JP3789997B2 - Waveform converter - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a waveform conversion apparatus used for a musical sound signal conversion process for generating an output waveform by changing the pitch or formant of an input waveform.
[0002]
[Prior art]
A kind of pitch conversion device that converts the pitch of an input musical sound signal to a pitch selected by a keyboard is used for the conversion processing of the musical sound signal. An example thereof is disclosed in Japanese Patent Application Laid-Open No. 62-65098. There is technology. In this method, musical tone data that has been input and stored is analyzed to extract a phoneme including a required data section, and the extracted phoneme is output at a cycle selected by keyboard operation.
[0003]
In this prior art, when a phoneme is cut out, a position where the correlation coefficient is maximized in relation to the previous cutout position is selected and cut out. However, the process for obtaining such a correlation coefficient is complicated, and the load on the CPU and DSP (digital signal processor) becomes heavy. In the case of real-time processing in which processing needs to be completed within a predetermined time (for example, within a sampling period), simple processing with a light load on the CPU and DSP is preferable.
[0004]
On the other hand, as a simple processing method for cutting out music in real time, for example, there is a technique disclosed in Japanese Patent Laid-Open No. 3-288200. This is to detect the pitch of the input musical sound signal based on the time interval between the zero cross position and the peak position, and to set the reference position (start position) at the time of phoneme extraction as the reference position at the time of pitch detection. .
[0005]
[Problems to be solved by the invention]
However, the technique disclosed in Japanese Patent Laid-Open No. 3-288200 has the following problems. FIG. 13 is an explanatory diagram of the problem.
Now, as shown in FIG. 13A, it is assumed that input waveform data that includes harmonics and that changes temporally is input. The input waveform data is sequentially stored in the memory. FIG. 13B shows the fundamental wave component of this input waveform data. First, in FIG. 13A, a waveform surrounded by the detected zero-cross points (including a set of a plurality of waveforms surrounded by the zero-cross points) is surrounded by the previous zero-cross points. When there is a similarity between them, the pitch is detected with the waveform between the detected zero-cross points as one period. Here, it is assumed that the waveform before S1 (indicating the address) in FIG. 13A is similar to the waveforms from S1 to S2.
[0006]
However, the waveform portion from S3 to S4 is not similar to the waveform portion from S2 to S3 immediately before, and therefore the pitch c is regarded as an abnormal pitch. Similarly, the waveform portion from S4 to S5 is not similar to the waveform portion from S3 to S4 immediately before, and the pitch d is also regarded as an abnormal pitch. That is, regarding the pitches c and d, it is considered that the pitch detection has failed, and a, b, e, and f that are similar to the immediately preceding waveform are eventually adopted as the pitch detection values.
[0007]
Then, waveform reading is started from the reference point when the pitch detection was successful immediately before. That is, the waveform data of the pitch a with the reference point S1 of the waveform of FIG. 13A is read from the memory at the time S1 of FIG. 13C, and then at the time of S2 of FIG. The waveform data of pitch b with reference to S2 of the waveform of (a) is read. Reading of the waveform data of the pitch b with reference to S2 is repeated until the pitch detection is newly succeeded. Thereafter, the waveform data of the pitch e with reference to S5 of the waveform of FIG. 13A is read at the time of S5 in FIG. 13C, and further, at the time of S6 in FIG. The waveform data of the pitch f with reference to S6 of the waveform is read.
[0008]
In this way, the input waveform data is sequentially read out, but as is apparent from the S5 part of FIG. 13C, the connection between the fundamental wave components of the pitch b and the pitch e is poor. In a discontinuous state. If such a discontinuous portion of the fundamental wave component exists, the musical sound in this portion becomes unnatural and the value as a musical effect is greatly diminished.
[0009]
In view of the above circumstances, the present invention can prevent the occurrence of discontinuous portions of the fundamental wave component and always obtain a natural output waveform while adopting a simple processing method for real-time processing as in the past. An object of the present invention is to provide a waveform converter capable of performing the above.
