JP5335390B2 - Signal processing apparatus and signal processing method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a signal processing device and a signal processing method, accurately calculating a fundamental period of an input signal just before the current time. <P>SOLUTION: A template setting section 51 sets a voice signal of a specified time width in the past from the time RT when packet loss occurs, as a template TM. In a period detection section 52, the template TM set by the template setting section 51 is shifted to the past from the time RT of the voice signal, and the template TM is correlated with the voice signal, and the fundamental period of the voice signal is detected just before the time RT on the basis of the shift amount at the time when a peak of correlation of the template TM and the voice signal is the highest. Here, the template setting section 51 increases the time width of the template TM, as the period detection section 52 increases the shift amount of the template TM. <P>COPYRIGHT: (C)2010,JPO&amp;INPIT

Description

  The present invention relates to a signal processing device and a signal processing method for detecting a fundamental period of an input signal.

  In recent years, VoIP (Voice over Internet Protocol) communication in which voice communication is performed using a communication medium such as a PLC or a wireless LAN is known. Since communication quality such as PLC and wireless LAN tends to fluctuate, deterioration of voice quality due to packet loss becomes a problem. When a packet loss occurs, for example, a noise sound such as a pudding may occur during a call.

  Therefore, in the VoIP communication apparatus, when packet loss occurs, the basic period (pitch) of the immediately preceding audio signal is detected, and the loss period is interpolated using the audio signal for the basic period immediately before the occurrence of the packet loss. Concealment processing is performed.

  FIG. 7 is a waveform diagram of an audio signal for explaining a conventional concealment process. In FIG. 7, the vertical axis indicates the intensity of the audio signal input to the communication device, and the horizontal axis indicates time. When reception of a voice packet fails and packet loss occurs, the communication apparatus sets a voice signal for a predetermined period immediately before the packet loss occurs as a template.

  Next, the template is slid toward the past from the time when the packet loss occurs with respect to the audio signal. Next, correlation calculation between the template and the audio signal is executed, and the basic period of the audio signal immediately before the packet loss occurs is detected.

  Next, from the occurrence of packet loss, the audio signal for the basic period is extracted retroactively, and the audio signal is repeatedly applied to the loss period to interpolate the loss period.

  Here, the loss period is interpolated with the audio signal for this basic period. For example, when the caller utters the voice “A”, the voice “A” is divided into about 20 msec and is converted into one voice. This is because, since it is transmitted on a packet, there is a high possibility that the audio signal for the basic period immediately before the occurrence of the packet loss is repeated in the loss period.

  Therefore, when performing concealment processing, it is important to detect the basic period of the audio signal immediately before the occurrence of packet loss.

  Note that Patent Documents 1 to 4 shown below are known as conventional fundamental period detection methods. In Patent Document 1, an autocorrelation function R (τ) of an input speech waveform is obtained with a plurality of different analysis window widths, and the obtained maximum value is R (τ) max, and V = R (τ) max / R (0 ), And the pitch of the input speech waveform is determined from the most reliable value of τ in consideration of the magnitude of V and the variation in τ.

  In Patent Document 2, an autocorrelation function φ (d) of a digitally sampled sound signal is obtained, an approximate pitch of the audio signal is extracted, and the audio signal is an integer multiple of the basic interval using the extracted approximate pitch as a basic interval. A technique is disclosed in which a calculation interval is determined by dividing, and the pitch of an audio signal is obtained from the autocorrelation function φ ′ (d) of the determined calculation interval.

  In Patent Document 3, when the average value of the pitch extracted in the past frame is large, the analysis window length is set long and the thinning rate is set high. Conversely, when the average value is small, the analysis window length is set short. In addition, a technique for setting the thinning rate low is disclosed.

