JP2007010892A - Device for determining speech signal - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a device for determining speech signals capable of unfailingly determining the ending point of talking in a voice signal. <P>SOLUTION: A first signal generating means of the device generates a first signal E which is the envelope signal of the value of the voice signal level. A second signal generating means generates a second signal by smoothing the first signal E on a time axis on condition that the level value of the second signal F does not change temporally when it is above the value obtained by adding a predetermined value to the level value of the voice signal. A determination means determines the point when the state that the level value of the second signal F is above the value determined by adding a predetermined value to the level value of the voice signal has continued for a first predetermined duration as the ending point of talking in the voice signal. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention relates to an audio signal determination apparatus that determines the state and type of an audio signal.

  The “audio signal” mentioned here is a signal having the component in the audible frequency band, and the signal form is not limited. For example, it may be an analog signal or a digital signal. Further, it may be an electric signal or an optical signal. For example, a speech signal output from a microphone when a speaker's voice is received by a microphone is an audio signal. A music signal recorded on the music CD is also an audio signal.

  Some acoustic devices determine the start time and end time of an input audio signal. For example, an automatic recording device that starts recording at the beginning of a conversation and stops recording at the end of a conversation.

  In the conventional apparatus, a predetermined gate level is set for detection of the start time and the end time of the talk, and the level of the input audio signal is compared with this gate level, so that Judging the end of the talk.

  However, according to the above method based on a preset gate level, a judgment error often occurs.

  For example, since the voice input to the microphone is small, the audio signal level does not reach the gate level, and thus the sound device does not respond to the sound device even though the speaker has started a conversation or announcement. The start (speaking start) time may not be determined.

  In addition, the sound output from the microphone exceeds the gate level due to the loudness of the speaker's ambient noise, and thus the sound device ends (speaking end) even though the speaker ends the conversation or announcement. The time may not be detected.

  Furthermore, it is determined (identified) whether the input audio signal is a speech signal by conversation or announcement, or a music signal, and different gain adjustments in the loudspeaker are to be performed according to the determination result. There are cases.

  That is, when the audio signal is a speech signal, the level of the audio signal varies as the speaker approaches the microphone or moves away from the microphone. Further, in a conversation conducted by a plurality of people, there is a difference in the volume of the voice depending on the speaker, so that the voice signal level varies. Therefore, when the audio signal is a speech signal, it is desirable to perform relatively sensitive gain adjustment in accordance with fluctuations in the level of the audio signal.

  On the other hand, even when the audio signal is a music signal, the level of the audio signal varies. This is because a portion with a relatively large level and a portion with a relatively small level are mixed in one music. However, when the audio signal is a music signal, it is better not to perform a sensitive gain adjustment according to the level fluctuation in order not to impair the musicality.

  Thus, there are cases where it is desired to change the gain adjustment method depending on whether the audio signal is a speech signal or a music signal. For this purpose, a method for determining whether the audio signal is a speech signal or a music signal is necessary. However, no effective method has been proposed yet.

A technique for discriminating a speech signal from a music signal based on the autocorrelation of the signal is disclosed in Patent Document 1.
JP 2002-366189 A (first page, summary column)

  An object of the invention according to the present application is to provide an audio signal determination device that can reliably determine the end point of speech in an audio signal.

  Another object of the present invention is to provide an audio signal determination device capable of accurately determining whether an audio signal is a speech signal or a music signal.

  In order to solve the above-described problem, an audio signal determination device according to the present invention includes first signal generation means for generating a first signal, second signal generation means for generating a second signal, and determination means, The first signal generation means generates the first signal which is an envelope signal of the level value of the audio signal, and the second signal generation means sets the level value of the second signal to a predetermined value to the level value of the audio signal. The second signal is generated by smoothing the first signal on the time axis on the condition that it does not change in time when the value is equal to or greater than the value obtained by adding A point in time when the state where the level value is equal to or greater than the level value of the audio signal plus the predetermined value continues for the first predetermined period is determined as the end point of the audio signal. This device utilizes the fact that a state in which the level value of the second signal is equal to or greater than a value obtained by adding a predetermined value to the level value of the audio signal can be determined as a silent state. Judging that it is the end. Without using the gate level, the end of the talk is determined from the result of the detection of the silent state and the observation thereof, so that it is possible to accurately determine the end of the talk without being influenced by ambient noise.

