JP2009124460A - Frequency converter - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To apply an arbitrary frequency change to an arbitrary frequency component in an acoustic signal by a simple process using an FIR filter, and further to achieve an operation of an amplitude frequency characteristic and an operation of a frequency at the same time. <P>SOLUTION: A signal processing part 2 for processing a digital signal from an AD converting part 1 has a filter part 4 and a memory part 5 for storing filter coefficient data. An FIR filter 40 and an FIR filter 41 included in the filter part 4 update a filter coefficient by data supplied from the memory part 5 during one sampling cycle. An amplitude of each frequency component of the respective FIR filters fluctuates independently with time. A cycle of an amplitude fluctuation is common to the FIR filter 40 and the FIR filter 41, but a phase is different at 90 degrees. The filter part 4 outputs the difference or total of output signals of the two FIR filters. <P>COPYRIGHT: (C)2009,JPO&INPIT

Description

  The present invention is a technique for performing arbitrary frequency conversion on a signal in a recording / playback device, a karaoke apparatus, an electronic musical instrument, etc., and can also be used as a frequency conversion technique used as a howling reduction method in a loudspeaker or a hearing aid. It is a signal processing technique that can also be used as part of a conversion device.

  In the field of handling audio signals, converting the pitch of audio is widely performed. Pitch conversion is a form of frequency conversion. For example, in a karaoke apparatus or the like, a signal processing technique for controlling a pitch of live performance or singing voice, that is, a pitch, is widespread. The pitch of the audio signal can be manipulated by changing the reproduction speed of the signal, but if the reproduction speed is changed, the length of time required for reproduction also changes. For this reason, a method of converting the pitch without changing the time length by thinning (deleting) or interpolating (inserting) data after expanding and contracting the voice waveform in the time axis direction (Patent Document 1), A method (Patent Document 2) is widely known in which a periodic waveform is cut out from a periodic portion of the waveform, expanded and contracted, and repeatedly recombined as many times as necessary.

However, if data that is originally provided is deleted or data that does not exist is inserted, an error from the original data occurs. In particular, when the signal frequency is high, errors due to thinning and interpolation become large. In the method of extracting and re-synthesizing periodic components, discontinuous points are formed in the connected portions, and noise is added to the reproduction signal.
In order to compensate for these drawbacks, a complicated signal processing such as a technique of using a superior signal interpolation technique or cross-fading after aligning the phases of two waveforms to be connected (Patent Documents 3 and 4) is required. .

As a method not involving an error or discontinuity in the time waveform, a frequency analysis such as FFT is performed, and after scaling operation is performed on the frequency axis, the time waveform is restored by inverse transformation (Patent Documents 5, 6, and 7). However, since the conversion from the time axis to the frequency axis and the conversion from the frequency axis to the time axis are required, the delay associated with the signal processing becomes relatively large.
In addition to the pitch conversion of the karaoke device, the frequency conversion of the audio signal can be used for a howling reduction method of a loudspeaker or a hearing aid (Patent Documents 9 and 10).

On the other hand, in the field of wireless communication, Hilbert transform or SSB (single sideband) modulation is often used as a carrier frequency conversion method (Patent Documents 10 and 11). This divides the signal into a real part I _(t) (In-phase component) and an imaginary part Q _(t) (Quadrature component) that is 90 degrees out of phase with the real part I _(t) and Q _(t). Amplitude modulation by (ωt) and sin (ωt) is performed, and Q _(t) is subtracted from the real part I _(t), and a frequency change corresponding to the frequency Δf (Hz) = ω / 2π is given to the input signal. (Non-Patent Document 1).
This method is called single-sideband modulation or single-sided modulation because it leaves only one of the sidebands generated on both sides of the signal frequency due to amplitude modulation. This can be realized by only two FIR filters for processing the imaginary part and amplitude modulation. In addition to wireless communication, attempts have been made to use it as an optical frequency conversion method (Patent Documents 12 and 13).
However, frequency conversion by general Hilbert transform is to move all frequency components of the input signal by a uniform frequency width on the frequency axis, and what is pitch conversion of sound that shifts frequency components at a constant ratio? Different.