[0010]
[Means for Solving the Problems]
The waveform conversion apparatus of the present invention that achieves the above object detects the pitch of input waveform data that is sequentially input, cuts out and extracts a phoneme comprising a predetermined data section corresponding to the detected pitch of the input waveform data. In a waveform converter for converting phonemes to output waveform data having a pitch corresponding to predetermined reproduction pitch information and outputting the data,
(1) Memory that temporarily stores waveform data
(2) Writing means for sequentially writing input waveform data input sequentially to the memory
(3) A first buffer for storing the pitch information in an updatable manner is provided. When the pitch detection process is executed for the input waveform data and the pitch detection is established, the pitch information stored in the first buffer is newly set. Pitch detection means for updating to detected pitch information
(4) It has a second buffer for renewably storing a read reference address serving as a reference for extracting a phoneme from the waveform data stored in the memory, and corresponds to pitch information read last time from the first buffer. An address for reading the pitch information from the first buffer every time elapses and updating the read reference address stored in the second buffer by an address increment corresponding to the pitch information newly read from the first buffer Update means
(5) Reproduction pitch information input means for inputting reproduction pitch information
(6) Reading a read reference address from the second buffer, reading out waveform data in a predetermined data section from the memory based on the read read reference address, cutting out the phoneme, and inputting the cut out phoneme as reproduction pitch information Output waveform generating means for generating output waveform data by synthesizing at a period corresponding to the reproduction pitch information inputted from the means
It is provided with.
[0011]
The waveform conversion apparatus of the present invention updates the read reference address by sequentially adding the detected pitch when the pitch detection is established, that is, the correctly detected pitch, and thereby erroneously detecting the pitch. Even in this case, an output waveform without a discontinuous portion of the fundamental wave is obtained, and the quality of the musical effect is greatly improved.
Here, in the waveform conversion apparatus of the present invention, the output waveform generation means (6) is
(6_1) There is a third buffer to which the read reference address stored in the second buffer is transferred each time output waveform data corresponding to two playback pitches is generated, and stored in the third buffer. First reading means for sequentially generating read addresses for two pitches of input waveform data based on the read reference address and sequentially reading the waveform data stored in the sequentially generated read addresses in the memory
(6_2) The output waveform data corresponding to the reproduction pitch of 2 pitches is the timing at which the read reference address stored in the second buffer is shifted by the reproduction pitch of 1 pitch from the timing of the transfer to the third buffer. Each time it is generated, it has a fourth buffer to which the read reference address stored in the second buffer is transcribed, and the input waveform data 2 pitches based on the read reference address stored in the fourth buffer Second read means for sequentially generating the read addresses for the minutes and sequentially reading the waveform data stored in the sequentially generated read addresses of the memory.
(6-3) The waveform data read by the first reading means and the waveform data read by the second reading means are weighted using weights that periodically change with two pitches as one cycle. It is preferable that weighted addition means for generating output waveform data by adding is provided.
[0012]
When these (6_1) to (6_3) are provided, discontinuity of the output waveform at the time of pitch update is prevented, in other words, a smooth continuous output waveform can be obtained even if the pitch is updated sequentially, and the quality of the musical effect Is further improved.
[0013]
DETAILED DESCRIPTION OF THE INVENTION
Hereinafter, embodiments of the present invention will be described.
FIG. 1 is a block diagram showing a configuration of an embodiment of a waveform conversion apparatus of the present invention. The input waveform signal input through the low-pass filter 1 is A / D converted by the A / D converter 2 and sent to the DSP 3. The DSP 3 writes the input waveform data in the ring buffer memory (RAM) 4 and cuts out the written data with a predetermined time delay to generate output waveform data in which the pitch or formant is changed. The output waveform data output from the DSP 3 is D / A converted by the D / A converter 5 and output to the outside through the low pass filter 6. The operating element 7 sets various setting conditions such as parameters, and the setting signal is sent to the DSP 3 via the CPU 8. In this operator 7, the reproduction pitch and the reproduction fault man are set as the respective ratios with respect to the input pitch and the input fault man as setting conditions directly related to the present embodiment. Details will be described later.
