JP5821584B2 - Audio processing apparatus, audio processing method, and audio processing program - Google Patents

Description

  The present invention relates to a voice processing device, a voice processing method, and a voice processing program.

  Conventionally, for example, when filtering is performed on an audio signal, filtering in the frequency domain has been widely performed. This is because the amount of calculation required for the filter processing is small compared to the filter processing in the time domain.

  For example, in the filter processing in the frequency domain, a plurality of analysis frames are cut out from the audio signal in order to convert the audio signal in the time domain into the frequency domain. Here, when the extracted analysis frames are added, the ends of the analysis frames are not connected continuously, and noise is included in the original audio signal. For this reason, for example, a method called overlap addition is used in order to make the end of the analysis frame continuous and bring the sound quality closer to the original audio signal. This is, for example, a method of multiplying each analysis frame by a window function and adding each analysis frame so that the frame lengths overlap each other by 50%.

  Further, for example, as an improvement method of the above overlap addition, it is also proposed that the analysis frame overlap ratio is 87.5%. The rate at which the analysis frames overlap is called the overlap rate.

JP 2004-20679 A

Rose Curtis "Computer Music History, Technology, Art" Tokyo Denki University Press January 2001

  However, the above-described conventional technique has a problem in that the amount of calculation for the filter processing increases. For example, if the overlap ratio is increased to 50%, 75%, 87.5%, etc. in order to improve the sound quality, the calculation amount of the filter processing increases to 2 times, 4 times, 8 times, etc. End up. For this reason, for example, when filtering a low-quality sound source or when performing filtering in a noisy situation, even if the overlap ratio is increased more than a certain level, the sound quality is hardly improved. In some cases, only the amount of calculation increases.

  The disclosed technology has been made in view of the above, and an object thereof is to provide an audio processing device, an audio processing method, and an audio processing program capable of suppressing the amount of calculation of filter processing.

The technique which this application discloses is provided with a filter process part and a detection part in one mode. The filter processing unit performs frequency domain filter processing on the input signal using window function processing in which analysis frames overlap at a predetermined rate. Each time the detection unit increases the rate of overlap of the analysis frames, the detection unit calculates the similarity between the signal after the filtering process and the signal based on the input signal, and based on the calculated similarity, The ratio set in the filter processing unit is detected.

  According to one aspect of the technology disclosed in the present application, there is an effect that the amount of calculation of the filter processing can be suppressed.

FIG. 1 is a block diagram illustrating a functional configuration of the speech processing apparatus according to the first embodiment. FIG. 2 is a diagram for explaining an example of a signal flow in the sound processing apparatus according to the first embodiment. FIG. 3 is a diagram for explaining the overlap. FIG. 4 is a diagram for explaining the overlap. FIG. 5 is a diagram for explaining the overlap. FIG. 6 is a diagram for explaining a process of comparing the filtered sound and the original sound by 50% overlap addition. FIG. 7 is a diagram for explaining processing for comparing the filtered sound and the original sound by 75% overlap addition. FIG. 8 is a diagram for explaining processing for comparing the filtered sound and the original sound by 87.5% overlap addition. FIG. 9 is a diagram for explaining the relationship between the overlap ratio and the determination coefficient. FIG. 10 is a flowchart illustrating the processing procedure of the speech processing apparatus according to the first embodiment. FIG. 11 is a block diagram illustrating a functional configuration of the speech processing apparatus according to the second embodiment. FIG. 12 is a diagram for explaining an example of a signal flow in the sound processing apparatus according to the second embodiment. FIG. 13 is a flowchart illustrating the processing procedure of the sound processing apparatus according to the second embodiment. FIG. 14 is a diagram for explaining the relationship between the determination coefficient and the sound quality. FIG. 15 is a diagram for explaining the relationship between the determination coefficient and the sound quality. FIG. 16 is a flowchart illustrating the processing procedure of the speech processing apparatus according to the third embodiment. FIG. 17 is a flowchart illustrating the processing procedure of the speech processing apparatus according to the third embodiment. FIG. 18 is a diagram illustrating an example of a computer that executes a voice processing program.

  Hereinafter, embodiments of a voice processing device, a voice processing method, and a voice processing program disclosed in the present application will be described in detail with reference to the drawings. Note that the present invention is not limited to the embodiments. Each embodiment can be appropriately combined within a range in which processing contents do not contradict each other.

  An example of a functional configuration of the speech processing apparatus according to the first embodiment will be described. FIG. 1 is a block diagram illustrating a functional configuration of the speech processing apparatus according to the first embodiment. As illustrated in FIG. 1, the speech processing apparatus 100 includes a signal acquisition unit 110, a calculation unit 120, a filter processing unit 130, and a ratio determination unit 140. The audio processing device 100 corresponds to, for example, a mobile terminal device such as a mobile phone, a smartphone, or a PHS (Personal Handy-phone System) terminal, or an audio playback device such as a CD (Compact Disk) player or a digital audio player. The audio processing device 100 is connected to the speaker 10 and the microphone 20. The speaker 10 is a device that outputs sound around the speaker 10. The microphone 20 is a device that collects sounds around the microphone 20. In addition, although the case where the speaker 10 and the microphone 20 are connected to the audio processing device 100 as external devices has been described here, the present invention is not limited to this. For example, the speaker 10 and the microphone 20 may be built in the audio processing device 100.

  The processing functions performed by the signal acquisition unit 110, the calculation unit 120, the filter processing unit 130, and the ratio determination unit 140 are realized as follows. In other words, all or a part of these processing functions can be realized by a CPU (Central Processing Unit) and a program analyzed and executed by the CPU, or can be realized as hardware by wired logic. .

  FIG. 2 is a diagram for explaining an example of a signal flow in the sound processing apparatus according to the first embodiment. Each processing function shown in FIG. 2 corresponds to each processing function having the same reference numeral shown in FIG. The signal flow in the voice processing device will be described together with each processing function of the voice processing device 100.

