JPH0444760B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a device for artificially adding reverberation to a musical tone signal or the like. More specifically, the present invention relates to an improvement of a convolution-type reverberation adding device that generates a reverberation signal by convolving an input signal with a time function of reverberation characteristics.

Conventionally, it was the first convolutional reverberation adding device.
The configurations shown in FIGS. 3 to 3 are known.

This reverberation adding device includes a data memory 23 that sequentially stores a predetermined number of samples of an input signal X(t) obtained by digitizing an analog musical tone signal, etc., and stores reverberation characteristics corresponding to the impulse response of a room to be virtualized. It is provided with a parameter memory 20 and a convolution calculation means for performing a convolution calculation according to the data in both memories 23 and 20 to obtain a reverberation signal OUT.

The data memory 23 enters the write mode every time the clock C2 with the period T output from the timing controller 32 is generated, and the output of the counter 24, which is incremented by the clock C1 with the same period T, is given as the write address ( Note that the output of the counter 24 is given via the subtracter 27, but as will be described later, the subtraction input at this time is 0). In other words, the input signal X is stored in the data memory 23.
(t) are written sequentially at a constant sampling period T,
A predetermined number of latest sample data X ₁ to Xj (X ₁ , X ₂ , X ₃ . . . in order of newest) are stored while being updated sequentially.

The parameter memory 22 stores impulse response characteristics as shown in FIG. 3A in the data format shown in FIG. 3B. One parameter consists of a pair of delay time data Di and gain data Gi (i = 1 to m), and the desired reverberation is achieved using delay time data Di at m points on the time axis and gain data Gi corresponding to each point. Characteristics are expressed. Parameter memory 20 is m+1
Data 0 is written to the first address 0, and parameters are written to the subsequent addresses.
Di and Di are stored in order (note that data at address 0 is assumed to be D ₀ and G ₀ ).

The delay time data Di here is given as an integer value indicating how many times the sampling period T of the input signal is. That is, Di×T is the value of the time dimension.

Writing of the parameters to the memory 20 is performed in advance by the controller 22. Multiplexer 2 connected to address input of memory 20
6 is switched and controlled by the controller 22, and the address line is drawn into the controller 22 only when writing parameters. In other normal operating conditions, the output of the counter 26 is applied to the parameter memory 2 as a read address.

The counter 26 is cleared by a clock C4 that occurs immediately after the clock C1, and is also cleared by a clock having a period of 1/(m+1) of the sampling period T.
It is incremented at C5 and its count output is given to the parameter memory 20 as a read address.

Therefore, every sampling period T, all data stored in the parameter memory 20 are sequentially read out. The delay time data Di among the parameters read from the memory 20 becomes the subtraction input of the subtracter 27. Also, the gain data Gi is multiplier 2
8 and is multiplied by data Xi read out from the data memory 23 as described later. The multiplication output is input to an accumulator 29 consisting of an adder 30, a register 35, and an AND circuit 31.

In detail, as shown in Figure 2, clock C1
The counter 24 is incremented at the rising edge of the clock C4, and at the same time, the clock C4 goes to L level and the counter 25 is cleared. Therefore, the parameter memory 20 is set to address 0 by the output of the counter 25.
is specified, and the data D ₀ =0, G ₀ =0 stored at address 0 as described above is read out.

Then, when the next clock C5 rises, the clock C2 also rises, and the data memory 23 enters the write state. At this time, since the output of the parameter memory 20 is 0, the output of the subtracter 27 is the output of the counter 24 itself. Therefore, the input signal X(t) at this time is written to the address indicated by the output Ak of the counter 24. This is the latest sample data X ₁ , and a certain amount of latest past sample data X ₁ to Xj is similarly stored in the data memory 23 .

When writing of one sample data to the data memory 23 is completed, the data memory 23 enters the read mode. Furthermore, until the next rising edge of the clock C1, the counter 25 is incremented in synchronization with the clock C5, and the output of the counter 25 sequentially specifies addresses 1 to m of the parameter memory 20, and the parameters Di.Gi are sequentially read out. be done.

When the delay time data Di (which is an integer value as described above) is read from the parameter memory 20, the value Ak obtained by subtracting Di from the output Ak of the counter 24 is obtained.
-Di is output from the subtracter 27, which becomes the read address of the data memory 23, and the data corresponding to this address is read from the memory 23.