[0014]
FIG. 2 is a block diagram showing a detailed configuration of the DSP 3 in FIG. FIG. 2 shows an operator 7, and a setting signal obtained by operating the operator 7 is input to the DSP 3 via the CPU 8 as shown in FIG. 1. However, in FIG. 2 and FIG. 3 to be described later, the CPU is not shown in order to avoid the complexity of the illustration. As shown in FIG. 2, the DSP 3 includes a writing unit 31, a pitch detecting unit 32, an address updating unit 33, and an output waveform generating unit 34.
[0015]
The writing means 31 sequentially writes the digital input waveform data sequentially output from the A / D converter 2 into the ring buffer memory 4.
The pitch detector 32 sequentially detects the pitch of the input waveform data. The pitch detection means 32 has a buffer 32a for storing the pitch of the input waveform data in an updatable manner, executes the pitch detection process for the input waveform data, and detects the pitch detection process, that is, here, immediately before detection. When the new pitch detected this time is the correct pitch (the pitch is almost equal to the pitch detected immediately before) in light of the criterion that the new pitch detected this time and the new pitch detected this time are the same pitch. It is determined that pitch detection has been established, and the pitch stored in the buffer 32a is updated to the new pitch detected this time.
[0016]
The address updating means 33 updates the reference address for reading the waveform data from the ring buffer memory 4, which is a reference address for cutting out phonemes in the output waveform generating means 34 described later. The address update means 33 includes a buffer 33a for storing a read reference address as a reference for extracting a phoneme from the waveform data stored in the ring buffer memory 4 so that it can be updated. From the previous update of the buffer 33a, Each time the time corresponding to the pitch information in the buffer 32a of the pitch detection means 32 elapses, the address update means 33 is stored in the buffer 32a of the pitch detection means 32 at the read reference address stored in the buffer 33a. The pitch is read, the corresponding address increment is added to the newly read pitch to obtain a new read reference address, and the new read reference address is replaced by the buffer instead of the read reference address previously stored in the buffer 33a. 33a is stored. In other words, the read reference address stored in the buffer 33a is sequentially accumulated by address increments corresponding to the pitch detected by the pitch detecting means 32 and stored in the buffer 32a.
[0017]
Further, the output waveform generation means 34 reads out the read reference address from the buffer 33a of the address update means 33 every time output waveform data corresponding to one reproduction pitch is generated, and updates it sequentially based on the read read reference address. Is read out, the phoneme is cut out by reading out the waveform data stored in the read out address from the ring buffer memory 4, and the extracted phoneme is input to the pitch information input by operating the operator 7. And output waveform data having a reproduction pitch and a reproduction formant corresponding to the formant information. The output waveform represented by the output waveform data is a waveform in which one or both of the pitch and formant are changed (including the case where they are the same) as compared with the input waveform.
[0018]
FIG. 3 is a block diagram showing a detailed configuration of the output waveform generating means 34 shown as one block in FIG.
The output waveform generating means 34 shown in FIG. 3 includes a first reading means 341, a second reading means 342, and a weighted adding means 343.
The first reading means 341 has a buffer 341a to which the read reference address stored in the buffer 33a of the address updating means 33 is transcribed every time output waveform data corresponding to two reproduction pitches is generated. Read addresses are sequentially generated based on the read reference address stored in 341a, and the waveform data stored in the sequentially generated read addresses in the ring buffer memory 4a are sequentially read.