  The Hilbert transform is also used to extract instantaneous frequencies such as speech and envelopes (Patent Documents 14, 15, and 16). As a technique for changing the pitch of the sound using the Hilbert transform, there is Patent Document 17, but this is a technique of dividing the band by a filter bank and changing the frequency of the carrier wave for each band, and as the number of bands is increased The amount of calculation increases. A method of manipulating only the phase by separating the absolute value (amplitude information) and phase information of the signal by Hilbert transform has been proposed (Patent Documents 18 and 19), but complicated processing such as extracting the angular velocity for each sample is proposed. It is necessary.

Further, when handling voice or music, manipulation of amplitude frequency characteristics, so-called equalization, is frequently performed. In the case of a pitch converter that operates the signal reproduction speed, an equalizer that operates the amplitude frequency characteristic is required in addition to the pitch converter in order to perform equalization.
Japanese Patent Laid-Open No. 9-212193 Japanese Patent Laid-Open No. 9-258777 Japanese Patent Laid-Open No. 2002-169556 Japanese Patent Laid-Open No. 5-97891 Japanese Patent Laid-Open No. 11-133996 JP 2006-64799 A JP-A-9-185392 JP 2007-258985 A JP-A-60-28399 JP-A-8-97751 JP 2007-67851 A JP 2007-11125 A JP 2004-85602 JP JP 2006-261787 A JP 2000-181472 A JP-A-5-316597 JP 2003-330500 A Japanese Patent Laid-Open No. 9-50293 JP 62-159196 A Naoki Mikami, “First time learning digital filters and fast Fourier transform,” CQ Publishing, 2005

As described above, in order to realize an audio pitch converter with little deterioration in sound quality, complicated signal processing is required, or delay due to frequency analysis is inevitable. Further, in the method using Hilbert transform for pitch conversion of speech or the like, complicated processing with band division and extraction of pitch information is required.
In view of the above problems, an object of the present invention is to provide a frequency converter and a pitch converter that can perform an arbitrary amount of frequency conversion on an arbitrary frequency component using only two FIR filters. Another object of the present invention is to provide a frequency converter that can perform an amplitude frequency characteristic operation (equalizing) simultaneously with a frequency operation by using an FIR filter.

  The frequency converter of the present invention includes a signal processing unit that processes a digital input signal, performs frequency conversion that gives a frequency change to the frequency component of the digital input signal, and outputs the result. The signal processing unit includes a filter unit and a filter coefficient data supply unit to be supplied to the filter unit. An imaginary part FIR filter is included. The real part FIR filter and the imaginary part FIR filter are time-varying filters, and the filter coefficient is updated with data supplied from the filter coefficient data supply unit during one sampling period, and is updated. The amplitude of each frequency component constituting each FIR filter varies independently with time, and the period of this amplitude variation is common to the real part FIR filter and the imaginary part FIR filter, but the phase is 90 degrees different. . The filter unit outputs a difference or sum of output signals of the real part FIR filter and the imaginary part FIR filter.

  In addition, an AD conversion unit that converts a time-series analog sound signal into a digital signal is input, the converted digital signal is input to the signal processing unit, and the digital signal output from the filter unit is A DA conversion unit for converting into a series of analog acoustic signals is provided. The period of amplitude fluctuation of each frequency component constituting the real part FIR filter and the imaginary part FIR filter is configured to be arbitrarily controllable.

Further, the characteristics of the real part FIR filter and the imaginary part FIR filter are characteristics in which at least two types of filter characteristics having different timings of amplitude fluctuation of the respective frequency components are smoothly cross-faded. The amplitude frequency characteristics of the real part FIR filter and the imaginary part FIR filter can be arbitrarily controlled. As a result, the filter part can be used as a low pass filter, a high pass filter, a band pass filter, a comb filter, or the like. The filter coefficient data supply unit includes a memory unit that supplies stored filter coefficient data, or a filter coefficient calculation unit that sequentially calculates filter coefficients.
In addition, the frequency converter of the present invention is used as a voice pitch converter in which the amplitude spectrum of the input signal is expanded or contracted on the frequency axis, or a speech speed converter that changes the reproduction speed without changing the pitch. Can do.