The address Ak of the data memory 23 is the storage address of the latest sample data _X1 , whereas the address Ak-Di is the storage address of the past sample data Di times. This address Ak−
Let the sample data read from Di be Xi.
This data Xi is data obtained by delaying the input signal X(t) by a time Di×T.

The delay data Xi read from the data memory 23 is multiplied (weighted) by the gain data Gi read from the parameter memory 20 together with the delay time data Di in the multiplier 28.
The output Gi×Xi of the multiplier 28 is the accumulator 29
is input.

Accumulator 29 is synchronized with clock C5.
The output Gi×Xi of the multiplier 28 is sequentially accumulated by an adder 30 and a register 35. And the clock
Immediately after the next rise of C1, the cumulative result for one cycle, Y= _n 〓 ⁱ⁼¹ Gi×Xi, is obtained, and this becomes the reverberation signal output OUT.

Thereafter, during the period when the output of the counter 25 becomes 1, the clock C3 becomes L level, and the AND circuit 31 of the accumulator 29 is cut off. In other words, the contents of the accumulator 29 are once cleared at this point, and the accumulation for the next cycle is started again.

The above is the configuration and operation of the reverberation adding device shown in FIG. A problem with this type of convolution type reverberation adding device is that the number of convolution operations that can be performed by serial sequential processing within a certain period of time is greatly limited by hardware. In the previous example, 1
Although m-point convolution calculations are performed within the sampling period T, the number of calculation points m cannot be freely increased due to restrictions on calculation speed.

The fact that the number m of convolution calculation points cannot be increased means that reverberation with a very long delay time cannot be created. This is because the number of calculation points m is the number m of parameters Di and Gi used in calculation, and it is not possible to appropriately express reverberation characteristics extending to a long delay time with a small number of parameters.

This can be easily understood from FIG. 3A. If the delay time data Di is taken very roughly on the time axis, reverberation characteristics over a long time period can be expressed with a small number of parameters. but,
In this case, even if the length of the reverberation is satisfied, the quality of the reverberation cannot be satisfied (this method is called parameter thinning). Achieving the desired natural reverberation generally requires a large number of parameters that are spaced sufficiently close together in time.

For these reasons, in most conventional systems, reverberation of only the initial delay time period of a desired reverberation characteristic is created by convolution calculation. As long as there is only a limited initial delay time period, the reverberation characteristics can be expressed with high density even with m parameters. However, in this case, the latter part (later delay time range) of the originally desired reverberation characteristics over a large time range is ignored and omitted.

By the way, even if the latter delay time period is omitted, if the input sound (input signal) is continuous, it will not cause much of a problem in terms of auditory sensation. Since the reverberation level in the later delay time period is extremely low, even if a minute level of reverberation exists in the latter half, it is masked by the continuous input sound and has only a slight effect on the auditory sense. Therefore, even if the reverberation in the latter half is omitted, there is no negative effect.

However, if the reverberation in the late delay time period is omitted, when the continuous input sound is interrupted, the subsequent reverberation effect will give a markedly unnatural feel.
Even minute level reverberations can be clearly recognized when the input sound is interrupted. As a result, the unnatural feeling that the reverberation, which should continue while attenuating, disappears midway becomes very noticeable.

This invention was made against the above-mentioned background, and its purpose is to provide a reverberation adding device using a convolution method that allows a more natural reverberation effect to be obtained under the constraint that the number of convolution calculation points is limited. It is about providing. Specifically, the purpose is to generate natural reverberation that extends to a sufficiently large delay time when the input sound is interrupted.

In order to achieve the above object, the present invention includes a data memory that reads input signal data at a constant sampling period T and stores data for a predetermined period of time while sequentially updating; a parameter memory that stores reverberation characteristics expressed by delay data at multiple points on the time axis and gain data corresponding to each point, using a pair of time data and gain data as one parameter; Every time sample data is written, a predetermined number m of consecutive parameters are read out from the parameter memory within the time T, and the input signal data stored in the data memory is read from the parameter memory. a convolution calculation means that performs a convolution calculation of adding weights specified by the parameters to the specified m pieces of data, and outputs the result as a reverberation signal at each time T; the input signal level and minute reference; level discrimination means for comparing the input signal level with the reference level; when the input signal level is equal to or higher than the reference level, the m parameters to be processed by the convolution calculation means are selected from the earliest delay time period; When the level continues to be below the reference level, processing target time period shifting means sequentially shifts the delay time period for selecting the m parameters to a side with a larger delay time.