In Patent Document 4, when the pitch of the input speech signal is obtained for each of a plurality of predetermined pitch candidates, the evaluation range on the time axis of the input speech signal used for cycle evaluation becomes wider as the pitch candidates become longer cycles. In addition, a technique for narrowing the sampling period of the input audio signal is disclosed.
Japanese Patent No. 3219868 Japanese Patent No. 3323448 Japanese Patent No. 345941 JP 2004-317533 A

  However, none of the methods of Patent Documents 1 to 4 has the problem of performing the concealment process described above, and simply detects the basic period from the input signal with high accuracy. Therefore, there is a problem that the basic period of the input signal immediately before the packet loss cannot be detected with high accuracy. In particular, in the methods of Patent Documents 1 to 4, when the basic period of the input signal changes immediately before the packet loss occurs, if this basic period and the time width of the template do not match, the basic period is accurately determined. The problem that it cannot be detected well occurs.

  An object of the present invention is to provide a signal processing apparatus and a signal processing method capable of accurately calculating the basic period of an input signal immediately before the present time.

  (1) A signal processing device according to one aspect of the present invention includes a reference signal setting unit that sets an input signal having a time width from the current time to the past as a reference signal, and the reference signal is set to the input signal from the current time. And a period detection unit that detects a basic period of the input signal by sliding toward the past and obtaining a correlation between the reference signal and the input signal, and the reference signal setting unit is configured to slide the reference signal. As the quantity increases, the time width of the reference signal is increased.

  A signal processing method according to another aspect of the present invention includes a reference signal setting step of setting an input signal having a time width from the present time to the past as a reference signal, and changing the reference signal from the current input signal to the past input signal. And a period detection step for detecting the fundamental period by obtaining a correlation between the reference signal and the input signal, and the period detection step includes the reference signal as the reference signal slide amount increases. Increase the time width of the signal.

  According to this configuration, an input signal having a time width from the present time to the past is set as the reference signal. Then, the set reference signal is slid toward the past from the present time with respect to the input signal. Then, the correlation between the reference signal and the input signal is obtained, and the fundamental period of the input signal is detected.

  Here, the time width of the reference signal is increased as the slide amount increases. Therefore, at a relatively early stage where the slide amount is small, a timing is generated at which the input signal for the basic period immediately before the current time is used as the reference signal. At this time, a strong correlation peak appears between the reference signal and the input signal. On the other hand, when the slide amount increases, the time width of the reference signal is increased accordingly, and the reference signal includes a plurality of frequency components. Therefore, it becomes impossible to obtain a stronger correlation peak as the correlation peak obtained at the above timing. Therefore, the basic period of the input signal immediately before the current time can be detected with high accuracy.

  Furthermore, in a relatively early stage where the slide amount is small, the time width of the reference signal is short, so that the calculation amount can be reduced.

  (2) A packet loss detection unit that detects whether or not a packet loss has occurred in the input signal, and a concealment processing unit, wherein the period detection unit is a loss in which a packet loss has occurred by the packet loss detection unit The basic period is detected with the occurrence time as the current time, and the concealment processing unit extracts the input signal for the basic period from the loss occurrence time to the past, and a loss period in which a packet loss has occurred in the extracted input signal Is preferably interpolated.

  According to this configuration, when packet loss occurs, the loss occurrence time is set as the current time, and the basic period of the input signal immediately before the loss occurrence time is detected. Then, an input signal corresponding to the basic period is extracted from the point of occurrence of loss toward the past, and the loss period is interpolated using this input signal to perform concealment processing. Therefore, the concealment process can be performed with high accuracy.

  (3) It is preferable that the reference signal setting unit sets a time width of the reference signal to a predetermined initial time width until a slide amount of the reference signal reaches a predetermined slide reference value.

  According to this configuration, when the slide amount of the reference signal is relatively small, the time width of the reference signal is set to the initial time width. Therefore, even when the slide amount is small, the time width of the reference signal is constant. It is possible to ensure a size equal to or greater than that, and the correlation between the reference signal and the input signal can be obtained with higher accuracy. As the initial time width, for example, a value about the minimum value of the assumed basic period of the input signal may be adopted. As the slide reference value, for example, an initial time width may be adopted.

  (4) It is preferable that the period detection unit obtains a correlation between the reference signal and the input signal by cross-correlation.

