JP4349415B2 - Sound signal processing apparatus and program - Google Patents

Description

  The present invention relates to a technique for processing a signal (hereinafter referred to as “sound signal”) indicating various sounds such as voice and musical sound, and in particular, a section (hereinafter referred to as “sound generation”) of a sound signal in which a desired sound is actually generated. (Referred to as “section”).

In voice analysis such as voice recognition and voice authentication (speaker authentication), a technique is used in which a sound signal is divided into a sounding period and a non-sounding period (a period in which only noise corresponding to the environment exists). For example, a section where the S / N ratio of the sound signal exceeds a predetermined threshold is specified as the sound generation section. Further, in Patent Document 1, each section is divided into a sound generation section and a non-sound generation by comparing the S / N ratio of each section into which a sound signal is divided and the S / N ratio of a section determined as a non-sound generation section in the past. A technique for determining which of the sections corresponds is disclosed.
JP 2001-265367 A

  However, in the technique of Patent Document 1, the distinction between the sounding period and the non-sounding period is determined only by comparing the S / N ratio of each section of the sound signal with the S / N ratio in the past non-sounding period. For example, a section where a momentary noise such as a coughing sound of a speaker, lip noise, or a mouth sound (a section that should be determined as a non-sounding section) may be erroneously detected as a sounding section. Against the background of the above circumstances, the present invention aims to solve the problem of improving the specific accuracy of the pronunciation interval.

In order to solve the above problems, a sound signal processing device according to one aspect of the present invention includes frame information generating means for generating frame information for each frame of a sound signal, and frame information generated by the frame information generating means. Storing means, first section specifying means for specifying the first sounding section (for example, the sounding section P1 in FIG. 2) of the sound signal, and each of the first sounding sections specified by the first section specifying means. And second section specifying means for specifying a second sound generation section (for example, the sound generation section P2 in FIG. 2) obtained by shortening the first sound generation section based on the frame information stored by the storage means for the frame.
In one aspect of the present invention, the frame information generating unit generates different types of first frame information and second frame information for each frame of the sound signal, and the first section specifying unit includes the first frame information of the frame. The first sounding section of the sound signal is specified based on the first frame information, and the second section specifying means specifies the second sounding section based on the second frame information of each frame in the first sounding section. .

  According to the above configuration, the second sound generation interval is specified by shortening the first sound generation interval based on the frame information of each frame. Therefore, it is possible to improve the accuracy of specifying the sounding section as compared with a configuration in which the sounding interval is determined in one stage of processing (for example, a configuration in which only the first sounding interval is specified). In addition, although the specific method of specifying the second sound generation section based on the specific contents of the frame information and the frame information is arbitrary in the present invention, for example, the following aspects are adopted.

  In the first aspect, the frame information includes a signal index value corresponding to the signal level of the sound signal in each frame (for example, the signal level HIST_LEVEL and the S / N ratio R in the embodiment). The second section specifying means is at least one of one or more frames continuous from the start point of the first sounding section and one or more frames continuous from the end point of the first sounding section to the near side, and is a signal included in the frame information By excluding a frame whose index value is lower than a threshold value corresponding to the maximum value of the signal index value in the first sounding period (for example, the threshold value TH1 in FIG. 6) from the plurality of frames in the first sounding period, the second sounding period Is identified.

Further, in the first aspect, the second section specifying means is configured such that the added value of the signal index value over a plurality of consecutive frames from the start point of the first sounding section corresponds to the maximum value of the signal index value in the first sounding section. When the threshold value is below the threshold value (for example, the threshold value TH2 in FIG. 6), the second sound generation interval is specified by excluding two or more frames on the start point side from the plurality of frames. Similarly, the second section specifying means is configured such that the added value of the signal index values over a plurality of consecutive frames from the end point of the first sounding section to the near side is a threshold corresponding to the maximum value of the signal index values in the first sounding section. The second sound generation section is specified by excluding two or more frames on the end point side among the plurality of frames.

  As described above, according to the configuration in which the second sound generation interval is specified in accordance with the maximum value of the signal index value in the first sound generation interval, noise (for example, coughing and lip of the speaker) generated before and after the actual sound generation interval. Noise, etc.) can be effectively eliminated. A specific example of the first aspect will be described later as the first embodiment.

  In the second aspect, the frame information includes pitch data indicating a result of detecting the pitch of the sound signal of each frame. The second section specifying means is at least one of one or more frames continuous from the start point of the first sounding section and one or more frames continuous from the end point of the first sounding section to the near side, and the pitch included in the frame information The second sounding interval is specified by excluding the frame whose data is not detected from the first sounding interval. According to the above aspect, it is possible to effectively eliminate noise whose pitch is not clearly specified, such as wind noise. A specific example of the second aspect will be described later as a second embodiment.

  In the third aspect, the frame information includes the number of zero crossings of the sound signal in each frame. The second section specifying means, when a frame in which the number of zero crosses included in the frame information exceeds a threshold value continues over a plurality of frames from the end point of the first sound generation section to the near side, a predetermined number of the start point side of the plurality of frames By excluding frames other than the frame, the second sound generation interval is specified. According to the above aspect, frames (unvoiced consonants) that are a plurality of frames on the near side from the end point of the first sounding section and whose zero-cross number exceeds the threshold value are excluded except for a predetermined number. It is possible to adjust the consonant) to a predetermined time length.

  The sound signal processing device according to a preferred aspect of the present invention includes an acquisition unit (for example, the switching unit 583 in FIG. 3) for acquiring a start instruction, and a noise level of a frame before acquisition of the start instruction by the acquisition unit among the sound signals. The S / N ratio is calculated based on the noise level calculation means to be calculated, and the signal level of each frame of the sound signal after acquisition of the start instruction by the acquisition means and the noise level calculated by the noise level calculation means. The first section specifying means specifies the first sound generation section based on the S / N ratio calculated for each frame by the S / N ratio calculating means. According to the above aspect, since the S / N ratio of each frame after obtaining the start instruction is calculated using each frame before obtaining the start instruction as noise, it is possible to specify the first sound generation section with high accuracy. It is.

The sound signal processing device according to a preferred aspect of the present invention is a feature amount calculation for sequentially calculating, for each frame of a sound signal, a feature amount used by the sound analysis device separate from the sound signal processing device for analysis of the sound signal. And an output control means for sequentially outputting the feature quantity of each frame corresponding to the first sounding section specified by the first section specifying means to the sound analysis apparatus every time the feature quantity calculation means calculates , The second section specifying means notifies the sound analysis device of the second sound generation section. In the above aspect, since the feature values calculated by the feature value calculation means are sequentially output to the sound analysis device, the feature values of all the frames belonging to the first sounding section are held in the sound signal processing device. There is no need. Therefore, there is an effect that the circuit size and processing load of the sound signal processing device are reduced. The above effects are particularly remarkable when the data amount of the frame information of each frame is smaller than the data amount of the feature amount of each frame. Further, since the second sound generation section specified by the second section specifying means is notified to the sound analysis device, the sound analysis device uses the feature value of the frame belonging to the second sound generation section among the feature values acquired from the output control device. Can be selectively used for analysis of sound signals. Therefore, there is an advantage that the accuracy of the analysis of the sound signal by the sound analyzer is improved.