[0019]
Similarly to the first reading unit 341, the second reading unit 342 transfers the reading reference address stored in the buffer 33a of the address updating unit 33 every time output waveform data corresponding to two playback pitches is generated. A buffer 342a. However, the transfer timing is the timing at which the read reference address stored in the buffer 33a of the address update means 33 is transferred to the buffer 341a of the first read means 341 is shifted by one reproduction pitch. . That is, the read reference address stored in the buffer 33a of the address updating unit 33 is alternately output to the buffer 341a of the first reading unit 341 or the second reading every time output waveform data corresponding to one reproduction pitch is generated. The data is transferred to the buffer 342a of the means 342. Similar to the first reading unit 341, the second reading unit 342 sequentially generates a reading address based on the reading reference address stored in the buffer 342 a and stores the reading address in the ring buffer memory 4 at the sequentially generated reading address. Sequentially read the waveform data.
[0020]
The waveform data read by the first reading unit 341 and the waveform data read by the second reading unit 342 are input to the weighting addition unit 343, and the weighting addition unit 343 converts the waveform data into Output waveform data is generated by weighting and adding the reproduction pitch using a weight that periodically changes with two pitches as one period.
[0021]
Here, the ring buffer memory 4 will be described. For example, when the first address is “0000” (hexadecimal) and the last address is “0FFF”, the ring buffer memory 4 is “0000”. ”To“ 0FFF ”is a memory in which a series of addresses conceptually continues in a ring shape, and the next address after“ 0000 ”is“ 0FFF ”.
[0022]
Next, the operation of the present embodiment will be described based on a flowchart and waveform diagrams. Before that, the gist of the present invention will be schematically described with reference to FIG.
FIG. 4 is a schematic explanatory diagram of the present invention and corresponds to FIG. 13 referred to when describing the problems of the conventional example. 4 (a) and 4 (b) are the same as FIGS. 13 (a) and 13 (b), respectively, and show input waveform data and its fundamental wave components, respectively. However, some symbols such as SS4 and SS5 are added.
[0023]
FIG. 4C is a waveform diagram obtained by the processing of the present invention. Conventionally, as described above, when the waveform reading is started, the reading is started from the reference point when the last pitch detection is established. However, in the present invention, the pitch when the pitch detection is finally established is started. Are sequentially added to the previous reference point (read address) to obtain the read reference point.
[0024]
First, at the time of S1 in FIG. 4C, the waveform data of the pitch a having the reference point S1 of the waveform in FIG. 4A is read. Then, the calculation of S1 + a is performed at the time of S2 in FIG. 4C, and the waveform data of the pitch b with the position of S1 + a (= S2) in FIG. 4A as the reference point for reading is read out. Similarly, the calculation of S2 + b is performed at the time of S3 in FIG. 4C, and the waveform data of the pitch b with the position of S2 + b (= S3) as the reference point in FIG. 4A is read.
[0025]
Next, in FIG. 4A, the position of SS4 is obtained by S3 + b, and the waveform data of the pitch b with the position of S3 + b (= SS4) as the reference point for reading is read. Similarly, in FIG. 4A, the SS5 position is obtained by SS4 + b, and waveform data having a pitch e with the position of SS4 + b (= SS5) as the reference point for reading is read. Further, similarly, in FIG. 4A, the position of SS6 is obtained by SS5 + e, and the waveform data of the pitch f with the position of SS5 + e (= SS6) as the reference point for reading is read.
[0026]
According to the above method, since the reference point for reading is updated according to the detection pitch, the fundamental wave component is not discontinuous and an output waveform with a natural waveform connection can be obtained. become.
Next, the operation of this embodiment will be described.
5 to 7 are waveform diagrams showing temporal changes in waveform data and functions created by a program described later in the present embodiment.
[0027]
In this embodiment, two variables, that is, a pitch change coefficient PITCH_VR and a formant change coefficient FORMAT_VR are input from the operator 7 (see FIG. 1). In this embodiment, the pitch change coefficient PITCH_VR can be arbitrarily set within the range of 0.5 to 2.0. For example, changing the value from 1.0 to 2.0 changes the pitch to 1. It means lowering the octave. Similarly, the formant change coefficient FORMAT_VR can be arbitrarily set within a range of 0.5 to 2.0. For example, changing the value from 1.0 to 2.0 means that the formant change coefficient It means to raise the shape one octave.