  In the audio signal determination apparatus, the determination means has a level value of the second signal that is smaller than a value obtained by adding the predetermined value to the level value of the audio signal after the end of speaking of the audio signal. The time point may be determined as the time when the voice signal starts to be spoken. Since the end point of the talk is accurately determined, the start point of the talk can be accurately determined by this device.

  In the audio signal determination device, the second signal generating means matches the level value of the second signal with the level value of the first signal within a second predetermined period following the start time of speaking of the audio signal. You may do it.

  In order to solve the above-described problem, another audio signal determination device according to the present invention includes a first signal generation unit that generates a first signal, a second signal generation unit that generates a second signal, and a determination unit. The first signal generating means generates the first signal that is an envelope signal of the level value of the audio signal, and the second signal generating means has the level value of the second signal of the audio signal. The second signal is generated by smoothing the first signal on the time axis on the condition that when the level value is equal to or greater than a value obtained by adding a predetermined value to the level value, the determination means includes: A state in which the level value of the second signal is equal to or higher than the level value of the audio signal plus the predetermined value is determined to be a silent state, and the audio is based on the frequency of occurrence of the silent state. It is determined whether the signal is a speech signal or a music signal. In this apparatus, it is possible to accurately determine whether the voice signal is a voice signal or a music signal based on the fact that the frequency of the silent state that appears between the voice signal and the music signal is different.

  In the audio signal determination apparatus, the determination means counts the number of times that a silent state occurs within a third predetermined period, and determines that the audio signal is a speech signal when the counted number is equal to or greater than the predetermined number. The audio signal may be determined to be a music signal when the counted number is less than the predetermined number.

  In the audio signal determination device, the determination means may determine whether the audio signal is based on the time from the end of the previous silent state to the start of the subsequent silent state. You may make it judge whether it is a speech signal or a music signal.

  In the audio signal determination device, the first signal generating means detects the maximum value of the level value of the audio signal in the latest fourth predetermined period for each fourth predetermined period, and the detected maximum value is 4 The first signal may be generated by maintaining only for a predetermined period.

  According to the voice signal determination device according to the present application, it is possible to reliably detect the end of the talk without being affected by the ambient noise of the speaker.

  Moreover, according to another audio signal determination apparatus according to the present application, it is possible to accurately determine whether the audio signal is a speech signal or a music signal.

  Hereinafter, an embodiment of the present invention will be described with reference to the drawings.

  First, an automatic gain adjustment device 20 that employs the configuration of an audio signal determination device according to the present application will be described with reference to FIG.

  FIG. 1 is a block diagram showing a schematic configuration of the acoustic system S. As shown in FIG. The sound system S is installed in a loudspeaker space, and is used to provide announcements and music to the loudspeaker space. A typical example of a loudspeaker space is an airport campus.

  The acoustic system S mainly includes a CD (compact disc) player 11, microphones Ma, Mb, Mc, a changeover switch 15, an automatic gain adjusting device 20, a power amplifier 17, and a speaker 18. In the acoustic system S, the sound source can be switched to any one of the CD player 11, the microphone Ma, the microphone Mb, and the microphone Mc by the changeover switch 15.

  When the announcer's announcement is provided to the loudspeaker space, the microphone Ma, the microphone Mb, or the microphone Mc is selected as a sound source by the changeover switch 15. When the announcer makes an announcement using these microphones Ma, Mb, Mc, the microphones Ma, Mb, Mc generate a speech signal as a voice signal to be input to the automatic gain adjustment device 20.

  When the music is provided to the loud space, the changeover switch 15 selects the CD player 11 as a sound source. Since the music CD is loaded in the CD player 11, the CD player 11 generates a music signal as an audio signal to be input to the automatic gain adjustment device 20.

  The automatic gain adjusting device 20 inputs an audio signal from the sound source (CD player 11, microphones Ma, Mb, Mc) via the changeover switch 15, gives a gain to the audio signal, and sends it to the power amplifier 17. The gain of the automatic gain adjusting device 20 is automatically adjusted based on the input audio signal.

  The power amplifier 17 amplifies the audio signal from the automatic gain adjustment device 20 and sends it to the speaker 18. Announcements and music are radiated from the speaker 18 to the loudspeaker space.