The present invention is also specified as an operation method (sound signal processing method) of the sound signal processing device according to each of the above aspects. The sound signal processing method according to one aspect of the present invention generates frame information for each frame of a sound signal, specifies a first sounding section (for example, the sounding section P1 in FIG. 2) of the sound signal, and specifies the first section. Based on the frame information generated for each frame in the first sounding period specified by the means, the second sounding period (for example, the sounding period P2 in FIG. 2) obtained by shortening the first sounding period is specified. According to the above method, the same operation and effect as the sound signal processing apparatus according to the present invention are exhibited.

The sound signal processing apparatus according to each aspect described above is realized by hardware (electronic circuit) such as a DSP (Digital Signal Processor) dedicated to each process, and general-purpose arithmetic processing such as a CPU (Central Processing Unit). This is also realized by cooperation between the apparatus and the program. The program according to the present invention includes a frame information generation process for generating frame information for each frame of a sound signal, a first section specifying process for specifying a first sound generation section (for example, the sound generation section P1 in FIG. 2) of the sound signal, Based on the frame information generated by the frame information generation process for each frame in the first sound generation section specified by the first section specifying process, the second sound generation section (for example, the sound generation section P2 in FIG. 2) obtained by shortening the first sound generation section. The second section specifying process for specifying is executed by the computer. Even with the above program, the same operation and effect as the sound signal processing apparatus according to the present invention are exhibited. The program of the present invention is provided to a user in a form stored in a portable recording medium such as a CD-ROM and installed in a computer, or provided from a server device in a form of distribution via a network. Installed on the computer.

<A: First Embodiment>
<A-1: Configuration>
FIG. 1 is a block diagram showing a configuration of a sound signal processing system according to one embodiment of the present invention. As shown in the figure, the sound signal processing system includes a sound collection device (microphone) 10, a sound signal processing device 20, an input device 70, and a sound analysis device 80. In the present embodiment, a configuration in which the sound collection device 10, the input device 70, and the sound analysis device 80 are installed separately from the sound signal processing device 20 is illustrated, but a part or all of the above elements are a single device. May be configured.

  The sound collection device 10 generates a sound signal S indicating a waveform of surrounding sound (speech and noise). FIG. 2 illustrates the waveform of the sound signal S. The sound signal processing device 20 specifies a sounding section in which the speaker has actually uttered from the sound signal S generated by the sound collection device 10. The input device 70 is a device (for example, a keyboard or a mouse) that outputs a signal corresponding to a user's operation. The user appropriately operates the input device 70 to input an instruction TR (hereinafter referred to as “start instruction”) that triggers the sound signal processing apparatus 20 to start specifying the sounding section. The sound analysis device 80 is used for analyzing the sound signal S. The sound analysis device 80 according to the present embodiment is a voice authentication device that authenticates the legitimacy of a speaker by comparing a feature amount extracted from the sound signal S with a pre-registered feature amount.

  The sound signal processing device 20 includes a first section specifying unit 30, a second section specifying unit 40, a frame analysis unit 50, an output control unit 62, and a storage unit 64. The first section specifying unit 30, the second section specifying unit 40, the frame analyzing unit 50, and the output control unit 62 may be realized by an arithmetic processing unit such as a CPU executing a program, or a DSP or the like. It may be realized by a hardware circuit.

  The first section specifying unit 30 is a means for specifying the sound generation section P1 shown in FIG. On the other hand, the second section specifying unit 40 is means for specifying the sound generation section P2 in FIG. The method in which the first section specifying unit 30 specifies the sounding section P1 is different from the method in which the second section specifying unit 40 specifies the sounding section P2. The second section specifying unit 40 of the present embodiment specifies the sound generation section P2 by a method with higher accuracy than the specification of the sound generation section P1 by the first section specifying unit 30. Therefore, as shown in FIG. 2, the sound generation period P2 is shorter than the sound generation period P1.

  The frame analysis unit 50 in FIG. 1 includes a division unit 52, a feature amount calculation unit 54, and a frame information generation unit 56. As shown in FIG. 2, the dividing unit 52 divides the sound signal S supplied from the sound collecting device 10 into frames having a predetermined time length (for example, several tens of milliseconds) and sequentially outputs them. Each frame is set to overlap each other on the time axis.

  The feature amount calculation unit 54 calculates a feature amount C for each frame F of the sound signal S. The feature amount C is a parameter used by the sound analysis device 80 to analyze the sound signal S. The feature amount calculation unit 54 of the present embodiment calculates a mel cepstrum coefficient (MFCC) as a feature amount C by frequency analysis including FFT (Fast Fourier Transform) processing. The feature amount C is calculated in real time in synchronization with the supply of the sound signal S of each frame F (that is, sequentially calculated each time each frame of the sound signal S is supplied).

  The frame information generation unit 56 generates frame information F_HIST for each frame F of the sound signal S output from the dividing unit 52. Further, the frame information generation unit 56 of the present embodiment includes a calculation unit 58 that calculates the S / N ratio R for each frame F. The S / N ratio R is information used by the first section specifying unit 30 to specify the sound generation section P1. On the other hand, the frame information F_HIST is information used by the second section specifying unit 40 to shorten the sound generation section P1 to the sound generation section P2. The frame information F_HIST and the S / N ratio R are calculated in real time in synchronization with the supply of the sound signal S of each frame F.

  FIG. 3 is a block diagram illustrating a specific configuration of the calculation unit 58. As shown in the figure, the calculation unit 58 includes a level calculation unit 581, a switching unit 583, a noise level calculation unit 585, a storage unit 587, and an S / N ratio calculation unit 589. The level calculation unit 581 is means for calculating the level (intensity) sequentially for each frame F of the sound signal S supplied from the division unit 52. The level calculation unit 581 of this embodiment is a level for each component FRAME_LEVEL [1] to the level of each component when the sound signal S of one frame F is divided into n (n is a natural number of 2 or more) frequency bands. Calculate FRAME_LEVEL [n]. Therefore, the level calculation unit 581 is realized by a plurality of bandpass filters (filter banks) having different passbands, for example. However, a configuration in which the level calculation unit 581 calculates the band-specific levels FRAME_LEVEL [1] to FRAME_LEVEL [n] by frequency analysis such as FFT processing is also adopted.

The frame information generation unit 56 in FIG. 1 calculates a signal level HIST_LEVEL for each frame F of the sound signal S. The frame information F_HIST of one frame F includes the signal level HIST_LEVEL calculated for the frame F. The signal level HIST_LEVEL is a total value of the band-specific levels FRAME_LEVEL [1] to FRAME_LEVEL [n] as expressed by the following equation (1). The frame information F_HIST of one frame F has a smaller data amount than the feature amount C (for example, MFCC) of one frame F.