[0028]
5 shows PITCH_VR = 1.0 and FORMAT_VR = 1.0. FIG. 6 shows PITCH_VR> 1.0 and FORMAT_VR = 1.0. FIG. 7 shows PITCH_VR = 1.0. , FORMANT_VR> 1.0.
In FIG. 7, since the pitch P and WIDTH of the input waveform are the same as in FIG. 5, the input waveform is not shown in FIG.
[0029]
Hereinafter, the flowchart of the present embodiment will be described with reference to these drawings. Hereinafter, when any of FIGS. 5 to 7 may be referred to, FIG. 5 is typically referred to.
FIG. 8 is a flowchart showing the operation of the present invention. The operation shown in this flowchart is repeatedly executed at a predetermined sampling period.
[0030]
In the following description of the flowchart, for the sake of simplicity, the same name may be used without distinguishing between the name given to the buffer and the name assigned to the value stored in the buffer.
First, the writing means 31 shown in FIG. 2 sequentially writes sampling data, that is, input waveform data input at a predetermined sampling period, into the ring buffer memory 4 (step 8_1).
[0031]
At the same time, the pitch detection means 32 performs a pitch detection process on the input sampling data (step 8_2). The pitch detection process in this case is performed by the same technique as that disclosed in Japanese Patent Laid-Open No. 3-288200 described above. That is, the zero cross position or peak position of the input waveform is detected, and the time interval is compared with the time interval of the previously detected zero cross position or peak position to determine that the pitch is correctly detected. Sometimes the address and pitch of the zero-cross position that is the previous reference is output. Specifically, assuming that input waveform data as shown in FIG. 5A is input, when the zero cross point Z2 is detected, the pitch P0 from Z0 to Z1 and the pitch P1 from Z1 to Z2 are detected. Are determined to be equal to each other, and the zero-cross point Z1 and the detection pitch P1 are output.
[0032]
If it is determined in step 8_2 that the pitch has been detected, the pitch detection means 32 updates and holds the value of the pitch 1 period length (PITCH) in the buffer 32a named PITCH instead of the value detected last time. (Step 8_3). If no pitch is detected in step 8_2, the process immediately proceeds to step 8_4.
[0033]
The address update means 33 is based on the pitch 1 period length value (PITCH) currently held by the pitch detection means 32 and every time corresponding to the pitch 1 period length value (PITCH) has elapsed since the previous update. Then, the read reference address (SADRS) in the ring buffer memory 4 is updated (step 8_4). The update of the read reference address will be described later with reference to FIGS.
[0034]
When the read reference address is updated in step 8_4, the output waveform generation means 34 increments the PHASE, PH1, and PH2 functions shown in FIGS. 5B, 5D, and 5E, respectively (step S4). 8_5). Then, the magnitude relationship between PHASE and WIDTH (reproduction pitch) is compared (step 8_6). While PHASE remains at a value smaller than WIDTH, waveform readout processing is performed (step 8_16), and output waveform data is output. (Step 8_17). Further, when PHASE becomes WIDTH or more, PHASE is initialized to 0 (step 8_7).
[0035]
Then, the output waveform generation means 34 inputs parameters obtained from the result of the CPU 8 detecting the state of the operation element 7 (step 8_8). Here, the pitch change coefficient PITCH_VR and the formant change coefficient FORMAT_VR described above are input.
The output waveform generation means 34 calculates WIDTH and LENGTH based on the parameters obtained from the operator 7 (step 8_9). WIDTH is a period determined as a reproduction pitch, and LENGTH is a period half of the period during which waveform reading is actually performed. WIDTH can be obtained from the product of PITCH updated in step 8_3 and the pitch change coefficient PITCH_VR. LENGTH can be obtained by dividing PITCH by the formant change coefficient FORMAT_VR.
