JPH06266383A - Speech segmentation device - Google Patents

Abstract

PURPOSE:To improve the recognition rate of a speech by accurately detecting the end point of the speech and segmenting a signal into a speech signal. CONSTITUTION:This speech segmentation device is equipped with a start point detecting means 2 which detects a start point where speech power calculated from the speech signal increases, a temporary end point detecting means 3 which detects a temporary end point C where the speech power decreases below a 1st threshold value, and a temporary end point moving means 4 which moves the temporary end point sequentially to a position where the speech power becomes less than a 2nd threshold value beyond it after the temporary end point C is detected by the temporary end point detecting means 3. In response to the determination of a final temporary end point as the end point H by the temporary end point moving means 4, the speech signal from the start point A to the end point H is obtained by segmentation and regarded as an object of speech recognition.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a voice cutting device for cutting a voice signal from a signal.

[0002]

2. Description of the Related Art Conventionally, an analog voice signal picked up by a microphone is digitized, and voice power is calculated from the digital voice signal. Regarding this voice power shown in FIG. 5, when the voice power continues to be less than the first threshold for a certain period of time as compared with the first threshold shown in the figure, it is regarded as noise even if the voice power is not zero. Then, the point C ′ that is less than the first threshold value is set as the end point C ′ of the voice, and the voice from the voice start point to the end point C ′ is cut out and the voice recognition is performed. The configuration and operation of FIG. 5 will be briefly described below.

FIG. 5 shows an explanatory view of the prior art. In FIG. 5A, the voice power is the voice power calculated and calculated from the digital voice signal collected by the microphone and digitized. The vertical axis represents voice power strength, and the horizontal axis represents time.

A starting point A'when the voice power shown in the figure, which is calculated by collecting voices with a microphone, is increased to a predetermined value or more is obtained (not shown). Next, the point at which the voice power becomes gradually weaker and remains below the first threshold value for a certain period of time or more is determined as the end point C ′. Then, the voice signal in the section from the start point A ′ to the end point C ′ is cut out and is used as a voice recognition target.

[0005]

When performing the above-described conventional audio segmentation, as shown in FIG. 5B, even if the audio power continues to be less than the first threshold value for a certain period of time or more, There was a problem that it might not finish. As a result, a voice is cut out from only the voice cut-out start point A ′ to the voice cut-out point C ′, and voice recognition is performed, and a problem occurs that the recognition rate is lowered due to an incorrect voice cut-out point C ′. did.

The present invention solves these problems.
The purpose is to accurately detect the end point of the voice and cut out the voice signal to improve the recognition rate of the voice.

[0007]

FIG. 1 is a block diagram showing the principle of the present invention. In FIG. 1, the starting point detecting means 2 detects the starting point A of the audio signal 1.

The temporary end point detecting means 3 has the first voice power.
The provisional end point C that is less than or equal to the threshold is detected. After detecting the temporary end point C, the end point moving means 4 sequentially moves the temporary end point to a position where the temporary end point C exceeds the second threshold value and becomes less than the second threshold value.

The cut-out means 5 cuts out the audio signal from the start point A to the end point H.

[0010]

According to the present invention, as shown in FIG. 1, the starting point detecting means 2 detects the starting point A at which the audio power increases based on the audio power calculated from the audio signal 1 and detects the temporary end point. The position where the means 3 detects the provisional end point C which is less than or equal to the first threshold value for the voice power, and the provisional end point moving means 4 detects the provisional end point C and then exceeds the predetermined second threshold value and becomes less than the predetermined second threshold value. The tentative end point C is repeatedly moved to, and in response to the detection of the final end point H, the cut-out means 5 cuts out a voice signal from the start point A to the end point H, and the voice recognition I am trying to target it.

At this time, the second threshold value is set so that the value gradually increases with the passage of time from the temporary end point C. Therefore, by accurately detecting the end point H of the voice and cutting out the voice signal from the start point A to the end point H to be the target of voice recognition, it is possible to improve the voice recognition rate.

[0012]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Next, the construction and operation of an embodiment of the present invention will be described in detail with reference to FIGS.

FIG. 1 is a block diagram showing the principle of the present invention. Figure 1
In the above, the voice input 1 is a voice signal collected by a microphone and is a voice input that is cut out and is a target of voice recognition.