  The switching unit 583 in FIG. 3 selectively switches the supply destination of the band-specific levels FRAME_LEVEL [1] to FRAME_LEVEL [n] calculated by the level calculating unit 581 in accordance with the start instruction TR input from the input device 70. It is. More specifically, the switching unit 583 outputs the band-specific levels FRAME_LEVEL [1] to FRAME_LEVEL [n] to the noise level calculation unit 585 before obtaining the start instruction TR, and after obtaining the start instruction TR, the S / S It outputs to N ratio calculation part 589.

  As shown in FIG. 2, the noise level calculation unit 585 is means for calculating noise levels NOISE_LEVEL [1] to NOISE_LEVEL [n] in the period P0 immediately before the switching unit 583 acquires the start instruction TR. The period P0 is a period whose end point is the time point of the start instruction TR, and is composed of a plurality of frames (6 in the example of FIG. 2). The noise level NOISE_LEVEL [i] corresponding to the i-th frequency band is an average value of the band-specific levels FRAME_LEVEL [i] over a predetermined number of frames F within the period P0. The noise levels NOISE_LEVEL [1] to NOISE_LEVEL [n] calculated by the noise level calculation unit 585 are sequentially stored in the storage unit 587.

The S / N ratio calculation unit 589 in FIG. 3 calculates the S / N ratio R for each frame F of the sound signal S and outputs it to the first section specifying unit 30. The S / N ratio R is a numerical value corresponding to the relative ratio between the intensity of each frame F after the start instruction TR and the intensity of noise within the period P0. The S / N ratio calculation unit 589 of this embodiment includes the level FRAME_LEVEL [1] to FRAME_LEVEL [n] for each frame F supplied from the switching unit 583 after the start instruction TR and the noise level NOISE_LEVEL stored in the storage unit 587. The S / N ratio R is calculated from [1] to NOISE_LEVEL [n] based on the following equation (2).

  The S / N ratio R calculated by the above equation (2) is an index indicating the magnitude of the current voice level relative to the level of noise existing around the sound pickup device 10. That is, when the user is not speaking, the S / N ratio R is a value close to “1”, and the S / N ratio R increases from “1” as the volume of the utterance by the user increases. Therefore, the first section specifying unit 30 specifies the sound generation section P1 of FIG. 2 based on the S / N ratio R of each frame F. That is, generally, a set of frames F in which the S / N ratio R exceeds a predetermined value is specified as the sound generation section P1. In this embodiment, since the S / N ratio R is calculated based on the noise level of a predetermined number of frames F immediately before the start instruction TR (that is, immediately before utterance by the speaker), It is possible to reduce the influence of noise.

  As shown in FIG. 1, the first section specifying unit 30 includes a start point specifying unit 32 and an end point specifying unit 34. The start point specifying unit 32 specifies the start point P1_START (FIG. 2) of the sound generation section P1, and generates start point data D1_START for identifying the start point P1_START. The end point specifying unit 34 specifies the end point P1_STOP (FIG. 2) of the sound generation section P1, and generates end point data D1_STOP for identifying the end point P1_STOP. The start point data D1_START is a number assigned to the first frame F of the sound generation interval P1, and the end point data D1_STOP is a number assigned to the last frame F of the sound generation interval P1. As shown in FIG. 2, the sound generation section P1 includes M1 frames (M1 is a natural number). A specific example of the operation of the first section specifying unit 30 will be described later.

  The storage unit 64 is means for storing the frame information F_HIST generated by the frame information generation unit 56. Various storage devices such as a semiconductor storage device, a magnetic storage device, and an optical disk storage device are preferably employed as the storage unit 64. Note that the storage unit 64 and the storage unit 587 may be separate storage areas defined in one storage device, or may be separate storage devices.

  The storage unit 64 of this embodiment selectively stores only the frame information F_HIST of the M1 frames F belonging to the sound generation section P1 among the many pieces of frame information F_HIST calculated sequentially by the frame information generation unit 56. That is, the storage unit 64 starts storing the frame information F_HIST from the frame F corresponding to the start point P1_START when the start point specifying unit 32 specifies the start point P1_START, and when the end point specifying unit 34 specifies the end point P1_STOP. Then, the storage of the frame information F_HIST is terminated with the frame F corresponding to the end point P1_STOP.

  The second section specifying unit 40 specifies the sound generation section P2 of FIG. 2 based on the M1 frame information F_HIST (signal level HIST_LEVEL) stored in the storage unit 64. As shown in FIG. 1, the second section specifying unit 40 includes a start point specifying unit 42 and an end point specifying unit 44. As shown in FIG. 2, the start point specifying unit 42 specifies a time point when the time length (number of frames) corresponding to the frame information F_HIST has elapsed from the start point P1_START of the sound generation interval P1 as the start point P2_START of the sound generation interval P2, and the start point P2_START The start point data D2_START for identifying is generated. The end point specifying unit 44 specifies a time point before the end point P1_STOP of the sound generation period P1 by the time length (number of frames) according to the frame information F_HIST as the end point P2_STOP of the sound generation period P2, and end point data for identifying the end point P2_STOP D2_STOP is generated. The start point data D2_START is the number of the first frame F of the sounding section P2, and the end point data D2_STOP is the number of the last frame F of the sounding section P2. The start point data D2_START and the end point data D2_STOP are output to the sound analyzer 80. As shown in FIG. 2, the sound generation period P2 includes M2 frames (M2 is a natural number) F (M2 <M1). A specific example of the operation of the second section specifying unit 40 will be described later.

  The output control unit 62 in FIG. 1 is means for selectively outputting to the sound analysis device 80 the feature amount C that the feature amount calculation unit 54 sequentially calculates for each frame F. The output control unit 62 of the present embodiment outputs the feature value C of each frame F belonging to the sound generation section P1 to the sound analysis device 80, while discarding the feature value C of each frame F other than the sound generation section P1 (sound analysis device). No output to 80). That is, when the start point specifying unit 32 specifies the start point P1_START, the output control unit 62 starts outputting the feature amount C from the frame F corresponding to the start point P1_START, and for each subsequent frame F, the feature amount calculating unit. The feature amount C is output in real time in synchronization with the calculation by 54 (that is, the feature amount C of each frame F is output to the sound analysis device 80 each time the feature amount C is supplied from the feature amount calculation unit 54). Then, when the end point specifying unit 34 specifies the end point P1_STOP, the output control unit 62 ends the output of the feature amount C with the frame F corresponding to the end point P1_STOP.

  As shown in FIG. 1, the sound analysis device 80 includes a storage unit 82 and a control unit 84. The storage unit 82 stores in advance a set of feature amounts (hereinafter referred to as “registered feature amounts”) extracted from the voice of a specific speaker. Further, the storage unit 82 stores the feature amount C output from the output control unit 62. That is, the feature amount C of each of the M1 frames F belonging to the sound generation section P1 is stored in the storage unit 82.