  The signal acquisition unit 110 acquires an audio signal from the sound source and outputs the acquired audio signal. For example, the signal acquisition unit 110 acquires an audio signal from an antenna that receives FM (Frequency Modulation) radio. For example, the signal acquisition unit 110 acquires an audio signal from a memory in which audio data is stored. For example, as illustrated in FIG. 2, the signal acquisition unit 110 outputs the acquired audio signal to the speaker 10, the calculation unit 120, the window function processing unit 131, and the detection unit 141. When the acquired audio signal is an analog signal, the signal acquisition unit 110 converts the analog signal into a digital signal and outputs the digital signal. In the following description, the audio signal output from the signal acquisition unit 110 is also referred to as “original sound”.

  The calculation unit 120 calculates a FIR (Finite Impulse Response) filter. For example, the calculation unit 120 calculates a filter count X (s) to be applied to the FIR filter 133 described later. For example, the calculation unit 120 calculates the filter count X (s) using the audio signal output from the signal acquisition unit 110 and the signal collected by the microphone 20, and calculates the calculated filter count X (s). Set to FIR filter 133. For example, as illustrated in FIG. 2, the calculation unit 120 outputs information for setting the calculated filter count X (s) in the FIR filter 133 to the FIR filter 133.

  Here, a process in which the calculation unit 120 calculates the filter count X (s) will be described. For example, the calculation unit 120 calculates an impulse response H (s) that is an acoustic characteristic from the speaker 10 to the microphone 20. For example, the calculation unit 120 converts the audio signal output from the signal acquisition unit 110 and the signal collected by the microphone 20 into frequency domains. The calculation unit 120 calculates the impulse response H (s) from the ratio between the audio signal output from the signal acquisition unit 110 in the frequency domain and the signal collected by the microphone 20.

For example, the calculation unit 120 calculates the inverse characteristic X (s) = H (s) ⁻¹ of the calculated impulse response H (s). The calculation unit 120 sets the calculated inverse characteristic X (s) as the filter count X (s) of the FIR filter 133. Here, the reverse characteristic of the acoustic characteristics from the speaker 10 to the microphone 20 is used as the filter count because the original sound can be reproduced at the position of the microphone 20. It should be noted that X (s) H (s) = 1 is not guaranteed due to the effect of preventing zeroing when calculating the filter count X (s).

  For example, the filter processing unit 130 performs a frequency domain filtering process on the input signal using window function processing in which analysis frames overlap at a predetermined rate. For example, the filter processing unit 130 receives an original sound as an input signal, and executes a frequency domain filter process using window function processing in which analysis frames overlap at a predetermined rate with respect to the received original sound. The filter processing unit 130 includes, for example, a window function processing unit 131, a conversion unit 132, an FIR filter 133, an inverse conversion unit 134, and an addition unit 135.

  For example, the window function processing unit 131 performs window function processing on an audio signal in the time domain. For example, the window function processing unit 131 performs window function processing in which analysis frames overlap each other at a predetermined rate with respect to the audio signal, and cuts out a plurality of analysis frames from the time domain audio signal. For example, as shown in FIG. 2, the window function processing unit 131 outputs a plurality of cut out analysis frames to the conversion unit 132.

  For example, the window function processing unit 131 receives an audio signal output from the signal acquisition unit 110. The window function processing unit 131 multiplies the received audio signal by a Hanning window so that adjacent analysis frames overlap at a rate set by the setting unit 142 described later, and cuts out a plurality of analysis frames. For example, the window function processing unit 131 sets the Hanning window so that adjacent analysis frames overlap each other by 50%. The window function processing unit 131 outputs the extracted plurality of analysis frames to the conversion unit 132. Although the case where the window function processing unit 131 uses a Hanning window as the window function has been described here, the present invention is not limited to this. The Hanning window has the property that the end of the window function has converged to 0 and that it is exactly 1 when shifted by 50%, and is suitable for suppressing noise generated at the window end. For example, if the window function has the above properties, the window function processing unit 131 may use another window function such as a Bartlett window as the window function.

  Here, the overlap will be described with reference to FIGS. 3 to 5 are diagrams for explaining the overlap. FIG. 3 shows an example of a waveform of an audio signal in the time domain. In FIG. 3, the horizontal axis indicates time [ms], and the vertical axis indicates amplitude. For example, when a signal of one sample acquired by 8 kHz sampling is expressed by 16 bits, the range of the value is −32767 to +32768. FIG. 3 shows a case where the analysis frame 1a, the analysis frame 1b, and the analysis frame 1c are cut out from the audio signal in the time domain. The analysis frame 1a and the analysis frame 1c overlap 50%, and the analysis frame 1b and the analysis frame 1c overlap 50%. In addition, 8 kilohertz sampling represents sampling every 1/8000 second.

  FIG. 4 shows a case where analysis frames that do not overlap are added. The signal of the analysis frame 2a in FIG. 4 indicates a signal that has been subjected to frequency domain filtering on the signal of the analysis frame 1a and then inversely transformed to the time domain. The signal of the analysis frame 2b indicates a signal that has been subjected to frequency domain filtering on the signal of the analysis frame 1b and then inversely transformed to the time domain. As shown in FIG. 4, even if the analysis frame 2a and the analysis frame 2b are added, a gap 2c exists between the waveforms. For this reason, the ends of the analysis frames are not continuously connected, and noise is included in the original audio signal.

  FIG. 5 shows a case where overlapping analysis frames are added. The signal of the analysis frame 3a in FIG. 5 indicates a signal that has been subjected to frequency domain filtering after the Hanning window is applied to the signal of the analysis frame 1a and then inversely transformed to the time domain. The signal of the analysis frame 3b indicates a signal that has been subjected to frequency domain filtering after the Hanning window is applied to the signal of the analysis frame 1b and then inversely transformed to the time domain. The signal of the analysis frame 3c indicates a signal that has been subjected to frequency domain filtering after the Hanning window is applied to the signal of the analysis frame 1c and then inversely transformed to the time domain. As shown in FIG. 5, when the analysis frame 3a multiplied by the window function and the analysis frame 3b are added, there is no gap at the end of the waveform. For this reason, the end of the analysis frame is continuously connected. Further, the waveform suppressed by the window function can be compensated by further adding the analysis frame 3c that overlaps the analysis frame 3a and the analysis frame 3b by 50%. Note that the method of adding so that adjacent analysis frames overlap in this way is referred to as “overlap addition”, and the overlapping ratio is referred to as “overlap ratio”. Further, for example, adding overlap at a 50% overlap ratio is also referred to as “50% overlap addition”.