  According to the present invention, an input signal is passed through two FIR filters, and an arbitrary amount of frequency change is given to an arbitrary frequency component of the input signal by simple signal processing that simply obtains the difference or sum of the output signals of the two FIR filters. A frequency converter is obtained. Further, if the frequency change width is set to a constant ratio with respect to the frequency, a pitch converter that gives an arbitrary pitch change to the input signal can be obtained.

  An embodiment of the present invention will be described with reference to FIGS. FIG. 1 is a block diagram of a pitch converter according to the present invention. The input acoustic signal is digitized by the AD conversion unit 1 and sent to the signal processing unit 2. The signal processing unit 2 includes a filter unit 4 and a memory unit 5. The signal frequency-converted by the signal processing unit 2 is converted to an analog signal by the DA conversion unit 3. When mounted on a hearing aid or a loudspeaker, a signal from a microphone or the like is input to the AD conversion unit 1, and an output signal from the DA conversion unit 3 is sent to a transducer such as a speaker or a receiver. Further, if frequency conversion is performed on data that is originally digitized and recorded on various media or the like, the digital signal may be directly input to the signal processing unit 2 without going through the AD conversion unit 1.

  FIG. 2 is a diagram showing a configuration of the signal processing unit 2 in the present invention. The signal processing unit includes a filter unit 4 and a memory unit 5. The filter unit 4 includes a real part FIR (Finite Impulse Response) filter 40 and an imaginary part FIR filter 41. As filter coefficients of the FIR filter 40 and the FIR filter 41 are sequentially read from the memory unit, the filter characteristics change with time.

  The FIR filter 40 and the FIR filter 41 are time-varying filters composed of common frequency components whose phases are different by 90 degrees, and the filter coefficients are updated with data supplied from the memory unit 5 during one sampling period. By being updated, the amplitude of each frequency component constituting each FIR filter varies independently with time. This period of amplitude fluctuation is common to the FIR filter 40 and the FIR filter 41, but the phase is 90 degrees different. The filter unit 4 outputs a difference or sum of output signals from the FIR filter 40 and the FIR filter 41.

FIG. 3 is a configuration diagram of a general frequency converter using the Hilbert transform. In frequency conversion using the Hilbert transform, a common signal is input to the FIR filter 42 and the FIR filter 43. The amplitude frequency characteristics of the two FIR filters are common, but the phase is 90 degrees different at all frequencies. The transmitter 7 generates two signals cos (ω _(t) ) and sin (ω _(t) ) whose phases are different from each other by 90 degrees, and outputs cos (ω _{( t)} ) and amplitude modulation by sin (ω _(t) ) is performed. When the difference between the modulated signals is obtained, an output signal obtained by shifting the frequency of the input signal by Δf (Hz) = ω _(t) / 2π is obtained.

  The general frequency converter shown in FIG. 3 has a function of moving all frequency components of the input signal uniformly by Δf (Hz). When the FIR filter 42 and the FIR filter 43 are all-pass filters having a gain of 1 at all frequencies, this operation is as shown in FIG. That is, all frequency components of the input signal move on the frequency axis by Δf (Hz) while maintaining the amplitude.

On the other hand, FIG. 5 is a figure which shows the effect | action calculated | required by pitch converters, such as an audio | voice. As shown in FIG. 5, in the pitch conversion of voice or the like, conversion is performed such that the amplitude spectrum of the input signal is expanded or contracted on the frequency axis. FIG. 5 illustrates a case where the spectrum of the input signal is expanded on the frequency axis. At this time, the individual frequency components are not shifted by a uniform frequency width, but are frequency-converted at a constant ratio according to the frequency.
The present invention provides a signal processing method for realizing a pitch conversion action as shown in FIG. 5 and a frequency converter using the method with the same amount of computation as the frequency converter shown in FIG. .