  The start point data D2_START and the end point data D2_STOP generated by the second section specifying unit 40 are supplied to the control unit 84. The control unit 84 uses the M2 feature values C in the sound generation section P2 defined by the start point data D2_START and the end point data D2_STOP among the M1 feature values C stored in the storage unit 82 to generate the sound signal S. Is analyzed. For example, the control unit 84 uses various pattern matching techniques such as DP matching to calculate the distance between each feature quantity C in the pronunciation section P2 and each registered feature quantity, and based on this calculated distance, the current utterance The legitimacy of the user (whether or not the speaker is an authorized user registered in advance) is determined.

  As described above, in the present embodiment, since the feature value C of each frame F is output to the sound analysis device 80 in real time in parallel with the specification of the sound generation section P1, all of the sounds in the sound generation section P1 are output. It is not necessary for the sound signal processing device 20 to hold the feature amount C of the frame F until the sounding section P1 is confirmed (end point P1_STOP is confirmed). Therefore, the scale of the sound signal processing device 20 can be reduced. Further, in the sound analysis apparatus 80, since each feature quantity C in the sound generation section P2 further narrowing down the sound generation section P1 is used for the analysis of the sound signal S, all the feature quantities C in the sound generation section P1 are targeted. Compared to the configuration in which the analysis of the sound signal S is performed, there are advantages that the processing load by the control unit 84 is reduced and the accuracy of the analysis (for example, the accuracy of authenticating the validity of the speaker) is improved.

<A-2: Operation>
Next, a specific operation of the sound signal processing device 20 will be described with a focus on the process of specifying the sound generation section P1 and the sound generation section P2.

  When the sound signal processing device 20 is activated, the level calculation unit 581 in FIG. 3 continuously calculates the band-specific levels FRAME_LEVEL [1] to FRAME_LEVEL [n] for each frame F of the sound signal S. When the user inputs the start instruction TR from the input device 70 prior to his / her utterance, the noise level calculation unit 585 causes the band-specific levels FRAME_LEVEL [1] to FRAME_LEVEL [n] of a predetermined number of frames F immediately before the start instruction TR. ], Noise levels NOISE_LEVEL [1] to NOISE_LEVEL [n] are calculated and stored in the storage unit 587. On the other hand, the S / N ratio calculation unit 589 sets the level FRAME_LEVEL [1] to FRAME_LEVEL [n] of each frame F after the start instruction TR and the noise level NOISE_LEVEL [1] to NOISE_LEVEL [n] of the storage unit 587. The corresponding S / N ratio R is calculated.

(A) Operation of the first section specifying unit 30
The first section specifying unit 30 starts a process for specifying the sound generation section P1 in response to the start instruction TR. That is, a process in which the start point specifying unit 32 specifies the start point P1_START (FIG. 4) and a process in which the end point specifying unit 34 specifies the end point P1_STOP (FIG. 5) are executed. The details of each process are as follows.

  As shown in FIG. 4, the start point specifying unit 32 clears the start point data D1_START and initializes the variables CNT_START1 and CNT_START2 to zero (step SA1). Next, the start point specifying unit 32 acquires the S / N ratio R of one frame F from the S / N ratio calculating unit 589 (step SA2), and adds “1” to the variable CNT_START2 (step SA3).

  Next, the start point specifying unit 32 determines whether or not the S / N ratio R acquired in Step SA2 exceeds a predetermined threshold value SNR_TH1 (Step SA4). A frame F in which the S / N ratio R exceeds the threshold value SNR_TH1 is likely to be the frame F in the sound generation section P1, but the S / N ratio R suddenly becomes a threshold value due to ambient noise or electrical noise. May exceed SNR_TH1. Therefore, in the present embodiment, as will be described below, among the predetermined number of frames F (hereinafter referred to as “candidate frame group”) starting from the frame F in which the S / N ratio R first exceeds the threshold value SNR_TH1, S / When the number of frames F in which the N ratio R exceeds the threshold value SNR_TH1 exceeds N1, the first frame F is specified as the start point P1_START of the sound generation period P1.

  If the result of step SA4 is affirmative, the start point specifying unit 32 determines whether or not the variable CNT_START1 is zero (step SA5). The variable CNT_START1 being zero means that the current frame F is the first frame F of the candidate frame group. Therefore, if the result of step SA5 is affirmative, the start point specifying unit 32 temporarily sets the start point data D1_START to the number of the current frame F (step SA6) and initializes the variable CNT_START2 to zero (step SA7). That is, the current frame F is assumed as the start point P1_START of the sound generation section P1. On the other hand, if the result of step SA5 is negative, the start point specifying unit 32 proceeds to step SA8 without passing through step SA6 and step SA7.

  The start point specifying unit 32 adds “1” to the variable CNT_START1 (step SA8), and determines whether or not the added variable CNT_START1 exceeds a predetermined value N1 (step SA9). If the result of step SA9 is affirmative, the start point specifying unit 32 determines the number of the frame F temporarily set in the immediately preceding step SA6 as the official start point data D1_START (step SA10). That is, the starting point P1_START of the sound generation section P1 is specified. In step SA10, the start point specifying unit 32 outputs the start point data D1_START to the second section specifying unit 40, and notifies the output control unit 62 and the storage unit 64 of the confirmation of the start point P1_START. With the notification from the first section specifying unit 30, the output of the feature value C by the output control unit 62 and the storage of the frame information F_HIST by the storage unit 64 are started.

  When the result of step SA9 is negative (that is, when the number of frames F whose S / N ratio R exceeds the threshold value SNR_TH1 is still N1 or less in the candidate frame group), the start point specifying unit 32 performs S for the next frame F. After the / N ratio R is acquired (step SA2), the processing after step SA3 is executed. As described above, since the start point P1_START is not determined only by the S / N ratio R of one frame F exceeding the threshold value SNR_TH1, for example, an increase in the S / N ratio R caused by ambient noise or electrical noise is indicated as a sound generation interval. The possibility of misidentifying the start point P1_START of P1 is reduced.

  On the other hand, when the result of step SA4 is negative (that is, when the S / N ratio R is equal to or less than the threshold value SNR_TH1), the start point specifying unit 32 determines whether or not the variable CNT_START2 exceeds a predetermined value N2 (step SA11). ). The fact that the variable CNT_START2 exceeds the predetermined value N2 means that the number of frames F in which the S / N ratio R exceeds the threshold value SNR_TH1 among the N2 frames F of the candidate frame group is N1 or less. Therefore, when the result of step SA11 is affirmative, the start point specifying unit 32 initializes the variable CNT_START1 to zero (step SA12), and proceeds to step SA2. If the S / N ratio R exceeds the threshold value SNR_TH1 immediately after step SA12 (step SA4: YES), the result of step SA5 becomes affirmative and steps SA6 and SA7 are executed. That is, the candidate data group is updated so that the frame F in which the S / N ratio R newly exceeds the threshold value SNR_TH1 starts. On the other hand, if the result of step SA11 is negative, the start point identification unit 32 proceeds to step SA2 without passing through step SA12.

  When the start point P1_START is specified in the process of FIG. 4, the end point specifying unit 34 executes a process of specifying the end point P1_STOP of the sound generation section P1 (FIG. 5). When the number of frames F whose S / N ratio R is lower than the threshold SNR_TH2 exceeds N3, the end point specifying unit 34 specifies the frame F whose S / N ratio R first falls below the threshold SNR_TH2 as the end point P1_STOP.