  The automatic gain adjustment apparatus 20 includes an audio signal input unit 21, a signal amplification unit 22, a control signal generation unit 23, and an audio signal output unit 24. The audio signal input unit 21 is a part that inputs an audio signal from a sound source. The audio signal input to the audio signal input unit 21 is sent to the signal amplification unit 22 and the control signal generation unit 23. The signal amplifying unit 22 gives a gain to the input audio signal, and then sends the audio signal to the audio signal output unit 24. The gain of the signal amplifying unit 22 is controlled by an external control signal. The control signal generation unit 23 generates a control signal by performing various calculations and processes based on the input audio signal, and sends the control signal to the signal amplification unit 22. The audio signal output unit 24 sends the audio signal from the signal amplification unit 22 to the power amplifier 17.

  Next, various calculations and processes in the control signal generator 23 will be described with reference to FIG.

  The control signal generation unit 23 generates various signals (level signal, first signal, second signal, control signal to be given to the signal amplification unit 22) from the input audio signal. Various flags (silence detection flag, talk end detection flag, talk start detection flag) are set in the control signal generator 23. Further, the control signal generation unit 23 makes various determinations (determination at the end of speech, determination at the start of speech, determination of whether the audio signal is a speech signal or a music signal) based on these signals and flags. .

  FIG. 2 is a diagram illustrating various signals and various flags generated by various calculations and processes in the control signal generation unit 23. The horizontal axis in FIG. 2 indicates time, and the vertical axis indicates the level of each signal and the state of the flag.

  The control signal generator 23 generates a level signal D indicating the level of the input audio signal. Further, the first signal E is generated based on the level signal D. Further, the second signal F is generated based on the level signal D and the first signal E. Furthermore, the state of the silence detection flag J is determined based on the level signal D and the second signal F. Further, based on the silence detection flag J, the state of the talk end detection flag K is determined. Further, based on the silence detection flag J and the talk end detection flag K, the state of the talk start detection flag L is determined. Note that the second signal F is corrected based on the speech start detection flag L and the first signal E.

  Further, a control signal is generated based on the second signal F. The control signal generator 23 sends this control signal to the signal amplifier 22.

  In FIG. 2, the level signal D is indicated by a broken line, the first signal E is indicated by a solid line, and the second signal F is indicated by a one-dot chain line.

  The level signal D will be further described. The level signal D in the figure is generated based on the speech signal generated when the announcer provides an announcement to the sound space while changing.

  A period indicated by an arrow Pa in the figure (hereinafter referred to as “period Pa”) is a period during which the announcer A made an announcement using the microphone Ma. A period indicated by an arrow Pb in the drawing (hereinafter referred to as “period Pb”) is a period in which the announcer B made an announcement using the microphone Mb. A period indicated by an arrow Pc in the figure (hereinafter referred to as “period Pc”) is a period in which the announcer C made an announcement using the microphone Mc.

  Focusing on the level signal D, it can be seen that there is a difference in the magnitude of the audio signal depending on the announcer. The audio signal in the period Pa is relatively small, and the peak value of the level signal D at that time is about −30 dB. The audio signal in the period Pb is relatively large, and the peak value of the level signal D at that time is about +5 dB. The audio signal in the period Pc is larger, and the peak value of the level signal D at that time is about +10 dB.

  Next, the first signal E will be described.

  The first signal E appears as an envelope of the level signal D. That is, the first signal E is an envelope signal of the level signal D.

  As a method for generating the first signal E, various methods can be employed. As the simplest method among them, a method of obtaining the envelope signal by passing the level signal D through a low-pass filter may be employed. However, in the present embodiment, the first signal E is generated by the following method.

  That is, every fourth predetermined period (P4 seconds), the maximum value of the level signal D in the latest P4 seconds is detected, and the detected maximum value is held for P4 seconds thereafter, thereby generating the first signal E. It is doing. The fourth predetermined period “P4 seconds” may be, for example, not less than 0.1 seconds and not more than 5 seconds. Since the first signal E is generated in this way, the waveform of the first signal E in the figure is stepped. Although it is a stepped waveform, the first signal E envelopes the level signal D.

  The waveform of the first signal E does not show a steep peak or dip seen in the waveform of the level signal D.

  Next, the second signal F will be described.