The processing of the filter unit for realizing pitch conversion with an equal frequency change width will be described with reference to FIGS. In order to realize an independent frequency change for each frequency in the present invention, it is necessary to give independent amplitude modulation to each frequency component constituting the FIR filter. As a result, the characteristics of the filter change from moment to moment. If pitch conversion is to be performed, the amplitude modulation frequency is set to a constant ratio with respect to the frequency of the components constituting the filter. Therefore, if the sampling frequency is f _s , the amplitude modulation frequency applied to the frequency component f constituting the filter is Δf × f (Hz), and the filter tap length is n, the i-th coefficient I _i of the FIR filter at time t. Is obtained by Equation 1.
By setting the frequency of amplitude modulation to Δf × f, pitch conversion is performed so that the frequency change width is a constant ratio to f.

If this I _i is a filter coefficient for the real part, the filter coefficient Q _i for the imaginary part is
It is.

  FIG. 6 is an example of frequency characteristics of a pitch conversion FIR filter that changes with time. In this example, when the tap length is set to 128, it is shown how the amplitude of each frequency component constituting the filter changes as the time t advances from 0 samples to 255 samples. The horizontal axis of each graph represents the frequency component f (0 to 64) constituting the filter, and the vertical axis represents the amplitude of each component.

For example, the component of f = 64 is amplitude-modulated once (360 degrees) every time the time t advances by 4 samples, but the component of f = 32 is 1/2 times (180 degrees) even if t advances by 4 samples. Only amplitude modulated. Therefore, if the sampling frequency of the signal is 8 kHz, the frequency of the component of f = 64 is shifted by 8000/4 = 2000 Hz, and the frequency of the component of f = 32 is shifted by 1000 Hz, which is half of that. Since the component of f = 1 requires 256 samples to be amplitude-modulated once, the frequency change of 8000/256 = 31.25 Hz is given to the component of f = 1.
After the time t reaches 256, the characteristics from t = 0 to 255 can be used repeatedly. Since two sets of filter coefficient data are required for the real part and for the imaginary part, in this example, a memory capacity for storing 128 pieces of coefficient data × 256 sets × 2 is required. Since the filter coefficient data stored in the memory unit is amplitude-modulated in advance, the transmitter shown in FIG. 3 is not necessary.

In this way, pitch conversion is performed in which the frequency change width is equal to the frequency of the component constituting the filter. FIG. 6 shows the characteristic change of one FIR filter. If this is used for the real part, the characteristics of the FIR filter for the imaginary part used as a pair with this are all the same as the filter used for the real part. The phase is 90 degrees different at the frequency of.
However, this method using an FIR filter with a finite impulse response (tap length) has a limit. There is no problem if the signal has a frequency whose frequency is 1 / integer of the number of taps of the FIR filter. However, when pitch conversion is performed on a signal having a frequency other than that, sidebands are generated above and below the signal frequency. A method for solving this problem will be described with reference to FIGS.

FIG. 7-A shows a spectrum before processing (left) and a spectrum after processing when a 20% pitch conversion is performed using the filter coefficients obtained by Formula 1 and Formula 2 with a 2 kHz sine wave as input (Figure 7-A). (Right), FIG. 7-B shows a spectrum (left) before processing when a 20% pitch conversion is performed using a filter coefficient obtained by Equation 1 and Equation 2 using a 2.4 kHz sine wave as input. The spectrum after processing (right). Both an example in which the 16kHz sampling frequency _{f s,} a tap length n of the FIR filter 1024. Since the pitch conversion is 20%, Δf is
Since f _s = 16000 and n = 1024, Δf = 3.125 Hz.

  FIG. 8 shows a waveform obtained as a result of 20% pitch conversion performed on the 2.4 kHz sine wave using the filter coefficients obtained by Equations 1 and 2.

  7A, no large sideband is generated in the pitch-converted signal, but in FIG. 7B, a large sideband is generated around the pitch-converted signal. The waveform at this time is shown in FIG. When the sampling frequency is 16 kHz, the period of the 2 kHz sine wave is 8 samples, which is 1/128 of the tap length 1024 of the FIR filter. On the other hand, the period of the 2.4 kHz sine wave is 6.666... Sample, which is not an integral fraction of the tap length of the FIR filter. For this reason, the frequency component of 2.4 kHz is decomposed into a plurality of adjacent frequency components in the FIR filter. Since two adjacent frequency components are subjected to amplitude modulation of different frequencies, an error occurs. In the example of FIG. 7-B, the frequency of amplitude modulation differs between two adjacent frequency components by 3.125 Hz. Therefore, a phase difference of 360 degrees occurs every 5120 samples. Large sidebands do not occur when the phase difference of amplitude modulation is around 0 degrees and around 360 degrees, but large sideband waves occur when the phase difference of amplitude modulation is around 180 degrees.