For example, the window function processing unit 131 performs window function processing on the audio signal at a predetermined overlap rate. For example, the window function processing unit 131 multiplies the audio signal with a Hanning window at an overlap ratio of (1-1 / 2 ⁿ ) × 100%. Note that n is an integer equal to or greater than 1, and is incremented by 1 each time the filter process is executed, for example. That is, each time the filter process is executed, the window function processing unit 131 executes the window function process on the audio signal while gradually increasing the overlap ratio to 50%, 75%, 87.5%,. . Note that the overlap ratio used by the window function processing unit 131 for the window function processing is not limited to the above example. For example, the window function processing unit 131 may store an integer n of 1 or more and an arbitrary overlap ratio in association with each other, and may select an overlap ratio corresponding to n.

  Returning to the description of FIG. For example, the conversion unit 132 performs Fourier transform on the audio signal. For example, the conversion unit 132 performs a Fourier transform on the audio signal for each of the plurality of analysis frames output by the window function processing unit 131. For example, as illustrated in FIG. 2, the conversion unit 132 outputs an audio signal subjected to Fourier transform to the FIR filter 133.

  For example, the FIR filter 133 performs frequency domain filtering on the audio signal. For example, the FIR filter 133 receives the audio signal that has been Fourier-transformed by the conversion unit 132. The FIR filter 133 performs filter processing for each analysis frame with the filter count X (s) set by the calculation unit 120. For example, as shown in FIG. 2, the FIR filter 133 outputs the audio signal subjected to the filter process to the inverse conversion unit 134. Although the case where the filter processing unit 130 uses the FIR filter 133 for reproducing the original sound at the position of the microphone 20 has been described here, the present invention is not limited to this. For example, instead of the FIR filter 133, the filter processing unit 130 may use a low-pass filter for suppressing noise included in the audio signal.

  For example, the inverse transform unit 134 performs inverse Fourier transform on the audio signal. For example, the inverse transform unit 134 performs inverse Fourier transform on the audio signal for each of the plurality of analysis frames output by the FIR filter 133. For example, as illustrated in FIG. 2, the inverse transform unit 134 outputs an audio signal obtained by performing the inverse Fourier transform to the adder unit 135.

  The adding unit 135 adds, for example, audio signals in the time domain. For example, the adder 135 receives the audio signal for each analysis frame that has been subjected to inverse Fourier transform by the inverse transformer 134, and overlaps the received audio signal. For example, as illustrated in FIG. 2, the adding unit 135 outputs the added audio signal to the speaker 10 and the detecting unit 141.

Thus, for example, the filter processing unit 130 performs window function processing on the audio signal at an overlap ratio of (1-1 / 2 ⁿ ) × 100%, and performs frequency domain filter processing. The filter processing unit 130 outputs the audio signal that has been subjected to the filter processing. The audio signal output by the filter processing unit 130 is also referred to as “filtered sound”. That is, the filter processing unit 130 outputs the filtered sound to the speaker 10 and the detection unit 141.

  For example, the ratio determining unit 140 determines an overlap ratio. For example, the ratio determining unit 140 determines the overlap ratio to be set in the filter processing unit 130. For example, the ratio determination unit 140 includes a detection unit 141 and a setting unit 142.

  For example, each time the overlap ratio is increased, the detection unit 141 calculates the correlation between the filtered sound and the original sound, and detects the overlap ratio set in the filter processing unit 130 based on the calculated correlation. For example, the detection unit 141 calculates a ratio between the calculated correlation and the previously calculated correlation among the calculated correlations, and when the calculated ratio is less than the threshold, the ratio when the previously calculated correlation is calculated Is detected. For example, the detection unit 141 outputs the detected overlap ratio to the setting unit 142 as illustrated in FIG.

  Hereinafter, processing of the detection unit 141 will be described. For example, the detection unit 141 compares the filtered sound output from the filter processing unit 130 with the original sound output from the signal acquisition unit 110. For example, the detection unit 141 compares the filtered sound by the 50% overlap addition with the original sound.

  FIG. 6 is a diagram for explaining a process of comparing the filtered sound and the original sound by 50% overlap addition. The horizontal axis of FIG. 6 shows the amplitude of the waveform of the original sound in the time domain, and the vertical axis shows the amplitude of the waveform of the filtered sound in the time domain. Each plot in FIG. 6 is a plot of the amplitude value of the sample in the waveform of the original sound and the amplitude value of the corresponding sample in the waveform of the filtered sound. That is, for example, a plot with a horizontal axis value of 8000 and a vertical axis value of 8020 indicates that the amplitude value of a predetermined sample in the waveform of the original sound has shifted from 8000 to 8020 by the filtering process.

For example, as illustrated in FIG. 6, the detection unit 141 compares the similarity between the filtered sound and the original sound by 50% overlap addition. For example, the detection unit 141 calculates a determination coefficient as the similarity between the filtered sound and the original sound based on the following equation (1). Incidentally, the coefficient of determination is both ^{R 2} notation. In this calculation, the determination coefficient is used as the similarity, but other means for calculating the similarity include the Euclidean distance, and other methods may be used.

In equation (1), i represents the number of samplings. y represents the sample value, that is, the amplitude value of the filtered sound. y _i indicates the sample value of the i-th sample. f _i indicates an estimated value based on a regression equation. Here, if the amplitude value of the original sound and x, for example, in FIG. 6, the estimated value f and the coefficient of determination R ² is as follows.
f = 1.000320669x + 0.081344144
R ² = 0.994887318

For example, the detection unit 141, by calculating the coefficient of determination R ² and filtering sound and original sound by 50% overlap-add, compare both. Incidentally, the coefficient of determination R ² are as waveform filtering sound is similar to the waveform of original sound, which is a value close to 1. The coefficient of determination R ² is an example of a "correlation."

