 1,
For example, if the pitch is down by 10%, N=1.1111.

As shown in the figure, the input signals to be pitched down are divided into unit sections, ..., ..., which overlap each other and whose starting points are slightly shifted, and these are divided into M systems (here, for convenience of explanation, M
= 4 will be described, but the pitch conversion circuits 12, 14, 16, and 18 (not limited to this) are sequentially and cyclically distributed as shown in the figure. The width U of each unit section is short, about 20 to 50 ms in time, and in this way, the input signal is processed by cutting it into small parts that do not cause any problems with hearing. It will be done. The starting position of each unit section is shifted by approximately N times M of the width U of each unit section, but it is not completely evenly shifted and is intentionally randomized to prevent so-called coloration. We are trying to

Coloration is a phenomenon in which a certain frequency component of the output signal is emphasized due to the periodicity of processing, that is, due to electrical processing that involves simple repetition, resulting in a so-called metallic sound to the auditory sense. Once this development occurs, the sound becomes extremely unnatural. If each of the unit sections mentioned above is simply divided into equal parts and shifted evenly, this development will inevitably occur, so here we will improve the processing by appropriately changing the starting position of each unit section within a certain range. Eliminates periodicity and reduces coloration. Specifically, a random function may be introduced into the processing address when distributing the input signal or reading out the waveform. In addition, the same thing is
This can also be achieved by providing randomness to the width of each unit section itself.

Each unit interval, ..., ... (each partial portion of the input signal) distributed to each pitch conversion circuit 12, 14, 16, 18 is subjected to 1/N pitch down processing, and the time width is N.U. is expanded to Specifically, a method is used in which the data read cycle is increased by N times, or the data is read twice once every 1/(N-1) times without changing the read cycle.

As a result, each pitch conversion circuit generates output waveforms ′, ′, ..., ′... which are pitch-converted from each unit section, . . . , as shown in the figure.
are obtained sequentially. If we look at a certain pitch conversion circuit, for example the conversion circuit 12, after converting the unit interval and outputting the output waveform ', it starts converting the unit interval and creates the output waveform ', and so on. The pitch conversion circuit performs conversion processing with almost no play, and is highly efficient. The reason for this is that, as mentioned above, the starting position of each unit interval, ..., ... is temporally randomized to some extent to prevent coloration.
Depending on the degree of the deviation, there may be a slight blank period at the joint between the waveforms sequentially generated in one conversion circuit, but this is practically negligible and is not a problem at all. .

In addition, the pitch conversion circuits 12, 14, 16, 1
Each coefficient multiplier 20, 2 provided on the output side of 8
2, 24, and 26, for the pitch-converted output waveform of each unit interval, as shown in the figure, the coefficient is gradually increased from 0 to 1 in the initial 1/4 interval, and the coefficient is kept at 1. 1/2
As the interval elapses, the coefficient is gradually decreased from 1 to 0 again in the final 1/4 interval. Then, the coefficient-added output waveforms existing at the same time are completely added together by an adder 28, and a final pitch-transformed signal is synthesized.

In this process, each waveform is replaced and crossfade processing is performed at that time. That is, when the oldest of multiple waveforms existing simultaneously at a certain point in time, for example, the output waveform', is removed from the addition by the adder 28,
The level of the same waveform ′ gradually decreases, but the latest waveform ′ newly added to the addition
On the contrary, it gradually increases in size. Since this complementary level control, or crossfade, is performed in the same way in all waveform exchange parts, the power average of the final pitch-converted signal waveform produced at the output of the adder 28 is always constant (in this case, 1
3 times the average power of the system), and so,
There are no unnatural peaks or dips that are aurally unpleasant.

Furthermore, for any one of the plurality of output waveforms added by the adder 28, there is always another output waveform generated from a portion of the input signal at the same time or at a time before or after the same time. For example, regarding the output waveform', the input signal content of the first half is common to the second half of the output waveform', and the input signal content of the second half is common to the first half of the output waveform'. Depending on the set pitch conversion ratio, there may not necessarily be many common parts, but in any case, there are always signals with extremely strong correlations in the output waveforms that are added. It is certain. Therefore, these components complement each other, and smoothness is not lost from the viewpoint of continuity of signal content.