When the filter coefficients obtained by the above formulas 1 and 2 are used, the waveform after the pitch conversion (FIG. 8) is influenced by sidebands at intervals of approximately 5120 samples. Assuming that the pitch converter used at this time is the first pitch converter, the second pitch converter in which the timing of the amplitude modulation applied to the FIR filter is shifted by 2560 samples is used to convert the 20% into a 2.4 kHz sine wave. When the pitch conversion process is performed, the waveform of FIG. 9 is obtained. At this time, the filter coefficient of the second pitch converter is obtained by the following equations 4 and 5.

  FIG. 9 is a waveform obtained by using the filter coefficient obtained as Δt = 2560 in Equations 4 and 5. As in FIG. 8, the influence of the sideband wave is generated at a constant interval, but the influence of the sideband wave is noticeable at a position shifted by 2560 samples in the waveforms of FIG. 8 and FIG. It is possible to reduce the influence of the sideband wave by performing a process of connecting only the part where the influence of the sideband wave is small from these two waveforms. That is, an FIR having a time-varying characteristic in which the characteristics of the first pitch converter using Expressions 1 and 2 and the second pitch converter using Expressions 4 and 5 are smoothly crossfade for each fixed period. A filter should be prepared. This is the third pitch converter.

  FIG. 10 shows a spectrum before processing (left) and a spectrum after processing (right) when a pitch conversion process of 20% is performed on a 2.4 kHz sine wave by the third pitch converter. A comparison of the right diagram in FIG. 7B and the right diagram in FIG. 10 clearly shows that the sidebands are reduced.

  An example of an audio signal that has been subjected to pitch conversion processing by the third pitch converter is shown in FIG. The spectrum of the original speech (upper) and the spectrum after increasing the pitch by 20% (lower). As shown by the arrow, it can be seen that the peak that was in the vicinity of 2.5 kHz has moved to near 3 kHz by 20% pitch conversion, and that no sideband that significantly deteriorated the sound quality has occurred.

  6 to 11, the pitch conversion method in which the frequency change width is set to a constant ratio with respect to the frequency has been described. However, the present invention makes it possible to provide an independent frequency change amount for each frequency. . Therefore, in addition to pitch conversion, for example, it can be used for special purposes such as changing the frequency of only a specific band component, or changing the frequency of the low range lower and the high range higher, such as piano tuning. Is also effective.

  If the present invention is used, pitch conversion is possible by using only two FIR filters for the real part and the imaginary part without extracting the phase information of the input signal as in the existing technology (Patent Documents 14 and 15). become. In order to obtain the characteristics of the time-varying filter described above, it is desirable to store the filter coefficients created in advance in the memory unit and read them sequentially to update the filter coefficients. Alternatively, a new filter coefficient may be obtained and updated every time. The configuration in this case is as shown in FIG. This is obtained by replacing the memory unit 5 in FIG.

  Further, since the FIR filter is used, the amplitude frequency characteristic can be freely set simultaneously with the frequency conversion. The filter coefficients obtained by Expression 1, Expression 2, Expression 4, and Expression 5 can be used as a low-pass filter, a high-pass filter, a band-pass filter, a comb filter, etc. by applying an arbitrary coefficient for each frequency.

  The main application of the present invention is to change the pitch without changing the playback speed, and to adjust the pitch to singing voice and live performance data, or as part of the method of reducing howling of hearing aids and loudspeakers It is. However, this is also effective as so-called speech speed conversion means for changing the playback speed without changing the pitch. In this case, it may be used as means for returning the pitch changed by changing the reproduction speed to the original height. For example, when listening to data recorded by a voice recorder or the like in a shorter time than when recording, it is possible to keep the pitch unchanged even when played back at a high speed.