The start point detecting means 2 calculates the voice power from the voice signal 1 and calculates the start point A of this voice power (see FIG. 4). The temporary end point detection means 3 detects the start point A of the audio power by the start point detection means 2 and then detects the point at which the audio power becomes equal to or less than a preset first threshold value as the temporary end point C. Yes (see FIG. 4).

The end point moving means 4 is a temporary end point detecting means 3
After detecting the tentative end point C, the tentative end point C is sequentially moved to a position where the tentative end point C exceeds a preset second threshold value and then becomes less. Here, the second threshold value is set such that its value gradually increases from the temporary end point C.

The cut-out means 5 cuts out a voice signal from the start point A detected by the start-point detection means 2 to the end point H of the final point moved by the end-point moving means 4. This cut out voice signal is the target of voice recognition.

Next, the operation of the configuration of FIG. 1 will be described in detail with reference to the flowchart of FIG. In FIG. 2, S1
Performs voice input. This is where the operator inputs speech into the macrophone. As a result, the audio signal collected by the microphone is captured.

In step S2, the sound is digitized. this is,
The analog voice signal captured in S1 is converted into a digital voice signal. S3 calculates the voice power. This calculates the voice power from the digital voice signal in S2.

In step S4, the starting point is detected. This is S3
The starting point A is detected from the voice power calculated from the voice signal, for example, the voice power of FIG. S5 is
The comparison with the first threshold value is performed. This compares the audio power with a preset first threshold as shown in FIG.

In step S6, the temporary end point C is detected. this is,
By the comparison in S5, a point at which the voice power becomes smaller than the first threshold value is detected as a temporary end point C (see FIG. 4A).

In step S7, the second threshold value is compared. This is because the provisional end point C is detected in S6, as shown in FIG. 4 (A) described later, and subsequently, the voice power and the second threshold value are compared.

In S8, the end point is moved. This is to move the provisional end point C to a point where the voice power becomes smaller than the second threshold value after the comparison in S7, for example, point F in FIG.
Then, the movement of the temporary end point is repeated.

In step S9, a voice is cut out. This is to repeat the movement of the temporary end point C in S8, obtain the final end point H, and then cut out the voice signal from the start point A to the end point H to be the target of voice recognition.

After converting the voice signal into a digital voice signal as described above, the voice power is calculated and the starting point A of this voice power is obtained. A point that is less than the voice power and the first threshold value is defined as a temporary end point C. Next, the voice power is compared with the second threshold value, and the temporary end point C is repeatedly moved to a point where the voice power becomes less than the second threshold value after exceeding the second threshold value, and the final end point H is obtained. Cut out the detected audio signal from the start point A to the end point H,
The target of voice recognition. With these, even if the signal level at the end of the audio signal becomes small, it is possible to accurately cut out the signal within the range of the audio signal without forgetting to cut out, and as a result, improve the voice recognition rate. Becomes

FIG. 3 shows an operation flowchart of the main part of the present invention. This is a detailed description of the procedure for detecting the final end point H after detecting the temporary end point C in S6 of FIG.

In FIG. 3, S11 detects a temporary end point of voice. This corresponds to S6 of FIG. 2 and detects the tentative end point C at which the audio power of FIG. 4A becomes equal to or lower than the first threshold value.

In step S12, a second threshold value for the time axis is set. For example, as shown in the second threshold in FIG. 4, the second threshold is set so that the value gradually increases with the passage of time from the temporary end point C.

In step S13, it is determined whether the period X has ended. This determines whether the preset period X has elapsed from the temporary end point C and the process has ended. If YES, end (EN
D). If NO, the process proceeds to S14.

In step S14, it is determined whether the voice power value is less than the second threshold value. If YES, the end point of the voice is set in S15 (that is, the temporary end point C is moved to this point and set), and S
Proceed to 16. On the other hand, if NO, the process returns to S13.

In step S16, it is determined whether the voice power value is the second threshold value or more. This determines whether or not the voice power value becomes less than the second threshold value and then becomes equal to or more than the second threshold value in YES in S14. If YES, the process returns to S13. In the case of NO, it is determined in S17 whether or not the period X has ended, when YES is ended (END), and when NO is returned to S16.