Further, for example, the detection unit 141 determines the determination coefficient when compared with the original sound for the 75% overlap addition and the 87.5% overlap addition for the filter process sound by the 50% overlap addition. to calculate the R ^2.

FIG. 7 is a diagram for explaining processing for comparing the filtered sound and the original sound by 75% overlap addition. The description of FIG. 7 is the same as the description of FIG. As illustrated in FIG. 7, for example, the detection unit 141 calculates the determination coefficient R ² between the filtered sound and the original sound by 75% overlap addition and the estimated value f based on the above equation (1). . 7, the estimated value f and the coefficient of determination R ² is as follows.
f = 0.999553234x + 0.3935565240
R ² = 0.999279900

FIG. 8 is a diagram for explaining processing for comparing the filtered sound and the original sound by 87.5% overlap addition. The description of FIG. 8 is the same as the description of FIG. As shown in FIG. 8, for example, the detection unit 141 obtains the determination coefficient R ² between the filtered sound and the original sound by the 87.5% overlap addition and the estimated value f based on the above equation (1). calculate. 8, the estimated value f and the coefficient of determination R ² is as follows.
f = 0.9994844307x + 0.634147404
R ² = 0.999097137

That is, when n = 1, the detection unit 141 calculates the determination coefficient R ² = 0.9994873018 of the filtered sound and the original sound by 50% overlap addition. In addition, when n = 2, the detection unit 141 calculates a determination coefficient R ² = 0.9999279900 between the filtered sound and the original sound by 75% overlap addition. When n = 1, the detection unit 141 calculates a determination coefficient R ² = 0.9999097137 between the filtered sound and the original sound by 87.5% overlap addition.

  Further, for example, the detection unit 141 compares the determination coefficient at the time of n−1 with the determination coefficient at the time of n, and determines whether or not it has increased by a certain percentage or more. If it has increased by a certain percentage or more, the detection unit 141 increments n set in the filter processing unit 130 by one. On the other hand, if it has not increased by a certain ratio or more, the detection unit 141 outputs the overlap ratio at the time of n−1 to the setting unit 142. Here, for example, 0.1% is set as the fixed ratio.

  For example, when n = 2, the detection unit 141 calculates (0.9999279900−0.994887318) /0.9994873018×100=0.443%. Since this value is 0.1% or more, the detection unit 141 increments n set in the filter processing unit 130 by one.

  For example, when n = 3, the detection unit 141 calculates (0.9999097137−0.9999279900) /0.9999279900×100=−0.018%. Since this value is less than 0.1%, the detection unit 141 outputs the overlap ratio “75%” when n = 2 to the setting unit 142.

  Here, the reason why the detection unit 141 detects the overlap ratio at which the determination coefficient is not improved is that even if the overlap ratio is further increased, only the amount of calculation for the filter processing increases. FIG. 9 is a diagram for explaining the relationship between the overlap ratio and the determination coefficient. The horizontal axis in FIG. 9 indicates the overlap ratio [%], and the vertical axis indicates the determination coefficient. FIG. 9 shows a case where the determination coefficients of FIGS. 6 to 8 are plotted. As shown in FIG. 9, when the overlap ratio is increased, the ends of adjacent analysis frames are smoothly connected, so that the determination coefficient, that is, the sound quality is improved. However, when the overlap ratio exceeds a predetermined value, the determination coefficient is not improved. This suggests that the ends of adjacent analysis frames are sufficiently smoothly connected by the overlap ratio at this time. On the other hand, when the overlap ratio is increased to 50%, 75%, 87.5%,..., The calculation amount of the filter processing increases to 2 times, 4 times, 8 times,. The overlap ratio at which the coefficient of determination is not improved depends on the sound quality of the original sound and the environment around the microphone 20. Therefore, by applying the overlap ratio detected by the detection unit 141 to the window function process, it is possible to prevent an increase in the amount of calculation related to the filter process.

  As described above, the detection unit 141 calculates the correlation between the filtered sound and the original sound every time the overlap ratio is increased. The detection unit 141 detects the overlap ratio set in the filter processing unit 130 based on the calculated correlation. Here, the method of detecting the overlap ratio by calculating the ratio between the correlation calculated this time by the detection unit 141 and the previously calculated correlation has been described, but the present invention is not limited to this. For example, the detection unit 141 may detect an overlap ratio when the calculated correlation is equal to or greater than a threshold value.

  Returning to the description of FIG. For example, the setting unit 142 sets the overlap ratio detected by the detection unit 141 in the filter processing unit 130. For example, the setting unit 142 receives the overlap ratio from the detection unit 141 and sets the received overlap ratio in the window function processing unit 131. In the example illustrated in FIG. 9, the setting unit 142 receives the overlap ratio “75%” when n = 2 from the detection unit 141 and sets the received overlap ratio in the window function processing unit 131. For example, the setting unit 142 outputs information for setting the overlap ratio to the window function processing unit 131 as shown in FIG.

  Next, a processing procedure of the speech processing apparatus 100 according to the first embodiment will be described. FIG. 10 is a flowchart illustrating the processing procedure of the speech processing apparatus according to the first embodiment. The process illustrated in FIG. 10 is executed at a predetermined interval while power is supplied from the power supply, for example, in the audio processing apparatus 100.

  As shown in FIG. 10, when the processing timing comes (step S101, Yes), the signal acquisition unit 110 outputs the original sound from the speaker 10 (step S102). Until the processing timing comes (No in step S101), the processing illustrated in FIG. 10 is in a standby state.

  The microphone 20 collects the original sound output from the speaker 10 (step S103). The calculation unit 120 calculates the FIR filter 133 (step S104). That is, the calculation unit 120 calculates the filter count X (s) applied to the FIR filter 133.