  According to this configuration, since the cross-correlation is employed, the correlation between the reference signal and the input signal can be calculated with high accuracy.

  (5) It is preferable that the period detection unit obtains a correlation between the reference signal and the input signal by AMDF.

  According to this configuration, since the AMDF is employed, it is possible to calculate the correlation between the reference signal and the input signal with high accuracy while using a relatively small amount of calculation.

  (6) Preferably, the input signal is a signal sampled at a predetermined sampling period, and the period detector performs the cross-correlation using equation (1).

  Where φ (τ) represents a correlation value, N represents a time width of the reference signal, x (j) represents the reference signal, x (j−τ) represents the input signal, and k + 1 represents the input signal. The starting point of the reference signal is indicated, a is a predetermined coefficient, τ is the sliding amount of the reference signal, and j is the sampling number of each sampling point of the input signal.

  According to this configuration, the cross correlation is calculated using Expression (1).

  (7) It is preferable that the input signal is a signal sampled at a predetermined sampling period, and the period detection unit performs the AMDF using Equation (2).

  Where φ (τ) represents a correlation value, N represents a time width of the reference signal, x (j) represents the reference signal, x (j−τ) represents the input signal, and k + 1 represents the input signal. The starting point of the reference signal is indicated, a is a predetermined coefficient, τ is the sliding amount of the reference signal, and j is the sampling number of each sampling point of the input signal.

  According to this configuration, AMDF is performed using equation (2).

  (8) The reference signal setting unit sets the a to a predetermined fixed value within a range of 1 ≦ a <2 until the slide amount of the reference signal exceeds a predetermined change reference value, When the slide amount exceeds the change reference value, it is preferable to decrease the value a so as to approach 1 as the slide amount approaches a predetermined maximum slide amount.

  According to this configuration, when the slide amount is small, the time width of the reference signal can be set larger than the slide amount, and when the slide amount is large, the time width of the reference signal is set to a value about the slide amount. can do. Therefore, when the slide amount is small, it is possible to prevent the time width of the reference signal from becoming too small.

  According to the present invention, in a relatively early stage where the slide amount is small, a timing is generated at which the input signal for the basic period almost immediately before the current point is used as the reference signal. At this time, a strong correlation peak appears between the reference signal and the input signal. On the other hand, when the slide amount increases, the time width of the reference signal is increased accordingly, and the reference signal includes a plurality of frequency components. Therefore, it becomes impossible to obtain a stronger correlation peak as the correlation peak obtained at the above timing. Therefore, it is possible to detect the basic period of the input signal almost immediately before the present time with high accuracy.

  Hereinafter, a case where a signal processing device according to an embodiment of the present invention is applied to a communication device will be described as an example. FIG. 1 is a block diagram showing an overall configuration of a communication apparatus to which a signal processing apparatus according to an embodiment of the present invention is applied. Note that this communication apparatus is a communication apparatus having a call function based on VoIP, for example, SIP (Session Initiation Protocol), H.264. Communication is performed with other communication devices connected via a predetermined communication network using a communication protocol such as H.323, IPv4, or IPv6.

  As shown in FIG. 1, the communication apparatus includes a packet receiver 1, a delay fluctuation absorbing buffer 2, a timer 3, a packet loss detector 4, a detection processor 5, a concealment processor 6, an audio output unit 7, and a speaker 8. It has.

  The packet receiver 1 receives a voice packet transmitted from another communication device, and outputs this voice packet to the delay fluctuation absorbing buffer 2. This voice packet is a voice packet compliant with, for example, RTP (Real-time Transport Protocol), and includes a 20 msec digital voice signal. The audio signal is a digital audio signal sampled at a sampling frequency of 8 kHz by, for example, PCM μ-Law or the like.

  Therefore, the packet receiving unit 1 outputs the received voice packet to the delay fluctuation absorbing buffer 2 in time-series order according to the sequence number included in the RTP header of the voice packet. Note that the RTP header includes a time stamp and the like in addition to the sequence number. The sequence number indicates the transmission order of voice packets, and the time stamp indicates the relative position of the voice signal in the original voice waveform.