[0036]
Next, the output waveform generation means 34 compares the magnitude relationship between LENGTH and WIDTH (step 8_10), and when LENGTH is greater than WIDTH, sets the value of LENGTH to WIDTH (step 8_11). When LENGTH is less than or equal to WIDTH in step 8_10, the process proceeds directly to step 8_12. As described above, the reason that LENGTH does not always exceed WIDTH by steps 8_10 and 8_11 is that, if LENGTH exceeds WIDTH, a state in which WINDOW1 and WINDOW2 (described later) do not fall down to 0 occurs.
[0037]
The output waveform generation means 34 further performs W_RATE calculation and FLAG inversion operation (step 8_12). W_RATE is obtained by 1 / LENGTH. Then, the positive / negative of FLAG is discriminated (stainless steel 8_13). If positive, PH1 and WINDOW1 are set to 0 by the first reading means 34 shown in FIG. 3, and SADRS1 is set to SADRS updated in step 8_4. (Step 8_14). If negative, PH2 and WINDOW2 are set to 0 by the second reading means 342 shown in FIG. 3, and SADRS2 is set to SADRS updated in step 8_4 (step 8_15). By setting PH1 and PH2 to zero in steps 8_14 and 8_15, resetting is performed at the rising and falling points of FLAG (see FIG. 5C), and sawtooth waves whose phases are shifted from each other by 180 degrees are generated (see FIG. 5). 5 (d), (e)).
[0038]
The processing from step 8_7 to step 8_15 including the above steps 8_14 and 8_15 is performed only when PHASE ≧ WIDTH is satisfied, and thus is performed only at the rising edge and falling edge of FLAG. As described above, after the PH1, WINDOW1, PH2, and WINDOW2 are reset and the SADRS1 and SADRS2 are set in Steps 8_14 and 8_15, the output waveform generation unit 34 performs a waveform reading process from the ring buffer memory 4 (Step 8_16). ). Then, output waveform data is generated based on the read waveform data, and the output waveform data is output to the D / A converter 5 (step 8_17).
[0039]
Next, the update of the read address (SADRS) in step 8_4 will be described.
FIG. 9 is a conceptual diagram showing a relationship between a write address and a read address in the ring buffer memory 4.
The vertical axis in FIG. 9 indicates the address position from “0000” to [0FFF], and the horizontal axis indicates time. IBUF_PTR is an address at which writing is performed by the writing means 31, and ADRS exists with a delay of DELAY from this IBUF_PTR. ADRS is an address to be read from the ring buffer memory 4, and the DELAY is provided to prevent the read address from overtaking the write address. As the DELAY value, one cycle of the lowest frequency of the input waveform is usually set.
[0040]
Inside the ADRS, the read reference address (SADRS) changes stepwise along the ADRS. When SADRS gradually rises and comes before the final address [0FFF], an interval where ADRS <SADRS occurs. Then, the SADRS returns to the lower first address side, and again changes in a stepped manner along the ADRS.
[0041]
FIG. 10 is a flowchart of the update operation of the read reference address (SADRS).
First, the address updating unit 33 sets ADRS by subtracting DELAY from IBUF_PTR. Further, the logic of (1000 (hex) −1) is set so that ADRS is within the range of the first address from the first address. Take the product (step 10_1). Here, (hex) represents a hexadecimal number.
[0042]
Next, the magnitude relationship between ADRS and SADRS is compared (step 10_2). If SADRS does not exceed ADRS, the process proceeds to step 10_4. If SADRS exceeds ADRS, this means that SADRS has come before the final address and ADRS has returned to the first address side below, as described with reference to FIG. The ADRS is corrected to an address exceeding the final address so that the comparison of the magnitude relationship in the next step 10_4 can be performed normally (step 10_3).