When the delay time period to be calculated is shifted to the latter half, the earliest delay time period is omitted. However, the input signal at this time during this period is 0.
(silence), even if this is the object of calculation, the result will be 0, and omitting this will have no practical effect at all. The amount omitted here can be effectively utilized to create more reverberations in the latter half.

Hereinafter, embodiments of the present invention will be described in detail based on the drawings.

FIG. 4 shows the configuration of a first embodiment of the invention. Many parts of the configuration of this reverberation adding apparatus are exactly the same as the apparatus shown in FIG. 1, and the common parts are given the same reference numerals. The previous explanation will be used loosely for the configuration and operation common to the device shown in FIG. 1, and the following description will focus on the parts of the device shown in FIG. 4 that are different from the device shown in FIG. Each clock C5 also occurs at the timing shown in FIG.

Further, it is assumed that the number of convolution calculation points that can be performed by serial sequential processing within one sampling period T is m points, which is the same as that in FIG. 1 in the apparatus of the present invention.

In the device of the present invention, the parameter memory 20 stores z parameters Di, Gi(i
=1~m~z) are stored. These large numbers of parameters Di and Gi express reverberation characteristics that extend to long delay times. This situation can be seen in the fifth
Shown in Figures A and B. In other words, the reverberation characteristics over a wide time period necessary to obtain the desired natural reverberation are expressed with a sufficiently high density by a large number of parameters, and are stored in the parameter memory 20. Further, in this embodiment, an absolute value detection circuit 41 for detecting the absolute value of the input signal X(t) as a level discrimination means for comparing the level of the input signal X(t) with a minute reference level; A comparator 42 that compares the output with a reference value N that is close to 0.
is provided. When the input signal level is equal to or higher than the reference value N, the comparator 42 outputs a counter 4.
3 is cleared, and when the input signal level is below the reference value N, the clearing is canceled.

When the counter 43 is cleared by the output of the comparator 42, its output is 1 (not 0). When the counter 43 is cleared, the counter 43 is incremented by the clock C1, and its output changes from 1 to
It increases from 2→3→.

The counter 25, which provides the read address of the parameter memory 20, is cleared by the clock C4 (its output becomes 0), and after one cycle of the clock C5, the counter 25 is cleared by the clock C2.
Read (preset) the output of step 3. Thereafter, it is sequentially stepped by clock C5.

That is, if the counter 43 is cleared and its output is 1, the output of the counter 25 becomes 0 in synchronization with clock C4, becomes 1 in synchronization with clock C2, and then continues until the next clock C1 occurs. In the meantime, m-1 clock C5 was received and m
increase to. In other words, the output of counter 25 is 0→
It changes from 1 → 2 →...m. This is exactly the same operation as the device shown in FIG.

Therefore, in the above state, parameter memory 2
Among the z parameters stored in 0, m parameters in the earliest delay time period are used to perform the convolution operation. Therefore, the reverberation signal output OUT at this time becomes exactly the same as the operation of the apparatus shown in FIG.

The above operation is an operation when the level of the input signal is continuously equal to or higher than the reference value N. In this operation, the second half of the reverberation characteristics stored in the parameter memory 20 is ignored, but this does not have much of an adverse effect on the reverberation effect due to the masking effect of continuous input sound.

When the input sound stops, that is, when the input signal level becomes smaller than the reference value N, the counter 43 is cleared, and the counter 43 is incremented by the clock C1. Counter 43 is 1
When the counter 25 is incremented by 2, the output becomes 2, and in this state, the output of the counter 25 is 0 → 2 → 3 → 4 →...m
Changes to +1. In other words, the parameters D ₂ , G ₂ ~Dm
The convolution operation is performed using m parameters of +1 and Gm+1.

If the silence continues further, the output of the counter 43 becomes 3.
The output of the counter 25 is 0 → 3 → 4 → 5 →
...changes to m+2. At this time, D ₃ , G ₃ ~Dm+
The convolution operation is performed using m parameters up to 2.Gm+2.

In this way, when the input signal level remains below the reference level, the delay time periods for selecting the m parameters are sequentially shifted to the side with larger delay times. Then, the parameters of the earliest delay time zone will be ignored in order from D ₁ to G ₁ .
However, at this time, the input signal level continues to be silent, so the sample data of the input signal corresponding to the ignored parameter is almost 0, 0×
Since the result of Gi is also 0, there is no need to perform this operation in the first place.