The filter processing unit 130 sets n = 1 (step S105). The filter processing unit 130 calculates a filter processing sound obtained by executing the window function processing at an overlap ratio of (1-1 / 2 ⁿ ) × 100% (Step S106).

  The ratio determining unit 140 compares the filtered sound with the original sound and calculates a determination coefficient (step S107). If n = 1 is not satisfied (step S108, No), the ratio determination unit 140 compares the determination coefficient at the time of n-1 with the determination coefficient at the time of n, and determines whether or not it has increased by a certain ratio or more. Is determined (step S109).

  If it has not increased by more than a certain ratio (No at Step S109), the ratio determining unit 140 sets the overlap ratio at the time of n-1 in the filter processing unit 130 (Step S111).

  On the other hand, if it has increased by a certain ratio or more (step S109, Yes), the ratio determination unit 140 increments n set in the filter processing unit 130 by 1 (step S110), and proceeds to step S106.

  On the other hand, when n = 1 (step S108, Yes), the ratio determination unit 140 increments n set in the filter processing unit 130 by 1 (step S110), and proceeds to step S106.

  Next, effects of the sound processing apparatus 100 according to the first embodiment will be described. The speech processing apparatus 100 performs frequency domain filtering on the input signal using window function processing in which analysis frames overlap at a predetermined rate. The speech processing apparatus 100 calculates the degree of similarity between the signal after the filter process is executed and an arbitrary signal each time the ratio of overlapping analysis frames is increased, and based on the calculated degree of similarity, the filter The ratio set in the processing unit is detected. For this reason, the speech processing apparatus 100 can suppress the calculation amount of the filter processing. For example, the speech processing apparatus 100 can suppress the calculation amount of the filter processing in an apparatus in which the detected ratio is set. For example, the sound processing apparatus 100 can improve the sound quality by increasing the overlap ratio when the original sound has high sound quality or in an environment in which noise is difficult to be mixed. In addition, for example, the speech processing apparatus 100 calculates the filter processing while improving the sound quality by suppressing an excessive increase in the overlap ratio when the original sound has low sound quality or in an environment where noise is likely to be mixed. The amount can be suppressed.

  Further, for example, the speech processing apparatus 100 calculates the ratio between the calculated correlation and the previously calculated correlation among the calculated correlations, and when the calculated ratio is less than the threshold, the previously calculated correlation is calculated. Detect the percentage of the moment. For this reason, the speech processing apparatus 100 can suppress the calculation amount of the filter processing in the apparatus in which the detected ratio is set.

  Further, for example, each time the analysis frame overlap ratio is increased, the speech processing apparatus 100 calculates the correlation between the signal after the filtering process and the original sound, and based on the calculated correlation, the filter The ratio set in the processing unit is detected. For this reason, the speech processing apparatus 100 can obtain a sound close to the original sound with a small amount of calculation in the apparatus in which the detected ratio is set.

  Further, for example, since the audio processing apparatus 100 sets the detected overlap ratio in the filter processing unit, it is possible to suppress the calculation amount of the filter processing without repeating the trial and error while the user views many times. .

  Further, for example, since the speech processing apparatus 100 suppresses the calculation amount of the filter processing, it is possible to suppress the power consumption related to the calculation amount. This is particularly effective in a device driven by a battery such as a cellular phone or a portable music player.

  For example, since the speech processing apparatus 100 suppresses the calculation amount of the filter processing, it is possible to suppress the heat generation of the apparatus related to the calculation amount. This is particularly effective in devices carried by the user, such as mobile phones and portable music players.

  In the first embodiment, the case where the filtered sound is brought close to the original sound has been described. However, the present invention is not limited to this, and for example, the filtered sound can be brought close to an arbitrary signal. Therefore, in the second embodiment, a case will be described in which the sound processing apparatus brings the filtered sound close to the sound signal that has been subjected to the time domain FIR filter processing.

  An example of a functional configuration of the speech processing apparatus according to the second embodiment will be described. FIG. 11 is a block diagram illustrating a functional configuration of the speech processing apparatus according to the second embodiment. As shown in FIG. 11, the speech processing apparatus 200 includes a signal acquisition unit 110, a calculation unit 120, a filter processing unit 130, a ratio determination unit 140, and an FIR filter 210. Among these, the description of the signal acquisition unit 110, the calculation unit 120, the filter processing unit 130, and the ratio determination unit 140 illustrated in FIG. 11 is the same as that of the signal acquisition unit 110, the calculation unit 120, the filter processing unit 130, and the ratio determination illustrated in FIG. Since it is the same as the description of the unit 140, the description is omitted.

  FIG. 12 is a diagram for explaining an example of a signal flow in the sound processing apparatus according to the second embodiment. Each processing function shown in FIG. 12 corresponds to each processing function having the same reference numeral shown in FIG. The signal flow in the voice processing device will be described together with each processing function of the voice processing device 200.

  For example, the FIR filter 210 performs time-domain FIR filter processing on an audio signal. For example, the FIR filter 210 performs a filter process with the filter count X (s) set by the calculation unit 120. For example, as shown in FIG. 12, the FIR filter 210 outputs an audio signal subjected to the time-domain FIR filter processing to the detection unit 141.

  Next, the processing procedure of the speech processing apparatus 200 according to the second embodiment will be described. FIG. 13 is a flowchart illustrating the processing procedure of the sound processing apparatus according to the second embodiment. The process shown in FIG. 13 is executed at a predetermined interval while power is supplied from the power supply in the audio processing device 200, for example.

  As shown in FIG. 13, when the processing timing comes (step S201, Yes), the signal acquisition unit 110 outputs the original sound from the speaker 10 (step S202). Until the processing timing comes (No in step S201), the processing illustrated in FIG. 13 is in a standby state.

  The microphone 20 collects the original sound output from the speaker 10 (step S203). The calculation unit 120 calculates the FIR filter 133 (step S204). That is, the calculation unit 120 calculates the filter count X (s) applied to the FIR filter 133.