  The second signal F is generated based on the first signal E and the like. A control signal applied to the signal amplifier 22 is generated based on the second signal F. That is, the control signal is generated so that the gain of the signal amplifier 22 is reduced when the second signal F is large, and the gain of the signal amplifier 22 is increased when the second signal F is small.

  The change of the second signal F varies depending on the state of a talk start detection flag L described later. When the state of the talk start detection flag L is “1”, the control signal generator 23 matches the second signal F with the first signal E. The reason will be described later.

  When the state of the talk start detection flag L is “0”, the second signal F is generated based on the level signal D and the first signal E. More specifically, the second signal F is generated as follows.

  That is, the control signal generation unit 23 constantly observes the second signal F and the level signal D. When the second signal F is larger than the level signal D by a predetermined value (W (dB)) or more, the second signal F is maintained at the current level without being temporally changed. That is, the second signal F does not increase or decrease. The predetermined value “W (dB)” may be, for example, 10 dB or more and 40 dB or less.

  When the level of the second signal F is smaller than the value obtained by adding W (dB) to the level value of the level signal D, the second signal F depends on the magnitude relationship between the second signal F and the first signal E. Change the size of. That is, if the first signal E is larger than the second signal F, the second signal F is increased at a predetermined first change rate (R1 (dB) / second). Here, the magnitude of R1 may be not less than 0.1 and not more than 3. If the first signal E is smaller than the second signal F, the second signal F is decreased at a rate of (R1 (dB) / second). When the second signal F and the first signal E are equal, the magnitude of the second signal F is not changed with time.

  As a result, in a period in which the state of the talk start detection flag L is “0”, the waveform of the second signal F becomes a horizontal line when the second signal F is larger than the level signal D by W (dB) or more. In other parts, the waveform of the second signal F is a waveform obtained by smoothing the waveform of the first signal E on the time axis.

  If the second signal F is larger than the level signal D by W (dB) or more, it can be determined that the announcement is silent, but at that time, the waveform of the second signal F becomes a horizontal line. The gain of the amplifier 22 does not change. Therefore, even if the announcement is suddenly resumed, an appropriate gain can be given to the audio signal.

  When the announcement is not silent, the waveform of the second signal F is a waveform obtained by smoothing the waveform of the first signal E on the time axis. Therefore, the gain of the signal amplifier 22 gradually changes in accordance with the level change of the speech signal due to the announcement. As a result, announcements that are natural and easy to hear can be provided to the loudspeaker space.

  Next, the silence detection flag J will be described.

  The state of the silence detection flag J is determined based on the level signal D and the second signal F. The control signal generator 23 constantly observes the level signal D and the second signal F, and sets the state of the silence detection flag J based on the level of these signals.

  More specifically, when the second signal F is greater than the level signal D by W (dB) or more, the silence detection flag J is set to “1”; otherwise, the silence detection flag J is set to “1”. The state is set to “0”. Since the state of the silence detection flag J is set in this way, when the state of the silence detection flag J is “1”, it can be determined that the audio signal is in the silence state.

  Such a silent state frequently appears in speech such as announcements and conversations. For example, it appears at the breaks of phrases in announcements and conversations, and also appears at the time of speaker change in conversations. Therefore, it is not appropriate to immediately determine that the speech signal has reached the end of the speech due to the detection of the silent state. In the present embodiment, the determination result as to whether or not the voice signal has reached the end point of the speech appears in the state of the end of speech detection flag K described below.

  Next, the talk end detection flag K will be described.

  The initial state of the talk end detection flag K is “0”. The state of the talk end detection flag K is determined based on the silence detection flag J. The control signal generation unit 23 constantly observes the state of the silence detection flag J, and when the state where the silence detection flag J is “1” continues for the first predetermined period (P1 second), the state of “0” Is set to a state of “1”. After that, when the silence detection flag J becomes “0”, the talking end detection flag K is returned to “0”. Here, the first predetermined period (P1 seconds) should be longer than the fourth predetermined period (P4 seconds). The first predetermined period (P1 seconds) may be not less than 0.5 seconds and not more than 10 seconds.

  Thus, the talk end detection flag K changes from “0” to “1” when the silent state continues for P1 seconds. Therefore, it can be determined that the point in time when the state of the talk end detection flag K changes from “0” to “1” is the talk end point in the speech signal.