  The delay fluctuation absorbing buffer 2 temporarily holds the voice packet output from the packet receiving unit 1 and outputs it after a predetermined time delay to absorb the voice packet delay fluctuation.

  The timer 3 is used when the packet loss detection unit 4 detects a packet loss. When the delay fluctuation absorbing buffer 2 outputs a voice packet to the detection processing section 5, the packet loss detecting section 4 causes the timer 3 to start timing, and before the delay fluctuation absorbing buffer 2 outputs the next voice packet, When the time measured by 3 exceeds a predetermined time when it is assumed that a packet loss has occurred, it is determined that a packet loss has occurred.

  The detection processing unit 5 extracts an audio signal according to the sequence number and the time stamp from the audio packets sequentially output from the delay fluctuation absorbing buffer 2. When the packet loss detection unit 4 detects a packet loss, the detection processing unit 5 performs a basic period (pitch) detection process on the extracted audio signal, and the packet loss detection unit 4 detects the packet loss. If not, the extracted audio signal is output to the audio output unit 7 as it is. It is assumed that the detection processing unit 5 holds an audio signal for a certain past period.

  The detection processing unit 5 includes a template setting unit 51 (an example of a reference signal setting unit) and a period detection unit 52. The template setting unit 51 sets, as a template (an example of a reference signal), an audio signal having a time width from the loss occurrence time point when the packet loss has occurred toward the past. Here, the template setting unit 51 increases the time width of the template as the period detection unit 52 increases the slide amount of the template.

  The period detection unit 52 slides the template set by the template setting unit 51 from the loss occurrence time to the past with respect to the audio signal, obtains the correlation between the template and the audio signal, and correlates the peak between the template and the audio signal. The basic period of the audio signal immediately before the point of occurrence of loss is detected from the amount of slide when appears most strongly.

  FIG. 2 is a waveform diagram of an audio signal for explaining the processing of the template setting unit 51 and the cycle detection unit 52. Note that the vertical axis shown in FIG. 2 indicates the intensity of the audio signal, and the horizontal axis indicates time as the number of samples. A template TJ shown in FIG. 2 indicates a template used for the conventional concealment process.

  When a packet loss occurs, the conventional communication apparatus sets, for example, an audio signal for a predetermined period in the past from the loss occurrence time RT as the template TJ. Then, by sliding this template TJ toward the past from the loss occurrence time RT with respect to the audio signal, the correlation between the audio signal and the template TJ is obtained, and the slide amount of the template TJ when the strongest correlation peak is obtained. The basic period of the audio signal was detected.

  FIG. 3 is a graph showing the calculation result of the correlation value between the template TJ and the audio signal when the conventional template TJ is used. In FIG. 3, the correlation value is calculated using AMDF. In FIG. 3, the vertical axis indicates the correlation value, and the horizontal axis indicates the time when the loss occurrence time RT is 0 as the number of samples. Also, since FIG. 3 shows the correlation value by AMDF, the smaller the value, the stronger the correlation between the audio signal and the template TJ.

  In FIG. 3, first, a convex correlation peak PK1 appears downward at the time of 37 samples, then a downward convex correlation peak PK2 appears at the time of 47 samples, and thereafter convex downward at a period of about 37 samples. The correlation peak of appears repeatedly. The correlation peak PK1 appears smaller than the correlation peak PK2. Therefore, in the conventional method, 37 samples are detected as the basic period of the audio signal.

  On the other hand, as shown in FIG. 2, the basic period of the audio signal immediately before the loss occurrence time RT is 47 samples. Therefore, it can be seen that the conventional method does not accurately detect the fundamental period of the audio signal immediately before the loss occurrence time RT.

  This is because the time width of the template TJ is much larger than 47 samples, and the template TJ includes only one period of a sound signal whose basic period to be detected is 47 samples. However, since the sound signal of 37 samples includes 3 periods, it is considered that a strong correlation peak appeared at 37 samples.