Instead of ignoring the parameter in the earliest delay time period, the convolution operation is performed using the parameters in the delay time period after parameters Dm and Gm, so in the silent state after the input sound has stopped, the parameter memory is According to the reverberation characteristics stored in 20, reverberation over a long delay period is created, and natural reverberation that gradually decays over a long period of time is generated.

FIG. 6 shows the configuration of a second embodiment of the invention. This device differs from FIG. 4 in that the absolute value detection circuit 41 in FIG. 4 is replaced by a power calculation circuit 44. This power calculation circuit 44 squares the input signal X(t) using a squaring circuit, integrates its output in synchronization with a clock C5 using an integrating circuit consisting of an adder 46 and a register 47, and outputs the output via a register 48. It is configured to apply the signal to the comparator 42 in synchronization with the clock C1.

Note that when the latest input signal is received, it is necessary to recognize the power at that time, so in this embodiment, input data for a certain period of time in the past is temporarily held in the pre-memory 49. The power is calculated in advance based on the contents of this holding memory. These pre-memory 4
In step 9, the input data is delayed by the number of pieces required for power calculation, and during this time, the power is calculated from these data, and at the same time, the corresponding input data is sent to the data memory via the buffer register 50 to synchronize the timing. For these controls, the aforementioned control signal _C2 and the address designation signal line of the data memory are also used.

In other words, it is determined whether the input signal is interrupted (sound state or silent state) based on the average value of the power of the input signal X(t). According to this embodiment, it is possible to more accurately discriminate between voiced and silent states in accordance with the statistical properties of musical tone signals and the like.

Note that the frequency band of the input signal is divided into several parts, and the reverberation adding device of the present invention is provided for each band.
These outputs may be combined to form a final reverberant signal. In this case, appropriate reverberation characteristics can be set for each frequency band. As is well known, high-frequency components of reverberant sound decay quickly and have a short reverberation time, while low-frequency components decay slowly and have a long reverberation time. A reverberation effect commensurate with such actual reverberation characteristics can be achieved by providing a reverberation load device for each of a plurality of bands.

In particular, when the present invention described above is applied to a system in which reverberation devices are provided in a plurality of frequency bands, the following effects are added. In other words, it is assumed that the low-frequency component of the musical tone signal has disappeared, or that the high-frequency component continues or is intermittent. Even in this case, according to the above configuration, the discontinued low frequency component is completely unaffected by the continuing high frequency component, and natural reverberation occurs for a long time with the low frequency component alone.

As described in detail above, according to the present invention, even under the constraint that the number of convolution calculation points is limited in terms of hardware, a more natural reverberation effect over a wide reverberation time zone can be obtained.

[Brief explanation of drawings]

Fig. 1 is a block diagram of a conventional reverberation adding device, Fig. 2 is a timing chart showing the operation of the device shown in Fig. 1, and Fig. 3 is an explanatory diagram of the reverberation characteristics stored in the parameter memory in the device shown in Fig. 1. , FIG. 4 is a block diagram of the first embodiment of the reverberation adding device according to the present invention, FIG. 5 is an explanatory diagram of the reverberation characteristics stored in the parameter memory in the device of FIG. 4, and FIG. FIG. 2 is a block diagram showing a second embodiment of such a reverberation adding device. 20...Parameter memory, 23...Data memory, 29...Accumulator, 41...Absolute value detection circuit, 42...Comparator, 43...Counter, 44
...Power calculation circuit.

Claims (1)

[Claims] 1. A data memory that reads input signal data at a constant sampling period T and stores the data for a predetermined time while sequentially updating; A parameter memory that stores reverberation characteristics expressed by delayed data at multiple points on the time axis and gain data corresponding to each point, with each pair as one parameter; each time one sample data is written to the data memory. Then, within the time T, read out a predetermined number of consecutive m parameters that are part of the parameter memory, and read m pieces of input signal data specified by the parameters out of the input signal data stored in the data memory. a convolution calculation means that performs a convolution calculation that adds weights to the data specified by the parameters, and outputs the result as a reverberant signal every time T; a level that compares the input signal level with a minute reference level; Discrimination means: When the input signal level is above the reference level, selects the m parameters to be processed by the convolution calculation means from the earliest delay time period, and when the input signal level is below the reference level. A reverberation adding device characterized by comprising: processing target time period shifting means for sequentially shifting the delay time period in which the m parameters are selected to a side with a larger delay time when the condition continues.