[0043]
Thereafter, the address updating means 33 compares the magnitude relationship between ADRS and (SADRS + PITCH) (step 10_4). PITCH is updated from the pitch detection means 32 in step 8_3 of FIG. If ADRS is greater than or equal to (SADRS + PITCH), the logical product of (SADRS + PITCH) and (1000 (hex) -1) is taken, and (SADRS + PITCH) is updated as a new SADRS (step 10_5). ). Further, when ADRS is smaller than (SADRS + PITCH), updating is not performed and the current SADRS is used as it is.
[0044]
Next, the waveform reading process in step 8_16 in FIG. 8 will be described with reference to the flowchart in FIG.
First, the first reading unit 341 shown in FIG. 3 sets a value obtained by adding W_RATE to WINDOW1 as a new WINDOW1 (step 11-1). W_RATE is obtained by 1 / LENGTH in step 8_12. WINDOW1 is incremented by W_RATE and is not shown in FIG. 5, but is a sawtooth wave similar to PH1, and has a slope different from that of PH1. Then, the value of WINDOW1 becomes 1 after the LENGTH period, and the value of WINDOW1 becomes 2 after the period twice LENGTH.
[0045]
Next, the first reading means 341 determines the current value of WINDOW1 (step 11_2). If the value of WINDOW1 is less than 1, envelope ENV1 is set to WINDOW1 (step 11_3), and if WINDOW1 is 1 or more and less than 2, envelope ENV1 is set to (2-WINDOW1) (step 11_4). Steps 11_3 and 11_4 mean that ENV1 is obtained by folding WINDO1 which is a sawtooth wave by 1 and converting it into a triangular wave. If there is a part where WINDOW1 is 2 or more, ENV1 of that part is set to 0 (step 11_5).
[0046]
Next, an address ADRS1 for reading input waveform data from the ring buffer memory 4 is set (step 11_6). This ADRS1 is obtained by adding (PH1 * FORMAT_VR) to the read reference address SADRS. Based on this ADRS1, the waveform data DATA1 is read from the ring buffer memory 4 (step 11_7). Note that the read address ADRS1 is an address of a decimal point expression that does not actually exist, but performs a process of reading the waveform data DATA1 corresponding to the decimal point address by performing a known interpolation operation.
[0047]
The same operation as described above is performed on the WINDOW 2 side, that is, the second reading means 342 (steps 11_8 to 11_14). In the weighted addition means 343 shown in FIG. 3, DATA1 * ENV1 (see FIG. 5 (h)) and DATA2 * ENV2 (see FIG. 5 (i)) are respectively obtained from DATA1 and DATA2 obtained in steps 11_7 and 11_14. Then, DATA1 * ENV1 + DATA2 * ENV2 is calculated. This is output from the output waveform generation means 34 as output waveform data (step 11_15).
[0048]
By the way, as described in Step 10_1 of FIG. 10, when the read address ADRS is obtained, a calculation of subtracting the delay amount DELAY from the write address IBUF_PTR is performed. This DELAY value is used to prevent the read address (SADRS1, SADRS2) of the ring buffer memory 4 from overtaking the current write address (IBUF_PTR), and is usually one cycle of the lowest frequency of the input waveform. As described above, is set. However, if the minimum frequency of the input waveform is 80 Hz, a value corresponding to 12.5 milliseconds is set for DELAY, resulting in a delay of 12.5 milliseconds. This is a value that can be recognized as a delay, and is not preferable.
[0049]
Therefore, even when an input waveform having a low minimum frequency is input, in order to avoid the need to add such a large delay as described above, it is shown in FIG. 12 between steps 8_9 and 8_10 in FIG. It is preferable to insert the steps 9a and 9b. That is, after step 8_9, the magnitude relationship between LENGTH and pitch P (100 Hz) when the lowest frequency is 100 Hz is compared (step 9a). When LENGTH is larger than P (100 Hz), LENGTH is set to P (100 Hz) (step 9_b). In this way, by limiting the DELAY value to a certain value or less, it is possible to prevent a delay in reading from being recognized.