  In this case, the concealment process is performed by taking out the audio signal of 37 samples retroactively from the loss occurrence time RT, and repeatedly applying this audio signal to the loss period for interpolation.

  Therefore, it is difficult to smoothly connect the waveform of the loss period and the waveform other than the loss period, and it is difficult to perform the concealment process with high accuracy.

  On the other hand, when the time width of the template is smaller than 47 samples, the basic period of 47 samples cannot be detected.

  Therefore, in the present embodiment, as shown in FIG. 2, the time width of the template TM is increased as the slide amount of the template TM is increased.

  Therefore, for example, when the template TM is slid to some extent as in the template TM shown in the third row of FIG. 2, the template includes only 47 samples of audio signals to be detected. On the other hand, the template TM in the fourth stage in FIG. 2 includes an audio signal having a basic period of 37 samples in addition to an audio signal having a basic period of 47 samples. For this reason, the correlation between the third-stage template TM and the audio signal appears stronger than the correlation between the fourth-stage template TM and the audio signal, and the basic period of the audio signal immediately before the loss occurrence time RT is accurately determined. It becomes possible to detect.

  Here, it is preferable that the period detection unit 52 employs, for example, cross-correlation represented by Expression (1) or AMDF represented by Expression (2) as the correlation calculation.

  Where φ (τ) indicates the correlation value, N indicates the time width of the template TM, x (j) indicates the template TM, x (j−τ) indicates the audio signal, and k + 1 indicates the start of the template TM. A point indicates a predetermined coefficient, τ indicates a slide amount of the template TM, and j indicates a sampling number of each sampling point of the audio signal.

  Further, it is preferable that the template setting unit 51 sets the time width of the template TM to a predetermined initial time width until the slide amount of the template TM reaches a predetermined slide reference value.

  In this way, when the slide amount of the template TM is relatively small, the time width of the template TM is set to the initial time width, and even if the slide amount is small, the time width of the template TM is greater than a certain amount. The correlation between the template TM and the audio signal (input signal) can be obtained more accurately.

  Further, the time width of the template TM is set to the initial time width until the slide amount of the template TM reaches the slide reference value, but the amount of calculation can be reduced by relatively shortening the initial time width. .

  Note that, as the initial time width, it is preferable to employ the estimated minimum value of the basic period of the audio signal. As the slide reference value, for example, an initial time width may be adopted.

  FIG. 4 is a diagram for explaining processing of the template setting unit 51 and the cycle detection unit 52. Each point on the straight line shown in FIG. 4 indicates a sampling point of the audio signal. The rightmost sampling point indicates a loss occurrence time RT, and each sampling point indicates a past sampling point as it goes to the left. Further, the loss occurrence time RT is set as the 0th sampling point. The basic period of the audio signal is about 3 msec when it is short. If the sampling frequency is 8 kHz, it corresponds to 24 samples. Therefore, the initial time width may be 24 samples, for example. In FIG. 4, for convenience of explanation, the initial time width of the template TM is set to 4, a = 1, and the slide reference value is set to 5.

  First, when a packet loss occurs, the period detection unit 52 sets τ = 0 and the initial time width of the template TM is 4. Therefore, the fourth sampling point on the left from the loss occurrence time RT is set as the reference sampling point k. And set the sampling number to each sampling point so that it increases by 1 from k to the loss occurrence time RT, and assign the sampling number to each sampling point so that it decreases by 1 from k to the past. To do.

  Then, the template setting unit 51 sets the audio signals x (k + 1) to x (k + 4) as the template TM0.

  Then, the cycle detection unit 52 calculates the correlation value φ (0) between the template TM0 and the audio signal x (j-0) using the formula (1) or (2). In this case, the template TM0 is applied to the audio signals x (k + 1) to x (k + 4).

  Next, the period detection unit 52 sets τ = 1, and in the same manner as τ = 0, the correlation between the template TM0 and the audio signal x (j−1) is obtained using Expression (1) or (2). The value φ (1) is calculated. In this case, the template TM0 is applied to the audio signals x (k) to x (k + 3).