  As described above, the end point of the talk is determined from the result of the detection of the silent state and the observation without providing a certain gate level. Therefore, it is possible to accurately determine the end point of the talk without being affected by the ambient noise of the speaker.

  Next, the talk start detection flag L will be described.

  The initial state of the talk start detection flag L is “0”. The state of the talk start detection flag L is determined based on the talk end detection flag K. The control signal generator 23 constantly observes the state of the talk end detection flag K, and when the talk end detection flag K changes from “1” to “0”, the talk start detection in the “0” state is detected. The flag L is set to “1”. Thereafter, when the second predetermined period (P2 seconds) has elapsed, the talking start detection flag L is returned to the state of “0”. Here, the second predetermined period (P2 seconds) should be longer than the fourth predetermined period (P4 seconds). The second predetermined period (P2 seconds) may be not less than 0.5 seconds and not more than 10 seconds.

  As described above, the speech start detection flag L changes from “0” to “1” when the speech signal is not silenced after the end of the speech is detected. Therefore, it can be determined that the time when the state of the talk start detection flag L changes from “0” to “1” is the talk start time. In this way, by properly detecting the end of the talk, it is possible to properly detect the start of the talk.

  Next, the relationship between the talk start detection flag L and the second signal F will be described.

  It can be determined that the time point when the talk start detection flag L changes from “0” to “1” is the time point when a new announcement is started. At this time, there is a high possibility that the announcer has been replaced or the microphone to be used has been replaced. Therefore, the gain of the signal amplifier 22 must be quickly adjusted in accordance with the level of the newly input audio signal. For this reason, the control signal generator 23 makes the second signal F coincide with the first signal E only for P2 seconds in which the state of the talk start detection flag L is set to “1”. Thereby, the 2nd signal F can be changed rapidly according to the level of an audio | voice signal. As a result, the gain of the signal amplifier 22 can be quickly changed to a level corresponding to the newly started announcement audio signal.

  Next, the gain of the signal amplifier 22 will be described.

  FIG. 3 is a diagram illustrating the second signal F and the gain G of the signal amplifier 22. The horizontal axis indicates time, and the vertical axis indicates level. The waveform of the second signal F in FIG. 3 is the same as the waveform of the second signal F in FIG.

  The control signal given to the signal amplifier 22 is generated based on the second signal F, and when the second signal F is large, the gain of the signal amplifier 22 is decreased, and when the second signal F is small, the signal is As described above, the gain of the amplifier 22 is increased.

  The gain G (dB) of the signal amplifier 22 is determined as follows. That is, assuming that the level of the second signal F is F (dB), when F is smaller than 0, “G (dB) = 12 (dB)” is set. When F is greater than 12, “G (dB) = 0 (dB)” is set. When F is not less than 0 and not more than 12, “G (dB) = (12−F) (dB)” is set.

  Based on the second signal F, the control signal generator 23 generates a control signal (a control signal to be given to the signal amplifier 22) so that the gain of the signal amplifier 22 is set as described above.

  From FIG. 3, it can be seen that the gain G of the signal amplifier 22 changes quickly in the initial period Pc. This is because the second signal F coincides with the first signal E for P2 seconds in which the state of the talk start detection flag L is set to “1”.

  Next, it will be described that by observing the silence detection flag J, it is possible to identify whether the audio signal input to the automatic gain adjustment device 20 is a speech signal or a music signal.

  The inventors of the present application analyzed changes in the state of the silence detection flag J obtained from various speech signals and changes in the state of the silence detection flag J obtained from various music signals. The present inventors have found that the change in the state of the silence detection flag J obtained from the speech signal and the change in the state of the silence detection flag J obtained from the music signal have different characteristics.

  That is, the silence detection flag J obtained from the speech signal has a higher frequency of changing from “0” to “1”, that is, the frequency of the flag rising, than the silence detection flag J obtained from the music signal. The reason can be estimated as follows.

  As understood from the level signal D in FIG. 2, when the audio signal is a speech signal, a sharp dip often occurs in the level signal D. This dip is caused by paragraph breaks in the announcement. In other words, there is no sound when a sentence breaks in an announcement. Such silence occurs frequently not only in announcements but also in conversations involving multiple speakers. This is because there are clause breaks in the conversation, and a silent state occurs when the speaker changes.