  As shown in FIG. 5, the end point specifying unit 34 clears the end point data D1_STOP and initializes the variable CNT_STOP to zero (step SB1), and acquires the S / N ratio R from the S / N ratio calculation unit 589. (Step SB2). Next, the end point specifying unit 34 determines whether or not the S / N ratio R acquired in Step SB2 is lower than a predetermined threshold value SNR_TH2 (Step SB3).

  If the result of step SB3 is affirmative, the end point specifying unit 34 determines whether or not the variable CNT_STOP is zero (step SB4). If the result of step SB4 is affirmative, the end point specifying unit 34 temporarily sets the end point data D1_STOP to the number of the current frame F (step SB5). On the other hand, if the result of step SB4 is negative, the end point specifying unit 34 proceeds to step SB6 without passing through step SB5.

  Next, the end point specifying unit 34 adds “1” to the variable CNT_STOP (step SB6), and determines whether or not the variable CNT_STOP after the addition exceeds a predetermined value N3 (step SB7). If the result of step SB7 is affirmative, the end point specifying unit 34 determines the number of the frame F temporarily set in the immediately preceding step SB5 as formal end point data D1_STOP (step SB8). That is, the end point P1_STOP of the sound generation section P1 is specified. In step SB8, the end point specifying unit 34 outputs the end point data D1_STOP to the second section specifying unit 40, and notifies the output control unit 62 and the storage unit 64 of the confirmation of the end point P1_STOP. With the notification from the first section specifying unit 30, the output of the feature value C by the output control unit 62 and the storage of the frame information F_HIST by the storage unit 64 are completed. Therefore, at the stage where the processing of FIG. 5 is completed, the frame information F_HIST (signal level HIST_LEVEL) is stored in the storage unit 64 and the storage unit of the sound analysis device 80 for each of the M1 frames F belonging to the sound generation period P1. 64, the feature amount C is stored.

  When the result of step SB7 is negative (that is, when the number of frames F in which the S / N ratio R falls below the threshold value SNR_TH2 is N3 or less), the end point specifying unit 34 acquires the S / N ratio R for the next frame F (Step SB2), the processing after Step SB3 is executed. As described above, if the S / N ratio R of one frame F is less than the threshold value SNR_TH2, the end point P1_STOP is not determined. Therefore, the possibility that the point when the S / N ratio R suddenly decreases is mistaken as the end point P1_STOP is reduced. Is done.

  On the other hand, when the result of step SB3 is negative, the end point specifying unit 34 determines whether or not the current S / N ratio R exceeds the threshold value SNR_TH1 used for specifying the start point P1_START (step SB9). When the result of step SB9 is negative, the end point specifying unit 34 proceeds to step SB2 and acquires a new S / N ratio R.

  By the way, the S / N ratio R when the user speaks basically exceeds the threshold value SNR_TH1. Therefore, when the S / N ratio R exceeds the threshold value SNR_TH1 after the processing of FIG. 5 is started (step SB9: YES), there is a high possibility that the user is speaking. Therefore, if the result of step SB9 is affirmative, the end point specifying unit 34 initializes the variable CNT_STOP to zero (step SB10), and then executes the processing after step SB2. If the S / N ratio R falls below the threshold value SNR_TH2 after execution of step SB10 (step SB3: YES), the result of step SB4 becomes affirmative and step SB5 is executed. That is, even when the end point data D1_STOP is provisionally set because the S / N ratio R is lower than the threshold value SNR_TH2, the number of frames F whose S / N ratio R is lower than the threshold value SNR_TH2 is at a stage where the predetermined value N3 or less. When the S / N ratio R of one frame F exceeds the threshold value SNR_TH1 (that is, when there is a high possibility that the user is speaking), the temporary setting of the end point data D1_STOP is cancelled.

(B) Operation of second section specifying unit 40
In order to reliably detect the section where the speaker actually uttered (that is, to reliably prevent detection omission), for example, the threshold value SNR_TH1 in FIG. 4 is set to a relatively small value and the threshold value SNR_TH2 in FIG. 5 is set. It must be set to a relatively large value. Therefore, for example, when noise such as coughing of the speaker, lip noise, or mouth sound is generated prior to the actual utterance, the time when the noise is generated may be recognized as the start point P1_START of the pronunciation period P1. Therefore, the second section specifying unit 40 sequentially selects the frames F that are highly likely to be noise after the first section specifying unit 30 specifies the sound generation section P1, starting from the first and last frames F of the sound generation section P1. By excluding (that is, shortening the sound generation section P1), the sound generation section P2 is specified.

  FIG. 6 is a flowchart showing the contents of processing executed by the start point specifying unit 42 of the second section specifying unit 40. The start point specifying unit 42 of the second section specifying unit 40 specifies the maximum value MAX_LEVEL of the signal level HIST_LEVEL from among the M1 pieces of frame information F_HIST stored in the storage unit 64 (step SC1). Next, the start point specifying unit 42 initializes the variable CNT_FRAME to zero and sets a threshold value TH1 corresponding to the maximum value MAX_LEVEL (step SC2). The threshold value TH1 in this embodiment is a multiplication value of the maximum value MAX_LEVEL specified in step SC1 and the coefficient α. The coefficient α is a numerical value less than “1” set in advance.

  Next, the start point specifying unit 42 selects one frame F from among the M1 frames F in the sound generation section P1 (step SC3). The start point specifying unit 42 of the present embodiment sequentially selects each frame F in the sound generation section P1 from the head toward the tail at every step SC3. That is, in the first step SC3 after starting the processing of FIG. 6, the first frame F of the sound generation section P1 is selected, and in the next step SC3, the frame immediately after the frame F selected in the previous step SC3. F is selected.

  Next, the start point specifying unit 42 determines whether or not the signal level HIST_LEVEL of the frame information F_HIST corresponding to the frame F selected in Step SC3 is lower than the threshold value TH1 (Step SC4). Since the noise level is smaller than the maximum value MAX_LEVEL, the frame F in which the signal level HIST_LEVEL is lower than the threshold value TH1 is likely to be noise generated immediately before the original utterance. Therefore, if the result of step SC4 is affirmative, the start point specifying unit 42 excludes the frame F selected in step SC3 from the sound generation interval P1 (step SC5). More specifically, the start point specifying unit 42 selects the frame F immediately after the frame F selected in step SC3 as a temporary start point p_START. Next, the start point specifying unit 42 initializes the variable CNT_FRAME to zero (step SC6), and proceeds to step SC3. In step SC3, a frame F immediately after the currently selected frame F is newly selected.