  Hereinafter, the template TM0 is slid toward the past with respect to the audio signal until τ = 4, and φ (2), φ (3), φ (4) are expressed by using the formula (1) or (2). Calculated.

  Next, when setting τ = 5, the cycle detection unit 52 sets τ ≧ slide reference value (= 5), and therefore sets the fifth sampling point to the left from the loss occurrence time RT as the reference sampling point k. Then, the template setting unit 51 sets the audio signals x (k + 1) to x (k + 5) as the template TM5. Then, the period detection unit 52 obtains a correlation value φ (5) between the template TM5 and the audio signal x (j-5) using the formula (1) or (2). In this case, the template TM5 is applied to the audio signals x (k-4) to x (k).

  Next, the period detection unit 52 sets τ = 6, and sets the sixth sampling point to the left from the loss occurrence time RT as the reference sampling point k. Then, the template setting unit 51 sets the audio signals x (k + 1) to x (k + 6) as the template TM6. Then, the period detection unit 52 obtains a correlation value φ (6) between the template TM6 and the audio signal x (j-6) using the formula (1) or (2). In this case, the template TM6 is applied to the audio signals x (k-5) to x (k).

  Thereafter, the cycle detection unit 52 repeats the above processing until τ reaches the maximum slide amount τmax, and obtains φ (τ). Thereby, the time width of the template TM is increased as the slide amount increases.

  FIG. 5 shows a graph of the correlation value φ (τ) when the correlation value φ (τ) is obtained for the audio signal shown in FIG. 2 using the method according to the present embodiment. In FIG. 5, the vertical axis indicates the correlation value φ (τ), and the horizontal axis indicates the time in terms of the number of samples. In FIG. 5, the correlation value φ (τ) is calculated by AMDF. Therefore, similarly to FIG. 3, the correlation peak with the lower correlation value has a stronger correlation between the audio signal and the template TM.

  In FIG. 5, a convex correlation peak PK1 appears when approximately 47 samples have elapsed from the loss occurrence time RT (= 0), and then when approximately 37 samples have elapsed since the correlation peak PK1 has appeared. A convex correlation peak PK2 appears, and thereafter a convex correlation peak appears every approximately 37 samples. The correlation peak value increases with time, and the correlation between the template TM and the audio signal is weakened. If the sampling frequency is 8 kHz, 37 samples correspond to 37 × 0.125 msec = 4.625 msec, and 47 samples correspond to 47 × 0.125 = 5.875 msec.

  That is, among the correlation peaks shown in FIG. 5, the correlation peak PK1 when the template TM is shifted by 47 samples is the minimum.

  Therefore, the period detection unit 52 detects 47 samples, which is the time when the minimum correlation peak PK1 appears, as the basic period of the audio signal immediately before the loss occurrence time RT. Therefore, it can be seen that the period detector 52 can detect 47 samples, which are the basic period of the audio signal immediately before the loss occurrence time RT shown in FIG.

  Returning to FIG. 1, the concealment processing unit 6 extracts voice signals for the basic period detected by the period detection unit 52 from the loss occurrence time RT to the past, and calculates a loss period in which packet loss has occurred in the extracted audio signals. Performs concealment processing for interpolation.

  Here, for example, if the audio signal shown in FIG. 2 is input to the concealment processing unit 6 and the period detection unit 52 detects 47 samples as the basic period, the audio of 47 samples from the loss occurrence time point RT to the past. The signal is extracted, the extracted audio signal is repeatedly applied to the end of the loss period, and the loss period is interpolated.

  The audio output unit 7 converts an audio signal that has been concealed or an audio signal that has not been concealed into an analog signal, and outputs the analog signal from the speaker 8.

  FIG. 6 is a flowchart showing processing of the communication apparatus shown in FIG. In the flowchart of FIG. 6, a = 1 is set for convenience of explanation. First, in step S1, when the packet loss detection unit 4 detects a packet loss (YES in step S1), the period detection unit 52 sets τ = 0 (step S2).