  Compared to speech such as announcements and conversations, music has less silence. Therefore, when the audio signal is a music signal, the frequency of the steep dip appearing in the level signal D is low. As a result, the frequency of the silence detection flag J changing from “0” to “1” is low.

  Using this fact, it is possible to determine (identify) whether the audio signal is a speech signal or a music signal. That is, the determination (identification) can be made based on the silence detection flag J. In other words, it is possible to detect (identify) whether the sound signal is a speech signal or a music signal based on the frequency of the silent state detected. There are several specific methods.

  For example, if the silence detection flag J rises (changes from “0” to “1”) for the third predetermined period (P3 seconds) or more for the first predetermined time (N1 times) or more, the input audio signal May be determined to be a speech signal, and when the number of rising times is less than the first predetermined time, it may be determined that the input speech signal is a music signal. In other words, based on the level signal D and the second signal F, the number of times of silence in a certain time is counted, and based on the counted number, whether the input voice signal is a speech signal or music It is judged (identified) whether it is a signal. Here, the third predetermined period (P3 seconds) should be longer than the fourth predetermined period (P4 seconds). The third predetermined period (P3 seconds) may be not less than 2 seconds and not more than 20 seconds. Further, the first predetermined number of times (N1 times) should be two times or more. The first predetermined times (N1 times) may be 2 times or more and 5 times or less.

  In order to determine that the input audio signal is a music signal, it is necessary to observe the state of the silence detection flag J for at least P3 seconds after the start of the measurement of the third predetermined period (P3 seconds). This is because only after the passage of P3 seconds, it can be known that the number of rises of the silence detection flag J in the P3 seconds is less than N1 times. As a result of observation for P3 seconds, if it is found that the number of rises of the silence detection flag J during that period (P3 seconds) is less than N1, it is determined that the input audio signal is a music signal at that time. be able to.

  In order to determine that the input voice signal is a speech signal, the state of the silence detection flag J is observed for P3 seconds after the start of the measurement of the third predetermined period (P3 seconds), and for P3 seconds. It can also be determined after detecting how many times the silent detection flag J has risen. If the counted number is, for example, (N1 + 2) times, it can be determined that the voice signal is a speech signal.

  However, in order to determine that the input voice signal is a speech signal, it is necessary to observe the state of the silence detection flag J for P3 seconds after the start of the third predetermined period (P3 seconds). There is no. It can be determined that the voice signal is a speech signal when the rise of the silence detection flag J is detected N1 times during the time counting of the third predetermined period (P3 seconds). Then, the timing of the next third predetermined period (P3 seconds) is started from that point.

  Further, as another specific method for detecting the frequency of the silent state and determining (identifying) whether the audio signal is a speech signal or a music signal based on the frequency, the following method is used. There is also.

  That is, after the silent detection flag J changes from “1” to “0”, the time T until the silence detection flag J returns to “1” is observed, and the time T is within T1 seconds which is a predetermined time. As long as the voice signal is input, it can be determined that the input voice signal is a speech signal. After that, when the time T exceeds T1 seconds is continuously observed N2 times, which is the second predetermined time, it can be determined that the voice signal has changed from a voice signal to a music signal. Further, after that, when a state in which the time T is within T1 seconds is continuously observed N3 times that is the third predetermined time, it can be determined that the voice signal has changed from a music signal to a speech signal. That is, the time from the end of the silence state to the start of the next silence state is continuously observed, and based on the observed time, the input speech signal is a speech signal or a music signal. Is identified. FIG. 2 illustrates a case where the time T from when the silence detection flag J changes from “1” to “0” until it returns to “1” is t seconds. Here, the predetermined time (T1 seconds) may be not less than 2 seconds and not more than 10 seconds. The second predetermined time (N2 times) may be 2 times or more and 10 times or less. The third predetermined time (N3 times) may be 2 times or more and 10 times or less.

  As described above, it is possible to detect (identify) whether the sound signal is a speech signal or a music signal based on the frequency of the silent state detected. Such a determination result can be reflected in the adjustment of the gain of the signal amplifier 22. For example, the change speed of the second signal F can be changed depending on whether the audio signal is a speech signal or a music signal. That is, as described above, the second signal F increases at a rate of (R1 (dB) / sec) or decreases at a rate of (R1 (dB) / sec) under a predetermined condition. This change speed is changed depending on the type of the audio signal.