[0050]
【The invention's effect】
As described above, according to the present invention, while adopting a simple processing method for real-time processing as in the past, the occurrence of discontinuous portions of the fundamental component is prevented, and a natural output waveform is always obtained. be able to.
[Brief description of the drawings]
FIG. 1 is a block diagram showing a configuration of an embodiment of a waveform conversion apparatus of the present invention.
FIG. 2 is a block diagram showing a detailed configuration of a DSP.
FIG. 3 is a block diagram showing a detailed configuration of output waveform generation means.
FIG. 4 is a schematic explanatory diagram of the present invention.
FIG. 5 is a waveform diagram showing temporal changes in waveform data and functions in the embodiment of the present invention.
FIG. 6 is a waveform diagram showing temporal changes in waveform data and functions in the embodiment of the present invention.
FIG. 7 is a waveform diagram showing temporal changes in waveform data and functions in the embodiment of the present invention.
FIG. 8 is a flowchart showing the operation of the embodiment of the present invention.
FIG. 9 is a conceptual diagram showing a relationship between a write address and a read address of a ring buffer memory.
FIG. 10 is a flowchart of an update operation of a read reference address (SADRS).
FIG. 11 is a flowchart of a waveform read process.
12 is a diagram showing a flowchart to be added to the flowchart of FIG. 8 according to another embodiment of the present invention.
FIG. 13 is an explanatory diagram of problems in the prior art.
[Explanation of symbols]
1 Low-pass filter
2 A / D converter
3 DSP
4 Ring buffer memory
5 D / A converter
6 Low-pass filter
7 Controller
8 CPU
31 Writing means
32 Pitch detection means
33 Address change means
34 Output waveform generation means
341 First reading means
342 second reading means
343 Weighted addition means

Claims (2)

Detects the pitch of input waveform data that is sequentially input, cuts out a phoneme consisting of a predetermined data section corresponding to the detected pitch of the input waveform data, and uses the extracted phoneme as a pitch corresponding to predetermined playback pitch information In the waveform converter for converting to output waveform data having
A memory for temporarily storing waveform data;
Writing means for sequentially writing input waveform data sequentially input to the memory;
A first buffer for storing the pitch information in an updatable manner is provided. When the pitch detection process is executed for the input waveform data and the pitch detection is established, the pitch information stored in the first buffer is newly detected. Pitch detection means for updating the pitch information;
A second buffer for renewably storing a read reference address serving as a reference for extracting phonemes from the waveform data stored in the memory, and a time corresponding to the pitch information read from the first buffer last time; Each time, the pitch information is read from the first buffer, and the read reference address stored in the second buffer is updated by an address increment corresponding to the pitch information newly read from the first buffer. Address updating means;
Reproduction pitch information input means for inputting reproduction pitch information;
A read reference address is read from the second buffer, and a phoneme is cut out by reading waveform data in a predetermined data section from the memory based on the read reference address. An output waveform generating means for generating output waveform data by synthesizing at a period corresponding to the input reproduction pitch information. The output waveform generating means is
A read buffer stored in the third buffer has a third buffer to which a read reference address stored in the second buffer is transcribed each time output waveform data corresponding to two playback pitches is generated. First reading means for sequentially generating read addresses for two pitches of input waveform data based on a reference address, and sequentially reading the waveform data stored in the sequentially generated read addresses of the memory;
Output waveform data corresponding to two playback pitches is generated at a timing deviated by one playback pitch from the timing at which the read reference address stored in the second buffer is transferred to the third buffer. Each has a fourth buffer to which the read reference address stored in the second buffer is transcribed, and reads out two pitches of input waveform data based on the read reference address stored in the fourth buffer. Second reading means for sequentially generating addresses and sequentially reading the waveform data stored in the sequentially generated read addresses of the memory;
The waveform data read by the first reading means and the waveform data read by the second reading means are weighted and added using a weight that periodically changes with a playback pitch of 2 pitches. The waveform conversion apparatus according to claim 1, further comprising weighted addition means for generating output waveform data.

Images