  Next, the template setting unit 51 sets a template TM having a time width corresponding to the value of τ from the audio signal (step S3). At this time, the template setting unit 51 sets the time width of the template TM to the initial time width if τ <slide reference value, and sets the time width of the template TM to N = τ if τ ≧ slide reference value. To do.

  Next, the period detection unit 52 sets the reference sampling point k so that k + 1 becomes the starting point of the template TM, and assigns a sampling number to each sampling point (step S4).

  Next, the period detection unit 52 calculates a correlation value between the template TM and the audio signal using the formula (1) or (2) (step S5).

  Next, the period detection unit 52 sets τ = τ + 1 (step S6). Next, when τ ≧ slide reference value (YES in step S7), that is, when the slide amount of the template TM exceeds the slide reference value, the cycle detection unit 52 advances the process to step S8, and τ <slide reference If it is a value (NO in step S7), the process returns to step S5. By repeating the processes of steps S5 to S7, the template TM having the initial time width is slid toward the past with respect to the audio signal until the slide TM reaches the slide reference value.

  In step S8, if τ <τmax (NO in step S8), the process returns to step S3, and the processes in steps S3 to S8 are repeated until τ ≧ τmax. As a result, the time width of the template TM is increased as the slide amount τ increases.

  In step S8, when τ ≧ τmax is satisfied (YES in step S8), the period detection unit 52 detects a correlation peak from the correlation value calculated in step S5, and the template TM and the audio signal among the detected correlation peaks. The slide amount of the correlation peak having the strongest correlation with is identified, and the basic period is detected from the identified slide amount (step S9). Here, when Expression (1) is adopted, the correlation peak having the maximum correlation value indicates the strongest correlation between the template TM and the audio signal. Further, when Expression (2) is adopted, a correlation peak showing a minimum correlation value indicates the strongest correlation between the template TM and the audio signal.

  The period detector 52 may calculate the basic period by multiplying the specified slide amount by the sampling period of the audio signal.

  Next, the concealment processing unit 6 extracts an audio signal according to the basic period detected in step S9, interpolates the loss period using the extracted audio signal, and performs concealment processing (step S10).

  In the description of FIG. 4, the template setting unit 51 sets a = 1. However, the present invention is not limited to this, and a is set to 1 ≦ a <2 until the slide amount of the template TM exceeds a predetermined change reference value. When the slide amount exceeds the change reference value, the value of a may be gradually decreased so as to approach 1 as the slide amount approaches the maximum slide amount (τmax). . As the change reference value, for example, the above-described slide reference value can be adopted.

  Thereby, when the slide amount is small, the time width of the template TM can be set larger than the slide amount. When the slide amount is large, the time width of the template TM can be set to a value about the slide amount. it can. Therefore, when the slide amount is small, it is possible to prevent the correlation calculation accuracy from being lowered due to the time width of the template TM becoming too small.

  Further, as the correlation calculation, a technique such as ASDF may be employed instead of the cross-correlation shown in Expression (1) or the AMDF shown in Expression (2).

  As described above, according to the present communication device, the audio signal having a certain time width from the loss occurrence time point RT to the past is set as the template TM. Then, the set template TM is slid toward the past from the present time with respect to the audio signal. Then, the correlation between the template TM and the audio signal is obtained, and the basic period of the audio signal is detected.

  Here, the time width of the template TM is increased as the slide amount increases. Therefore, at a relatively early stage where the slide amount is small, a timing is generated at which the audio signal for the basic period almost immediately before the current time is used as the template TM. At this time, a strong correlation peak appears between the template TM and the audio signal. On the other hand, as the slide amount increases, the time width of the template TM is increased accordingly, and the template TM includes a plurality of frequency components. Therefore, it becomes impossible to obtain a stronger correlation peak as the correlation peak obtained at the above timing. Therefore, it is possible to detect the basic period of the audio signal almost immediately before the present time with high accuracy.