Further, according to the above operation, the entire waveform can be used uniformly to perform pitch down without cutting off the rear part of the input signal waveform as in the conventional case.

(B) Pitch Up Figure 5 is an example of pitch up N times. Again, N is any number greater than or equal to 1; for example, in the case of a 10% pitch up, N=1.1.

As shown in the figure, the input signals to be pitched are divided into each unit section,...
4 will be described as an example, but is not limited to this). The width U of each unit section is approximately 20 to 50 ms in time. The starting position of each unit section is shifted by approximately one (NM) of the width U of each unit section, and is intentionally randomized as in the case of pitch down in FIG. 4.

Each unit section distributed to each pitch conversion circuit 12, 14, 16, 18 (each partial portion of the input signal) is subjected to N-times pitch up processing, and the time width is reduced to N/U. be done. As a method, the data read cycle may be increased by 1/N times, or the read cycle may remain the same and one piece of data may be skipped and read once every 1/(N-1) times.

As a result, each pitch conversion circuit generates output waveforms ′, ′, ..., ′, which are pitch-converted for each unit section, , , , and so on as shown in the figure.
are obtained sequentially, and the coefficient multipliers 20, 22, 24,
At step 26, in the same manner as described above, coefficients are added as shown in the figure in order to make the power average constant, and then the adder 28 performs full addition to synthesize the final pitch-transformed signal.

According to the above operation, the pitch can be increased without reading part of the input signal waveform twice as in the conventional case.

[Modification] In the embodiment described above, the number M of pitch conversion systems is set to 4, but generally speaking, the more the number of systems increases, the more uniformity of the pitch conversion signal can be obtained.

[Effects of the Invention] As explained above, the pitch conversion method of the present invention sequentially generates a plurality of signal waveforms by partially pitch-converting an input signal at different timings, and also generates a plurality of signal waveforms among the plurality of signal waveforms. By sequentially exchanging the oldest waveform and the latest waveform and continuously synthesizing these multiple signal waveforms to obtain a pitch-converted signal, for pitch-conversion, only a part of the signal is converted into 2 Reading it multiple times, etc.
There is no need to cut out a part, all parts of the input signal can be used universally, the weighting of the signal can be made more uniform, and the multiple signal waveforms to be synthesized can also be related to each other. Because they complement each other, the continuity of the signal is maintained sufficiently and it is possible to obtain natural-looking pitch changes. In other words, since other pitch-converted waveforms are output at the joint portion of the signal waveform, the joint portion is masked and becomes less noticeable to the audible sense.
Furthermore, since the pitch-converted waveform output at this joint portion is a signal that is temporally close to the joined signals, it does not feel unnatural. and,
This makes it possible to make the joints less noticeable without the need for traditional cross-fade processing, eliminating the need for troublesome phase alignment processing and easily resolving poor continuity at the joints. be able to.

Further, even when cross-fade processing is performed for the purpose of making the power average of the synthesized pitch-transformed signal waveform constant, there is no need for troublesome phase matching processing because of the masking effect at the joint portion.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing an embodiment of the pitch conversion method of the present invention. Figures 2 and 3 are
FIG. 6 is a waveform chart showing a conventional pitch conversion method. FIG. 4 is an explanatory diagram showing an example of the operation during pitch down by the pitch conversion method of the present invention. FIG. 5 is an explanatory diagram showing an example of the operation during pitch up by the pitch conversion method of the present invention. 10... Distribution control circuit, 12, 14, 16, 1
8... Pitch exchange circuit, 20, 22, 24, 26
... Coefficient multiplier, 10 ... Distribution control circuit.

Claims (1)

[Claims] 1. Partial pitch conversion of an input signal at different timings to sequentially generate at least three signal waveforms, and among these signal waveforms, the oldest waveform and the latest waveform By sequentially connecting these two waveforms while other waveforms are being output and without leaving a gap between the oldest waveform and the latest waveform, the signal waveforms can be sequentially exchanged and continuous. A pitch conversion method characterized in that a pitch conversion signal is obtained by composing these signal waveforms. 2. The method according to claim 1, characterized in that crossfade processing is performed when exchanging the oldest waveform and the latest waveform so that the power average of the synthesized pitch-transformed signal waveform is constant. Pituchi conversion method.