  For example, when it is determined that the audio signal is a speech signal, the change rate of the second signal F is set to the first change rate (R1 (dB) / sec), and it is determined that the audio signal is a music signal. When this is done, the change rate of the second signal F may be the second change rate (R2 (dB) / sec). The magnitude of R2 is smaller than R1. Therefore, in the case of a music signal, the change speed becomes slower. The magnitude of R2 may be not less than 0.01 and not more than 2. As described above, when the change rate of the second signal F is changed depending on the type of the audio signal, the change rate of the gain of the signal amplifying unit 22 also changes depending on the type of the audio signal.

  The reason why the change rate is made slower in the case of a music signal as in the above-described example is to prevent the musicality from being impaired.

  The embodiment of the present invention has been described above. In the above embodiment, an example in which the audio signal determination device according to the present application is used for an automatic gain adjustment device has been described. However, the audio signal determination device according to the present application has not only an automatic gain adjustment device but also other uses. For example, the present invention can be applied to an automatic recording apparatus that starts recording an audio signal at the beginning of a conversation and stops recording at the end of the conversation.

  By using the audio signal determination device of the present application, it is possible to determine the end point of speech in the audio signal, and it is possible to determine the type of the audio signal, which can be used in the technical field of electroacoustics.

It is a block diagram which shows schematic structure of an acoustic system. It is a figure which shows the various signals and various flags which are produced | generated by the various calculation and processing in a control signal production | generation part. It is a figure which shows the 2nd signal and the gain of a signal amplifier.

Explanation of symbols

11 CD player 15 changeover switch 17 power amplifier 18 speaker 20 automatic gain adjustment device 21 audio signal input unit 22 signal amplification unit 23 control signal generation unit 24 audio signal output unit Ma, Mb, Mc microphone D level signal E first signal F first 2 signal G gain S sound system

Claims (7)

A first signal generating means for generating a first signal; a second signal generating means for generating a second signal; and a determining means.
The first signal generating means generates the first signal which is an envelope signal of the level value of the audio signal,
The second signal generating means outputs the first signal as a time on condition that the second signal does not change in time when the level value of the second signal is not less than a value obtained by adding a predetermined value to the level value of the audio signal. Generating the second signal by smoothing on an axis;
The determination means determines a point in time when the state in which the level value of the second signal is equal to or higher than the level value of the audio signal plus the predetermined value continues for a first predetermined period, An audio signal determination device that determines that there is an audio signal.   The determination means determines a point in time when the level value of the second signal becomes smaller than a value obtained by adding the predetermined value to the level value of the voice signal after the voice signal ends. The audio signal determination apparatus according to claim 1, wherein the audio signal determination apparatus determines that it is a point in time when talking is started.   3. The voice according to claim 2, wherein the second signal generating means matches the level value of the second signal with the level value of the first signal within a second predetermined period following the start of speaking of the voice signal. Signal judgment device. A first signal generating means for generating a first signal; a second signal generating means for generating a second signal; and a determining means.
The first signal generating means generates the first signal which is an envelope signal of the level value of the audio signal,
The second signal generating means outputs the first signal as a time on condition that the second signal does not change in time when the level value of the second signal is not less than a value obtained by adding a predetermined value to the level value of the audio signal. Generating the second signal by smoothing on an axis;
The determination means determines that a state where the level value of the second signal is equal to or higher than a value obtained by adding the predetermined value to the level value of the audio signal is a silent state, and the frequency at which the silent state occurs. An audio signal determination device that determines whether the audio signal is a speech signal or a music signal based on   The determination means counts the number of times that a silent state occurs within the third predetermined period, and determines that the voice signal is a speech signal when the counted number is equal to or greater than the predetermined number. The audio signal determination apparatus according to claim 4, wherein the audio signal is determined to be a music signal when the number is less than the predetermined number of times.   The determination means determines whether the audio signal is a speech signal or a music signal based on the time from the end of the previous silent state to the start of the subsequent silent state. The audio signal determination apparatus according to claim 4, wherein it is determined whether there is any. The first signal generating means detects the maximum value of the level value of the audio signal in the latest fourth predetermined period every fourth predetermined period, and maintains the detected maximum value only for the fourth predetermined period. The audio signal determination apparatus according to claim 1, wherein the first signal is generated.

Images