The filter processing unit 130 sets n = 1 (step S205). The filter processing unit 130 calculates a filter processing sound that has been subjected to the window function processing at an overlap ratio of (1-1 / 2 ⁿ ) × 100% (step S206).

  The ratio determining unit 140 compares the audio signal that has been subjected to the time-domain FIR filter processing by the FIR filter 210 with the filtered sound, and calculates a determination coefficient (step S207). If n = 1 is not satisfied (step S208, No), the ratio determination unit 140 compares the determination coefficient at the time of n-1 with the determination coefficient at the time of n, and determines whether or not it has increased by a certain ratio or more. Is determined (step S209).

  If it has not increased by more than a certain ratio (No at Step S209), the ratio determining unit 140 sets the overlap ratio at n-1 in the filter processing unit 130 (Step S211).

  On the other hand, if it has increased by a certain percentage or more (step S209, Yes), the percentage determination unit 140 increments n set in the filter processing unit 130 by 1 (step S210), and proceeds to step S206.

  On the other hand, if n = 1 (step S208, Yes), the ratio determining unit 140 increments n set in the filter processing unit 130 by 1 (step S210), and proceeds to step S206.

  Next, effects of the sound processing device 200 according to the second embodiment will be described. The audio processing device 200 executes frequency domain filtering using an input signal that uses window function processing in which analysis frames overlap at a predetermined rate. Each time the speech processing apparatus 200 increases the rate of overlap of analysis frames, the speech processing apparatus 200 calculates the similarity between the signal after the filter processing is executed and an arbitrary signal, and based on the calculated similarity, the filter The ratio set in the processing unit is detected. For this reason, the audio processing device 200 can suppress the calculation amount of the filter processing while bringing the filtered signal close to an arbitrary signal. For example, the audio processing device 200 can suppress the calculation amount of the filter processing while bringing the filtered sound close to the audio signal that has been subjected to the time-domain FIR filter processing. For example, the sound processing device 200 can suppress the amount of calculation of the filter processing while bringing the filtered sound close to the ideal sound that the audience wants to hear.

  Although the embodiments of the present invention have been described so far, the present invention may be implemented in other embodiments besides the above-described embodiments. Therefore, other embodiments will be described below.

  For example, in the first embodiment and the second embodiment, the detection unit 141 detects the overlap ratio when the ratio between the determination coefficient at n-1 and the determination coefficient at n has not increased by a certain ratio or more. Explained. However, the present invention is not limited to this. For example, the detection unit 141 may detect the overlap ratio using a threshold value.

  For example, the detection unit 141 detects the overlap ratio when the determination coefficient calculated at n is equal to or greater than a threshold value. For example, when the determination coefficient calculated at n is equal to or greater than the threshold value “0.99994”, the detection unit 141 detects the overlap ratio at n. For example, the detection unit 141 outputs the detected overlap ratio to the setting unit 142. Here, the threshold value is described as 0.99994, but the present invention is not limited to this, and a person who uses the speech processing apparatuses 100 and 200 can set an arbitrary value.

Here, the reason why the detection unit 141 detects the overlap ratio using the threshold value is that the sound to be output may be low sound quality, or it may be meaningless to improve the sound quality too much. For example, if the sound to be output has a sound quality comparable to that of FM radio, “0.99994” may be used as the threshold value of the determination coefficient. FIG. 14 is a diagram for explaining the relationship between the determination coefficient and the sound quality. In FIG. 14, the original sound of 44.1k sampling 16-bit data having the same sound quality as that of the audio CD is converted into 44.1k sampling 8bit data having the same sound quality as that of FM radio, and the determination coefficient is obtained. . The horizontal axis of FIG. 14 indicates the amplitude of the waveform of 44.1k sampling 16-bit data, and the vertical axis indicates the amplitude of the waveform of 44.1k sampling 8-bit data. The description of FIG. 14 is the same as the description of FIG. 14, the estimated value f and the coefficient of determination R ² is as follows.
f = 1.00060006x-0.977042
R ² = 0.9999940

  That is, when the output sound may have the same sound quality as FM radio, it can be said that it is sufficient that the determination coefficient is “0.99994” or more. For this reason, the detection unit 141 detects the overlap ratio when the sound to be output has a sound quality equivalent to that of the FM radio, and the determination coefficient becomes “0.99999” or more, thereby filtering the filter. The calculation amount of processing can be suppressed.

As described above, the coefficient of determination can be arbitrarily changed, but it is considered that “0.999” or less is not preferable. FIG. 15 is a diagram for explaining the relationship between the determination coefficient and the sound quality. FIG. 15 shows a case where a low-pass filter in which suppression of 1 kHz is 0 dB and suppression after 10 kHz is −100 dB is applied to the sound quality of the audio CD for the purpose of noise removal. The horizontal axis in FIG. 15 indicates the amplitude of the waveform of the audio CD, and the vertical axis indicates the amplitude of the waveform after the low-pass filter is applied. The description of FIG. 15 is the same as the description of FIG. 15, the estimated value f and the coefficient of determination R ² is as follows.
f = 0.969675930x + 0.00729994
R ² = 0.999899060

  In FIG. 15, the cymbal sound included in the audio CD is almost inaudible. That is, it is considered that the sound quality of the original sound cannot be maintained when the determination coefficient is about “0.999”.

  Here, a processing procedure in a case where the detection unit 141 detects an overlap ratio using a threshold value will be described. FIG. 16 is a flowchart illustrating the processing procedure of the speech processing apparatus according to the third embodiment. FIG. 16 illustrates a case where the speech processing apparatus 100 described in the first embodiment detects an overlap ratio using a threshold value. The process shown in FIG. 16 is executed at a predetermined interval while power is supplied from the power supply, for example, in the audio processing apparatus 100.

  As shown in FIG. 16, when the processing timing comes (Yes in step S301), the signal acquisition unit 110 outputs the original sound from the speaker 10 (step S302). Until the processing timing comes (No in step S301), the processing illustrated in FIG. 16 is in a standby state.