  When the result of step SC4 is negative (that is, when the signal level HIST_LEVEL is greater than or equal to the threshold value TH1), the start point specifying unit 42 adds “1” to the variable CNT_FRAME (step SC7) and then adds the variable CNT_FRAME after the addition. It is determined whether or not exceeds a predetermined value N4 (step SC8). If the result of step SC8 is negative, the start point identifying unit 42 proceeds to step SC3 and selects a new frame F. On the other hand, when the result of step SC8 is affirmative, the start point specifying unit 42 shifts the process to step SC9. That is, if the determination in step SC4 (HIST_LEVEL <TH1) is denied continuously over the number of frames F exceeding N4, the process proceeds to step SC9.

  In step SC9, the start point specifying unit 42 sets the threshold value TH2 according to the maximum value MAX_LEVEL specified in step SC1. The threshold value TH2 in this embodiment is a multiplication value of the maximum value MAX_LEVEL and a predetermined coefficient β.

  Next, the start point specifying unit 42 generates a plurality of frames F after the provisional start point p_START at the current stage in the sound generation section P1 (that is, after the removal of some frames F on the head side after step SC5) A predetermined number of consecutive frames F are selected from the section P1) (step SC10). FIG. 7 is a conceptual diagram showing a set G (G1, G2, G3,...) Of frames F selected in step SC10. As shown in the figure, in the first step SC10 after the processing of FIG. 6 is started, a set G1 of a predetermined number of frames F from the head is selected.

  Next, the start point specifying unit 42 calculates an addition value SUM_LEVEL for the signal level HIST_LEVEL of the predetermined number of frames F selected in step SC10 (step SC11). Further, the start point specifying unit 42 determines whether or not the added value SUM_LEVEL calculated in step SC11 is lower than the threshold value TH2 calculated in step SC9 (step SC12).

  As described with reference to FIG. 4, in this embodiment, when the number of frames F in which the S / N ratio R exceeds the threshold value SNR_TH1 exceeds N1 in the candidate frame group, the first frame F is the start point P1_START of the sound generation interval P1. Identified as Therefore, when noise occurs over a plurality of frames F in the candidate frame group, the head of the candidate frame group can be recognized as the start point P1_START. On the other hand, since the noise level is sufficiently small compared to the maximum value MAX_LEVEL, the frame F in which the added value SUM_LEVEL of the signal level HIST_LEVEL over the predetermined number of frames F is lower than the threshold value TH2 is noise generated immediately before the original sound generation. Probability is high.

  Therefore, if the result of step SC12 is affirmative, the start point specifying unit 42 excludes the first half of the frames F from the set G selected in step SC10 as shown in FIG. 7 (step SC13). In other words, the first frame F in the latter half of the set G is selected as the temporary start point p_START. Next, the start point specifying unit 42 shifts the process to step SC10, and as shown in FIG. 7, selects a set G2 of a predetermined number of frames F from the beginning at the current stage, and executes the processes after step SC11.

  On the other hand, if the result of step SC12 is negative, the start point specifying unit 42 determines the start point p_START set at the current stage as the start point P2_START, and analyzes the start point data D2_START specifying the start point P2_START (frame number). The data is output to the device 80 (step SC14). For example, as shown in FIG. 7, when the result of step SC12 is negative when the set G3 is selected, the head of the set G3 (the head in the latter half of the set G2) is specified as the start point P2_START.

  The end point specifying unit 44 of the second section specifying unit 40 specifies the end point P2_STOP by sequentially excluding each frame F of the sound generation section P1 from the tail by the same process as in FIG. That is, the end point specifying unit 44 sequentially selects each frame F of the sound generation section P1 from the tail toward the top for each step SC3, and excludes the frame F when the signal level HIST_LEVEL is lower than the threshold value TH1 ( Step SC5). In addition, the end point specifying unit 44 selects a set G of a predetermined number of frames F continuous from the end to the near side (step SC10) and calculates an addition value SUM_LEVEL of the signal level HIST_LEVEL (step SC11). The end point specifying unit 44 excludes the frame F in the latter half of the set G when the addition value SUM_LEVEL is lower than the threshold value TH2 (step SC13), and when the addition value SUM_LEVEL is equal to or higher than the threshold value TH2, End point data D2_STOP that designates the last frame F as the end point P2_STOP of the sound generation section P2 is output to the sound analysis device 80 (step SC14).

  As described above, the maximum value MAX_LEVEL of the signal level HIST_LEVEL in the sounding section P1 is determined at the stage where the second section specifying unit 40 specifies the sounding section P2. Therefore, by using the maximum value MAX_LEVEL as exemplified above, the second section specifying unit 40 has to specify the pronunciation section P1 when the maximum value MAX_LEVEL is not yet determined. Therefore, it is possible to specify the pronunciation period P2 with higher accuracy. That is, the second section specifying unit 40 excludes the frame F included in the sound generation section P1 due to noise such as coughing of the speaker or lip noise. Therefore, in the sound analysis device 80, the sound signal S is analyzed with high accuracy by using each frame F of the sound generation section P2 from which the influence of noise is eliminated.

  In the above embodiment, the configuration in which the signal level HIST_LEVEL is used as the frame information F_HIST is illustrated, but the content of the frame information F_HIST is appropriately changed. For example, the signal level HIST_LEVEL in the above operation may be replaced with the S / N ratio R calculated by the S / N ratio calculation unit 589 for each frame F. That is, the frame information F_HIST used by the second section specifying unit 40 to specify the sound generation section P2 may be a numerical value (signal index value) corresponding to the level of the signal of the sound signal S. Is unquestionable.

<B: Second Embodiment>
Next, a second embodiment of the present invention will be described. In addition, about the element which an effect | action and function are common in 1st Embodiment in this form, the same code | symbol as the above is attached | subjected and each detailed description is abbreviate | omitted suitably.

  The sound signal S when the wind generated outside or the nasal breath of the speaker is blown to the sound collection device 10 (that is, when the wind noise is collected) maintains a high level for a long time. Therefore, the first section specifying unit 30 may recognize the section where the wind noise is generated as the sound generation section P1 even though the section is not actually speaking by the speaker. Therefore, the second section specifying unit 40 of the present embodiment specifies the sound generation section P2 by excluding the frames that have a high possibility of wind noise from the sound generation section P1.

  The frame information generation unit 56 of this embodiment detects the pitch for each frame F of the sound signal S, and generates pitch data HIST_PITCH indicating the result of this detection. The frame information F_HIST stored in the storage unit 64 includes pitch data HIST_PITCH together with the signal level HIST_LEVEL similar to that in the first embodiment. The pitch data HIST_PITCH indicates the pitch when a clear pitch is detected for the frame F of the sound signal S, and indicates non-detection of the pitch when a clear pitch is not detected for the sound signal S (for example, Set to zero). Since the human voice can basically detect the pitch if the level is high, pitch data HIST_PITCH including the pitch is generated. On the other hand, since a clear pitch is not detected for wind noise that does not have a regular overtone structure, pitch data HIST_PITCH indicating non-detection of pitch is generated when wind noise is collected.