1 is a block diagram showing an overall configuration of a communication apparatus to which a signal processing apparatus according to an embodiment of the present invention is applied. It is a wave form diagram of an audio signal for explaining processing of a template setting part and a cycle detection part. It is the graph which showed the calculation result of the correlation value of a template and an audio | voice signal when the conventional template is used. It is a figure explaining the process of a template setting part and a period detection part. The graph of the correlation value when the correlation value is calculated | required using the method by this Embodiment with respect to the audio | voice signal shown in FIG. It is a flowchart which shows the process of the communication apparatus shown in FIG. It is a wave form diagram of an audio signal for explaining conventional concealment processing.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Packet receiving part 2 Delay fluctuation absorption buffer 3 Timer 4 Packet loss detection part 5 Detection processing part 6 Concealment processing part 7 Audio | voice output part 8 Speaker 51 Template setting part 52 Period detection part

Claims (9)

A reference signal setting unit for setting an input signal having a time width from the present time to the past as a reference signal;
A period detection unit that detects the basic period of the input signal by sliding the reference signal toward the past from the present time with respect to the input signal, and obtaining a correlation between the reference signal and the input signal ;
A packet loss detector that detects whether or not a packet loss has occurred in the input signal ,
The reference signal setting unit increases the time width of the reference signal as the slide amount of the reference signal increases,
The signal processor according to claim 1, wherein the period detector detects the basic period with the loss occurrence time at which the packet loss detected by the packet loss detector has occurred as the current time . Toward the past from the previous SL loss occurs when taking out an input signal of the basic period, according to claim 1, further comprising a concealment processing unit for interpolating the loss period packet loss occurs in the input signal extracted Signal processing equipment.   The reference signal setting unit sets the time width of the reference signal to a predetermined initial time width until the slide amount of the reference signal reaches a predetermined slide reference value. Signal processing device.   The signal processing apparatus according to claim 1, wherein the period detection unit obtains a correlation between the reference signal and the input signal by cross-correlation.   The signal processing apparatus according to claim 1, wherein the period detection unit obtains a correlation between the reference signal and the input signal by AMDF. The input signal is a signal sampled at a predetermined sampling period,
The signal processing apparatus according to claim 4, wherein the period detection unit performs the cross-correlation using Equation (1).

Where φ (τ) represents a correlation value, N represents a time width of the reference signal, x (j) represents the reference signal, x (j−τ) represents the input signal, and k + 1 represents the input signal. The starting point of the reference signal is indicated, a is a predetermined coefficient, τ is the sliding amount of the reference signal, and j is the sampling number of each sampling point of the input signal. The input signal is a signal sampled at a predetermined sampling period,
The signal processing apparatus according to claim 5, wherein the period detection unit performs the AMDF using Equation (2).

Where φ (τ) represents a correlation value, N represents a time width of the reference signal, x (j) represents the reference signal, x (j−τ) represents the input signal, and k + 1 represents the input signal. The starting point of the reference signal is indicated, a is a predetermined coefficient, τ is the sliding amount of the reference signal, and j is the sampling number of each sampling point of the input signal.   The reference signal setting unit sets the a to a predetermined fixed value within a range of 1 ≦ a <2 until the slide amount of the reference signal exceeds a predetermined change reference value, and the slide amount of the reference signal is 8. The signal processing device according to claim 6, wherein when the change reference value is exceeded, the value of a is decreased so as to approach 1 as the slide amount approaches a predetermined maximum slide amount. A reference signal setting step for setting an input signal having a time width from the present time to the past as a reference signal;
Slide toward the past input signal the reference signal from the current input signal, by obtaining the correlation between the reference signal and the input signal, the period detection step of detecting a basic period of the input signal,
A packet loss detection step for detecting whether or not a packet loss has occurred in the input signal ,
The reference signal setting step increases the time width of the reference signal as the slide amount of the reference signal increases,
In the signal processing method , the period detection step detects the basic period with the loss occurrence time at which the packet loss detected in the packet loss detection step has occurred as the current time .

Images