  Although the basic embodiments of the present invention have been described above, the present invention is not limited to these embodiments. Appropriate changes can be made without departing from the scope of the present invention.

Configuration example of frequency converter according to the present invention Configuration example of signal processing unit in the present invention Diagram showing the configuration of a general frequency converter using the Hilbert transform Schematic diagram of constant width frequency conversion processing with constant frequency variation Schematic diagram of pitch conversion processing with frequency change width at a constant ratio to frequency The figure which shows the characteristic fluctuation | variation of the time-varying FIR filter used with the filter part of this invention. 7-A is the spectrum of the 2 kHz sine wave signal (left), and the spectrum of the signal obtained by applying 20% pitch conversion to the 2 kHz sine wave (right), and 7-B is the 2.4 kHz sine wave. Wave signal spectrum (left) and 2.4 kHz sine wave signal spectrum obtained by applying 20% pitch conversion using filter coefficients according to equations 1 and 2 (right) Waveform of the signal obtained as a result of 20% pitch conversion using 2.4 kHz sine wave with filter coefficients according to Equation 1 and Equation 2. Waveform of signal obtained as a result of 20% pitch conversion using 2.4 kHz sine wave with filter coefficients of equations 4 and 5 A spectrum of a signal obtained as a result of 20% pitch conversion using a filter coefficient obtained by cross-fading the filter coefficient according to expressions 1 and 2 and the filter coefficient according to expressions 4 and 5 to a 2.4 kHz sine wave. Comparison of speech spectrum before pitch conversion and speech spectrum after processing Configuration example of frequency converter using filter coefficient calculation unit instead of memory unit

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 AD conversion part 2 Signal processor 3 DA conversion part 4 Filter part 40 FIR filter 41 FIR filter 42 FIR filter 43 FIR filter 5 Memory part 6 Filter coefficient calculating part 7 Transmitter

Claims (7)

  A frequency converter that has a signal processing unit that processes a digital input signal, and that performs frequency conversion that gives a frequency change to the frequency component of the digital input signal and outputs the frequency component. The signal processing unit includes a filter unit and the filter unit A filter coefficient data supply unit that supplies a real part FIR filter and an imaginary part FIR filter composed of common frequency components having phases different from each other by 90 degrees. The FIR filter for imaginary part and the FIR filter for imaginary part are time-varying filters, and the filter coefficients are updated with the data supplied from the filter coefficient data supply unit during one sampling period, and each of them is updated. The amplitude of each frequency component constituting the FIR filter varies independently with time, and the period of the amplitude variation is the FIR for real part. Although it is common to the filter and the imaginary part FIR filter, the phase is 90 degrees different, and the filter part outputs the difference or sum of the output signals of the real part FIR filter and the imaginary part FIR filter. To frequency converter.   An AD conversion unit that converts a time-series analog sound signal into a digital signal, inputs the converted digital signal to the signal processing unit, and converts the digital signal output from the filter unit into a time-series The frequency converter of Claim 1 provided with the DA converter part converted into an analog acoustic signal.   3. The frequency converter according to claim 1, wherein an amplitude variation period of each frequency component constituting the real part FIR filter and the imaginary part FIR filter can be arbitrarily controlled.   The characteristics of the real part FIR filter and the imaginary part FIR filter are characteristics obtained by smoothly cross-fading at least two types of filter characteristics having different timings of amplitude fluctuation of each frequency component. The frequency converter described in 1.   The amplitude frequency characteristics of the real part FIR filter and the imaginary part FIR filter can be arbitrarily controlled, and as a result, the filter part can be used as a low pass filter, a high pass filter, a band pass filter, a comb filter, or the like. The frequency converter in any one of 1-4.   6. The frequency converter according to claim 1, wherein the filter coefficient data supply unit includes a memory unit that supplies stored filter coefficient data, or a filter coefficient calculation unit that sequentially calculates filter coefficients.   7. The voice speed converter for converting the amplitude spectrum of the input signal to be expanded or contracted on the frequency axis, or the speech speed converter for changing the playback speed without changing the pitch. Frequency converter.