  Next, FIG. 8 is a flowchart showing the operation of the start point specifying unit 42 in the second section specifying unit 40. The start point identifying unit 42 initializes the variable CNT_FRAME to zero (step SD1), and then selects one frame F from the sound generation section P1 (step SD2). Each frame F is selected in order for each step SD2 from the head to the tail of the sound generation section P1. Next, the start point specifying unit 42 determines whether or not the signal level HIST_LEVEL included in the frame information F_HIST of the frame F selected in Step SD2 exceeds a predetermined threshold L_TH (Step SD3).

  If the result of step SD3 is affirmative, the start point specifying unit 42 determines whether or not the pitch data HIST_PITCH included in the frame information F_HIST of the frame F selected in step SD2 indicates pitch non-detection (step SD4). . If the result of step SD4 is affirmative, the start point identification unit 42 adds “1” to the variable CNT_FRAME (step SD5), and then determines whether or not the variable CNT_FRAME after the addition exceeds a predetermined value N5 ( Step SD6). The sound signal S when only the wind noise is collected maintains a high level continuously over a plurality of frames F, and the pitch is not detected. Therefore, when the result of step SD6 is affirmative (that is, when the determinations of step SD3 and step SD4 are continuously affirmed over the frame F exceeding N5), the start point specifying unit 42 selects at this stage. A predetermined number ((N5 + 1)) of frames F up to frame F are excluded (step SD7), and the process proceeds to step SD1. That is, the start point specifying unit 42 selects the frame F immediately after the frame F selected in the immediately preceding step SD2 as the temporary start point p_START. On the other hand, when the result of step SD6 is negative (when the number of consecutive frames F satisfying the conditions of step SD3 and step SD4 is N5 or less), the start point specifying unit 42 proceeds to step SD2. After selecting a new frame F, the processing after step SD3 is executed.

  On the other hand, if the result of either step SD3 or step SD4 is negative (ie, it is unlikely that the sound of frame F is only a wind noise), the first frame F at the current stage is selected as the start point P2_START. The That is, the start point specifying unit 42 determines the provisional start point p_START as the start point P2_START, and outputs start point data D2_START specifying the start point P2_START to the sound analysis device 80 (step SD8).

  The end point specifying unit 44 of the second section specifying unit 40 specifies the end point P2_STOP by sequentially excluding each frame F of the sound generation section P1 from the tail by the same process as in FIG. That is, the end point specifying unit 44 sequentially selects each frame F of the sound generation section P1 from the tail toward the head for each step SD2, while in step SD7, the determinations in steps SD3 and SD4 are continuously affirmed. The predetermined number of frames F is excluded. In step SD8, end point data D2_STOP for designating the last frame F at the time as the end point P2_STOP is generated. According to the above embodiment, the frame F recognized as the sound generation section P1 due to the influence of wind noise is excluded. Therefore, the accuracy of the analysis of the sound signal S by the sound analysis device 80 can be improved.

<C: Third Embodiment>
Next, a third embodiment of the present invention will be described. In addition, about the element which an effect | action and function are common in 1st Embodiment in this form, the same code | symbol as the above is attached | subjected and each detailed description is abbreviate | omitted suitably.

  The sound analysis device 80 authenticates the speaker by comparing the registered feature value extracted when the authorized user utters a specific word (password) with the feature value C extracted from the sound signal S. . In order to maintain the accuracy of authentication, it is desirable that the time length of the phoneme at the end of the password is the same at the time of authentication and at the time of registration, but in practice the time length of the unvoiced consonant corresponding to the end of the password is Fluctuates with each certification. Therefore, in this embodiment, a plurality of frames F continuous from the end point P1_STOP of the sound generation interval P1 to the near side are excluded so that the unvoiced consonant at the end of the password at the time of authentication is unified to a predetermined time length.

  The frame information generation unit 56 of this embodiment generates the zero cross number HIST_ZXCNT of the sound signal S of each frame F as the frame information F_HIST. The zero cross number HIST_ZXCNT is the number of times that the level of the sound signal S in one frame F fluctuates across the reference value (zero). When the sound collected by the sound collection device 10 is an unvoiced consonant, the zero cross number HIST_ZXCNT of each frame F is a large numerical value.

  FIG. 9 is a flowchart showing the operation of the end point specifying unit 44 in the second section specifying unit 40, and FIG. 10 is a conceptual diagram for explaining the processing of the end point specifying unit 44. The end point specifying unit 44 initializes the variable CNT_FRAME to zero (step SE1), and then selects one frame F in the sounding section P1 (step SE2). Each frame F is selected in order for each step SE2 from the tail to the head of the sound generation section P1. Next, the end point specifying unit 44 determines whether or not the zero cross number HIST_ZXCNT included in the frame information F_HIST of the frame F selected in Step SE2 exceeds a predetermined threshold value Z_TH (Step SE3). The threshold value Z_TH is set experimentally or statistically so that the determination in step SE3 is affirmed when the sound signal S of the frame F is an unvoiced consonant.

  If the result of step SE3 is affirmative, the end point specifying unit 44 excludes the frame F selected in step SE2 from the sounding section P1 (step SE4). That is, the end point specifying unit 44 selects the frame F immediately before the frame F selected in step SE2 as the provisional end point p_STOP. Further, the end point specifying unit 44 shifts the process to step SE1, initializes the variable CNT_FRAME to zero, and executes the processes after step SE2.

  On the other hand, if the result of step SE3 is negative, the end point specifying unit 44 adds “1” to the variable CNT_FRAME (step SE5), and determines whether or not the variable CNT_FRAME after the addition exceeds the predetermined value N6 ( Step SE6). If the result of step SE6 is negative, the end point specifying unit 44 proceeds to step SE2.

  Since the variable CNT_FRAME is initialized to zero when the zero cross number HIST_ZXCNT exceeds the threshold Z_TH (step SE1), the determination at step SE6 is that the zero cross number HIST_ZXCNT is continuously below the threshold Z_TH over the frame F exceeding N6. If you are affirmed. If the result of step SE6 is affirmative, the end point specifying unit 44 determines the time point after a predetermined time length T from the last frame F (provisional end point p_STOP) at the current stage as the end point P2_STOP of the sound generation interval P2. After that, the end point data D2_STOP is output (step SE7). For example, when a plurality of (12) frames F are removed from the end point of the sound generation section P1 by repeating step SE4 as shown in FIG. 10, the end point is the time point that has passed the time length T from the last frame F after the removal. Confirm as P2_STOP.

  As described above, in the present embodiment, since the voice at the end of the password (unvoiced consonant) at the time of authentication is adjusted to a predetermined time length T regardless of the actual utterance of the speaker, Compared with the case where the feature amount C of all the frames F is used, it is possible to improve the accuracy of authentication by the sound analysis device 80.

<D: Modification>
Various modifications can be made to the above embodiment. An example of a specific modification is as follows. In addition, you may combine each following aspect suitably.

(1) Various known techniques can be employed to specify the sound generation section P1 by the first section specifying unit 30. For example, a configuration in which a set of a plurality of frames F whose sound volume (energy) exceeds a predetermined threshold in the sound signal S is specified as the sound generation section P1. In the configuration in which the start and end of sound generation are instructed by the user from the input device 70, the section from the start instruction to the end instruction may be specified as the sound generation section P1.