  The microphone 20 collects the original sound output from the speaker 10 (step S303). The calculation unit 120 calculates the FIR filter 133 (step S304). That is, the calculation unit 120 calculates the filter count X (s) applied to the FIR filter 133.

The filter processing unit 130 sets n = 1 (step S305). The filter processing unit 130 calculates a filter processing sound that has been subjected to the window function processing at an overlap ratio of (1-1 / 2 ⁿ ) × 100% (step S306).

  The ratio determining unit 140 compares the filtered sound with the original sound and calculates a determination coefficient (step S307). The ratio determining unit 140 determines whether or not the determination coefficient is equal to or greater than a threshold value (step S308). If the determination coefficient is not equal to or greater than the threshold (No at Step S308), the ratio determination unit 140 increments n set in the filter processing unit 130 by 1 (Step S309), and proceeds to Step S306.

  On the other hand, when the determination coefficient is greater than or equal to the threshold (step S308, Yes), the ratio determining unit 140 sets the overlap ratio in the filter processing unit 130 (step S310).

  FIG. 17 is a flowchart illustrating the processing procedure of the speech processing apparatus according to the third embodiment. FIG. 17 illustrates a case where the speech processing apparatus 200 described in the second embodiment detects an overlap ratio using a threshold value. The process shown in FIG. 17 is executed at a predetermined interval while power is supplied from the power supply in the audio processing device 200, for example.

  As shown in FIG. 17, when the processing timing comes (step S401, Yes), the signal acquisition unit 110 outputs the original sound from the speaker 10 (step S402). Until the processing timing comes (No in step S401), the processing illustrated in FIG. 17 is in a standby state.

  The microphone 20 collects the original sound output from the speaker 10 (step S403). The calculation unit 120 calculates the FIR filter 133 (step S404). That is, the calculation unit 120 calculates the filter count X (s) applied to the FIR filter 133.

The filter processing unit 130 sets n = 1 (step S405). The filter processing unit 130 calculates a filter processing sound that has been subjected to the window function processing at an overlap ratio of (1-1 / 2 ⁿ ) × 100% (step S406).

  The ratio determination unit 140 compares the sound signal that has been subjected to the time-domain FIR filter processing by the FIR filter 210 with the filter processing sound, and calculates a determination coefficient (step S407). The ratio determining unit 140 determines whether or not the determination coefficient is equal to or greater than a threshold (step S408). If the determination coefficient is not equal to or greater than the threshold (No at Step S408), the ratio determination unit 140 increments n set in the filter processing unit 130 by 1 (Step S409), and proceeds to Step S406.

  On the other hand, when the determination coefficient is equal to or larger than the threshold (step S408, Yes), the ratio determining unit 140 sets the overlap ratio in the filter processing unit 130 (step S410).

  Further, for example, among the processes described in the first and second embodiments, all or a part of the processes described as being automatically performed can be manually performed. Alternatively, all or part of the processes described as being manually performed among the processes can be automatically performed by a known method. In addition, the processing procedures, control procedures, specific names, and information including various data and parameters shown in the above-described document and drawings can be arbitrarily changed unless otherwise specified.

  1 and 11 are functionally conceptual elements, and need not be physically configured as illustrated. That is, the specific form of distribution / integration of the speech processing apparatuses 100 and 200 is not limited to the illustrated one, and all or a part thereof can be functionally or physically processed in arbitrary units according to various loads and usage conditions. Can be distributed and integrated. For example, the voice processing apparatus 100 does not necessarily have the setting unit 142. For example, the overlap ratio detected by the voice processing device 100 may be set in another device.

  The voice processing devices 100 and 200 can also be realized by mounting the functions of the voice processing devices 100 and 200 on a known information processing device. The known information processing apparatus corresponds to a device such as a personal computer, a mobile phone, a PHS (Personal Handy-phone System) terminal, a mobile communication terminal, or a PDA (Personal Digital Assistant).

  FIG. 18 is a diagram illustrating an example of a computer that executes a voice processing program. As illustrated in FIG. 18, the computer 300 includes a CPU 301 that executes various arithmetic processes, an input device 302 that receives data input from a user, and a monitor 303. The computer 300 also includes a medium reading device 304 that reads a program or the like from a storage medium, an interface device 305 for connecting to another device, and a wireless communication device 306 for connecting to another device wirelessly. The computer 300 also includes a RAM (Random Access Memory) 307 that temporarily stores various types of information and a hard disk device 308. Each device 301 to 308 is connected to a bus 309. Although not shown, the computer 300 is connected to a microphone and a speaker.

  The hard disk device 308 stores an audio processing program having the same functions as the processing units of the filter processing unit 130 and the detection unit 141 shown in FIGS. The hard disk device 308 stores various data for realizing the voice processing program.

  The CPU 301 reads each program stored in the hard disk device 308, develops it in the RAM 307, and performs various processes. In addition, these programs can cause the computer to function as the filter processing unit 130 and the detection unit 141 illustrated in FIGS.

  Note that the above-described voice processing program is not necessarily stored in the hard disk device 308. For example, the computer 300 may read and execute a program stored in a computer-readable recording medium. As the computer-readable recording medium, for example, a portable recording medium such as a CD-ROM, a DVD disk, and a USB memory, a semiconductor memory such as a flash memory, a hard disk drive, and the like are supported. Further, the program may be stored in a device connected to a public line, the Internet, a LAN (Local Area Network), a WAN (Wide Area Network), etc., and the computer 300 may read and execute the program therefrom. good.

  The following supplementary notes are further disclosed with respect to the embodiments including the above examples.

(Supplementary Note 1) A filter processing unit that performs frequency domain filtering using window function processing in which analysis frames overlap at a predetermined rate with respect to an input signal;
Each time the ratio is increased, the degree of similarity between the signal after the filter processing is executed and an arbitrary signal is calculated, and the ratio set in the filter processing unit is detected based on the calculated degree of similarity. A speech processing apparatus comprising: a detection unit.