As described above, when the voice power value becomes less than the second threshold value, the process of moving the temporary end point C to the point less than this value in S15 is repeated until the end of the period X. As a result, the temporary end point C is sequentially moved to a point that is less than the second threshold value within the period X, and the final value is determined as the end point H.

FIG. 4 shows an operation explanatory diagram of the present invention. Figure 4
(A) shows a change in voice power and a temporary end point C at that time.
The calculation of the end point H is shown. Here, the voice power is the voice power calculated from the voice signal.

The first threshold is for obtaining the temporary end point C. The point at which the voice power becomes less than the first threshold value is defined as the temporary end point C. The second threshold is for calculating the final end point H by sequentially moving the provisional end point C after the provisional end point C, and gradually starting with the provisional end point C as the start point. It has a value that increases.

The starting point A is the starting point of voice power. -Temporary end point C is a point at which the audio power becomes less than the first threshold value. The final end point H is a final end point that is obtained after the provisional end point C is obtained and sequentially moved to a point where the audio power becomes less than the second threshold value within the preset period X.

Next, under the voice power, the first threshold value, the second threshold value, and the period X shown in FIG.
The procedure for obtaining is explained. (1) The starting point A at which the voice power starts to increase is determined.

(2) A tentative end point C at which the voice power increases and is less than the first threshold value is obtained. (3) After the tentative end point C, the tentative end point C is sequentially moved to a point where the audio power becomes less than the second threshold value after exceeding the second threshold value,
Continue until the end of period X, and set the last moved point to the end point H.
To decide.

(4) The starting point A obtained in (1),
A voice signal within the range of the final end point H determined in (3) is cut out and is subjected to voice recognition. In FIG. 4B,
A second threshold example is shown. This is to have a value that gradually increases from the temporary end point C within the period X from the temporary end point C, and here, the change amount is gradually increased (FIG. 4). The second threshold of (A) is linearly and gradually increased). As a result, even if the voice signal is as small as possible in the vicinity of the temporary end point C, the voice recognition rate is improved by incorporating the audio signal into the cutout range. The range is abnormally wide so that the voice recognition rate does not decrease.

[0038]

As described above, according to the present invention,
After the provisional end point C where the voice power is less than the first threshold value is output, the provisional end point is sequentially moved to a position where it exceeds the second threshold value and becomes less than the second threshold value to obtain the final end point H. End point H
Since a configuration is adopted in which a voice signal is cut out within the range of 10 to be the target of voice recognition, the end point H of the voice is accurately detected, and the voice signal from the start point A to the end point H is cut out to perform voice recognition. As a target, the voice recognition rate can be improved. At this time, the second threshold value is set so as to gradually increase from the temporary end point C, a voice signal in the vicinity of the temporary end point C is captured to improve the voice recognition rate, and noise increases as the distance from the temporary end point C increases. It is possible to prevent the voice recognition rate from being lowered by not taking in such information.

[Brief description of drawings]

FIG. 1 is a principle configuration diagram of the present invention.

FIG. 2 is an operation flowchart of the configuration of the embodiment of the present invention.

FIG. 3 is an operation flowchart of a main part of the present invention.

FIG. 4 is an explanatory diagram of the operation of the present invention.

FIG. 5 is an explanatory diagram of a conventional technique.

[Explanation of symbols]

 1: Voice input 2: Starting point detecting means 3: Temporary ending point detecting means 4: Ending point moving means 5: Cutting means A: Starting point C: Temporary ending point H: Final ending point X: From temporary ending point C Period for cutting out audio signal

Claims (2)

[Claims] 1. A voice clipping device for clipping a voice signal from a signal, the voice power calculated from the voice signal, a start point detecting means (2) for detecting a start point A at which the voice power increases, and a voice power. Is a first threshold value or less, a temporary end point detecting means (3) for detecting the temporary end point C, and a predetermined second threshold value after the temporary end point C is detected by the temporary end point detecting means (3). A provisional end point moving means (4) for sequentially moving the provisional end point to a position above and below is provided, and the provisional end point moving means (4) determines the final provisional end point as the end point H. Correspondingly, a voice cut-out device, characterized in that the voice signal from the start point A to the end point H is cut out and used as a target of voice recognition. 2. The audio clipping device according to claim 1, wherein the second threshold value is set such that the value gradually increases with time from the temporary end point C.