  Similarly, the method by which the second section specifying unit 40 specifies the sound generation section P2 is changed as appropriate. For example, a configuration in which the second section specifying unit 40 includes only one of the start point specifying unit 42 and the end point specifying unit 44 is also employed. In the configuration in which the second section specifying unit 40 includes only the start point specifying section 42, the section from the start point P2_START to the end point P1_STOP obtained by retreating the start point P1_START of the sound generation section P1 is specified as the sound generation section P2. Similarly, in the configuration in which the second section specifying unit 40 includes only the end point specifying unit 44, the section from the start point P1_START to the end point P2_STOP of the sound generation section P1 is specified as the sound generation section P2.

  A configuration in which the second section specifying unit 40 (the start point specifying unit 42 or the end point specifying unit 44) executes only one of the processing up to step SC8 and the processing after step SC9 in FIG. 6 is also adopted. Furthermore, you may combine suitably the operation | movement of the 2nd area specific | specification part 40 in each form. For example, a configuration in which the second section specifying unit 40 specifies the start point P2_START or the end point P2_STOP based on both the signal level HIST_LEVEL (first embodiment) and the zero cross number HIST_ZXCNT (third embodiment) is employed.

  In the second embodiment, the frame F is excluded when both the condition that the signal level HIST_LEVEL exceeds the threshold L_TH (step SD3) and the condition that the pitch data HIST_PITCH indicates non-detection (step SD4) are satisfied. However, only the condition of step SD4 may be determined. As can be understood from the above examples, the second section specifying unit 40 may be any means for specifying a pronunciation section P2 shorter than the pronunciation section P1 based on the frame information F_HIST generated for each frame F.

(2) In the above embodiments, the storage unit 64 starts or ends the storage of the frame information F_HIST when the start point P1_START and the end point P1_STOP are determined. However, the frame information generation unit 56 sets the start point P1_START at the start point P1_START. The same effect can be obtained in the configuration in which generation of the frame information F_HIST is started with the confirmation and the generation of the frame information F_HIST is ended with the confirmation of the end point P1_STOP.

  However, the object stored in the storage unit 64 is not limited to the frame information F_HIST in the sound generation section P1. That is, the frame information F_HIST generated for all the frames F of the sound signal S may be stored in the storage unit 64. However, according to the configuration in which only the frame information F_HIST in the sound generation section P1 is stored in the storage unit 64 as in the above embodiments, there is an advantage that the capacity required for the storage unit 64 is reduced.

(3) Information for designating the start point (P1_START, P2_START) and the end point (P1_STOP, P2_STOP) is not limited to the frame F number. For example, the start point data (D1_START, D2_START) and the end point data (D1_STOP, D2_STOP) may be data that designates the start point and the end point at a time based on a predetermined time point (for example, when the start instruction TR is generated).

(4) The trigger for generating the start instruction TR is not limited to the operation on the input device 70. For example, when a notification (notification by an image or sound) that prompts the user to start pronunciation is executed from the sound signal processing system, a configuration in which a start instruction TR is generated using the notification as a trigger is also employed.

(5) The content of the sound analysis by the sound analysis device 80 is arbitrary. For example, speaker recognition that identifies a speaker by comparing registered feature values extracted for a plurality of users and a speaker's feature value C, or phonemes (character data) uttered by a speaker is represented by a sound signal S. The sound analysis device 80 may execute voice recognition specified from the following. The technique of specifying the sound generation section P2 (excluding the section of only noise from the sound signal S) as in each of the above forms is suitably employed for improving accuracy in any sound analysis. Further, the content of the feature amount C is appropriately selected according to the content of the processing by the sound analysis device 80, and the mel cepstrum coefficient in each of the above embodiments is merely an example of the feature amount C. For example, the sound signal S divided into the frames F may be output to the sound analysis device 80 as the feature amount C.

1 is a block diagram showing a configuration of a sound signal processing system according to a first embodiment of the present invention. It is a conceptual diagram which shows the relationship between the sound signal S and sound generation area (P1, P2). It is a block diagram which shows the specific structure of a calculating part. It is a flowchart which shows the process which specifies the starting point of the sound production area P1. It is a flowchart which shows the process which specifies the end point of the sound production area P1. It is a flowchart which shows the process which pinpoints pronunciation area P2. It is a conceptual diagram for demonstrating the process which specifies the sound production area P2. It is a flowchart which shows the process which specifies the sound generation area P2 in 2nd Embodiment. It is a flowchart which shows the process which specifies the sound generation area P2 in 3rd Embodiment. It is a conceptual diagram for demonstrating the process which specifies the sound generation area P2 in 3rd Embodiment.

Explanation of symbols

DESCRIPTION OF SYMBOLS 10 ... Sound collecting device, 20 ... Sound signal processing device, 30 ... 1st area specific | specification part, 40 ... 2nd area specific part, 32, 42 ... Start point specific part, 34, 44 ... End point specific part , 50... Frame analysis unit, 52... Division unit, 54... Feature amount calculation unit, 56... Frame information generation unit, 58 ... calculation unit, 581 ... level calculation unit, 583 ... switching unit, 585 ...... Noise level calculation unit, 587, 64 ... Storage unit, 589 ... S / N ratio calculation unit, 62 ... Output control unit, 70 ... Input device, 80 ... Sound analysis device, 82 ... Storage unit , 84: Control unit, S: Sound signal, F: Frame, TR: Start instruction, F_HIST: Frame information, R: S / N ratio, C: Feature quantity, P1, P2: Sound generation section.

Claims (4)

Frame information generating means for generating different types of first frame information and second frame information for each frame of the sound signal;
First section specifying means for specifying a first sound generation section of the sound signal based on the first frame information of each frame;
Second section specifying means for specifying a second sounding section obtained by shortening the first sounding section based on the second frame information of each frame in the first sounding section specified by the first section specifying means; Sound signal processing device. The second frame information includes the number of zero crossings of the sound signal in each frame,
The second section specifying means, when a frame in which the number of zero crosses included in the second frame information exceeds a threshold value continues over a plurality of frames from the end point of the first sound generation section to the near side, starts from the plurality of frames. The sound signal processing device according to claim 1, wherein the second sound generation section is specified by excluding frames other than the predetermined number of frames on the side. A feature amount calculating means for sequentially calculating, for each frame of the sound signal, a feature amount used by the sound analysis device separate from the sound signal processing device for analysis of the sound signal;
Output control means for sequentially outputting the feature quantities of each frame in the first sound generation section specified by the first section specifying means to the sound analysis apparatus every time the feature quantity calculation means calculates,
The sound signal processing apparatus according to claim 1, wherein the second section specifying unit notifies the sound analysis apparatus of the second sound generation section . Frame information generation processing for generating different types of first frame information and second frame information for each frame of the sound signal;
A first section specifying process for specifying a first sound generation section of the sound signal based on the first frame information of each frame;
A second section specifying process for specifying a second sounding section obtained by shortening the first sounding section based on the second frame information of each frame in the first sounding section specified by the first section specifying process; A program to be executed.

Images