(Additional remark 2) The said detection part calculates the ratio of the similarity calculated this time and the similarity calculated last time among the calculated similarities, and when the calculated ratio is less than a threshold value, the similarity calculated the last time The speech processing apparatus according to appendix 1, wherein the ratio when the value is calculated is detected.

(Additional remark 3) The said detection part detects the said ratio when the said similarity is calculated, when the calculated similarity becomes more than a threshold value, The audio processing apparatus of Additional remark 1 characterized by the above-mentioned.

(Supplementary note 4) The speech processing apparatus according to any one of supplementary notes 1 to 3, wherein the detection unit uses the input signal as the arbitrary signal.

(Additional remark 5) The said detection part uses the signal after the filter process of the time domain was performed with respect to the said input signal as said arbitrary signal, Any one of Additional remark 1 thru | or 3 characterized by the above-mentioned. The speech processing apparatus according to the description.

(Supplementary note 6) The speech processing apparatus according to any one of supplementary notes 1 to 5, further comprising a setting unit that sets the ratio detected by the detection unit in the filter processing unit.

(Supplementary note 7) A voice processing method executed by a computer,
Performs frequency domain filtering using the window function processing where the analysis frames overlap with the input signal at a predetermined rate,
Each time the ratio is increased, the degree of similarity between the signal after the filter process is executed and an arbitrary signal is calculated, and the ratio set in the filter process is detected based on the calculated degree of similarity. A voice processing method characterized by the above.

(Additional remark 8) The said process to detect calculates the ratio of the similarity calculated this time and the similarity calculated last time among the calculated similarities, and when the calculated ratio is less than a threshold value, the similarity calculated the last time The voice processing method according to appendix 7, wherein the ratio when the degree is calculated is detected.

(Supplementary note 9) The voice processing method according to supplementary note 7, wherein, when the calculated similarity is equal to or greater than a threshold value, the detection processing detects the ratio when the similarity is calculated. .

(Supplementary note 10) The voice processing method according to any one of supplementary notes 7 to 9, wherein the detection process uses the input signal as the arbitrary signal.

(Additional remark 11) The said process to detect uses the signal after the filter process of the time domain was performed with respect to the said input signal as said arbitrary signal, Any one of Additional remark 7 thru | or 9 characterized by the above-mentioned. The voice processing method described in 1.

(Supplementary note 12) The voice processing method according to any one of supplementary notes 7 to 11, wherein the ratio detected by the detection process is set in the filter process.

(Supplementary note 13)
Performs frequency domain filtering using the window function processing where the analysis frames overlap with the input signal at a predetermined rate,
Each time the ratio is increased, the similarity between the signal after the filter process is executed and an arbitrary signal is calculated, and the ratio set in the filter process is detected based on the calculated similarity. A voice processing program for executing a process.

(Additional remark 14) The said process to detect calculates the ratio of the similarity calculated this time and the similarity calculated last time among the calculated similarities, and when the calculated ratio is less than a threshold value, the similarity calculated last time 14. The voice processing program according to appendix 13, wherein the ratio when the degree is calculated is detected.

(Supplementary note 15) The voice processing program according to supplementary note 13, wherein, when the calculated similarity is equal to or greater than a threshold, the processing to detect detects the ratio when the similarity is calculated. .

(Supplementary note 16) The sound processing program according to any one of supplementary notes 13 to 15, wherein the detection process uses the input signal as the arbitrary signal.

(Supplementary note 17) Any one of Supplementary notes 13 to 15, wherein the detection process uses, as the arbitrary signal, a signal after a time domain filtering process is performed on the input signal. The voice processing program described in 1.

(Supplementary note 18) The audio processing program according to any one of supplementary notes 13 to 17, wherein the ratio detected by the detection processing is set in the filter processing.

DESCRIPTION OF SYMBOLS 10 Speaker 20 Microphone 100,200 Audio | voice processing apparatus 110 Signal acquisition part 120 Calculation part 130 Filter process part 131 Window function process part 132 Conversion part 133 FIR filter 134 Inverse conversion part 135 Addition part 140 Ratio determination part 141 Detection part 142 Setting part 210 FIR filter

Claims (8)

A filter processing unit that performs frequency domain filtering using window function processing in which analysis frames overlap at a predetermined rate with respect to an input signal;
Each time the ratio is increased, the degree of similarity between the signal after the filter processing is executed and the signal based on the input signal is calculated, and set in the filter processing unit based on the calculated degree of similarity. An audio processing apparatus comprising: a detection unit that detects a ratio.   The detection unit calculates a ratio between the calculated similarity and the previously calculated similarity, and when the calculated ratio is less than a threshold, the previously calculated similarity is calculated. The voice processing apparatus according to claim 1, wherein the ratio is detected.   The speech processing apparatus according to claim 1, wherein when the calculated similarity is equal to or greater than a threshold, the detection unit detects the ratio when the similarity is calculated. The audio processing apparatus according to claim 1, wherein the detection unit uses the input signal as a signal based on the input signal. The said detection part uses the signal after the filter process of the time domain was performed with respect to the said input signal as a signal based on the said input signal, The Claim 1 characterized by the above-mentioned. The speech processing apparatus according to the description.   The speech processing apparatus according to claim 1, further comprising a setting unit configured to set a ratio detected by the detection unit in the filter processing unit. An audio processing method executed by a computer,
Performs frequency domain filtering using the window function processing where the analysis frames overlap with the input signal at a predetermined rate,
Each time the ratio is increased, the degree of similarity between the signal after the filter process is executed and the signal based on the input signal is calculated, and the ratio is set for the filter process based on the calculated degree of similarity. A speech processing method characterized by detecting a signal. On the computer,
Performs frequency domain filtering using the window function processing where the analysis frames overlap with the input signal at a predetermined rate,
Each time the ratio is increased, the degree of similarity between the signal after the filter process is executed and the signal based on the input signal is calculated, and the ratio is set for the filter process based on the calculated degree of similarity. A voice processing program for executing each process.

Images