JPH06105400B2 - Voice recognition system - Google Patents

Description

Description: TECHNICAL FIELD The present invention relates to a speech recognition system, and in particular, pattern matching is performed while shifting an end point of an input pattern in the time direction at each time, and a distance from a standard pattern is calculated, and each time is calculated. The present invention relates to a voice recognition system that evaluates a distance for each and determines a recognition result.

2. Description of the Related Art Conventionally, only the condition defined for the evaluation distance is used as the end condition of the recognition result determination used by the recognition result determination unit of this type of voice recognition method.

In the recognition result determination unit of the conventional voice recognition method described above, only the condition defined regarding the evaluation distance is used as the end condition of the recognition result determination, and therefore the recognition result determination cannot be completed at the optimum position. Therefore, there is a drawback that incorrect recognition may be caused and a correct recognition result cannot be obtained.

OBJECT OF THE INVENTION It is an object of the present invention to provide a voice recognition system which ends recognition result determination at an optimum position and obtains a correct recognition result.

According to the present invention, a method of determining a voice recognition result by providing an evaluation distance threshold value and an evaluation distance time threshold value for an evaluation distance that is time-series data, that is, the power of an input voice that is time-series data. A power threshold value and a power duration time threshold value are provided for the input voice evaluation distance satisfying the condition defined by the evaluation distance threshold value and the evaluation distance duration time threshold value, and the input voice power is the power threshold value and the power duration time. A voice recognition system is obtained in which the recognition result determination is ended and the recognition result is output when the condition defined by the threshold value is satisfied.

Example Next, an example of the present invention will be described with reference to the drawings.

FIG. 1 is a block diagram of an embodiment of the present invention. The embodiment of the present invention includes a voice analysis unit 1, a standard pattern unit 2, a pattern matching unit 3, a recognition result determination unit 4, and a power calculation unit 5.

The input voice 10 is input to the voice analysis unit 1 and the power calculation unit 5, analyzed by the voice analysis unit 1 into a characteristic pattern, and input as an input pattern 11 to the pattern matching unit 3. The power calculation unit 5 calculates the power of the input voice 10 and outputs it as the input power 15 to the recognition result determination unit 4. The pattern matching unit 3 reads the standard pattern 12 from the standard pattern 2, performs pattern matching with the input pattern 11 for each frame, calculates the inter-pattern distance, and sets it as the evaluation distance 13.
It outputs to the recognition result determination unit 4.

The pattern matching is performed after shifting the end point of the input pattern for each frame by one frame behind the end point of the previous frame. The pattern matching method is not particularly limited, but in this embodiment, an example using the start end free DP matching will be described.

FIG. 2 is a diagram showing a DP path permissible area for pattern matching when the start end free DP matching is used, and the DP path shown in FIG. 3 is used. 2 and 3, the horizontal axis is the frame of the input pattern and the vertical axis is the frame of the standard pattern.
Indicates an input pattern frame, and j indicates a standard pattern frame.

The start-end free DP matching method is a method for finding a DP matching path that gives the minimum inter-pattern distance to the end point without limiting the start end point. In the example shown in FIG. When I is the end point,
It indicates that one of the frames from IS1 to IS2 is the starting point. Further, shifting the end point of the input pattern for each frame to perform pattern matching means that the allowable area of the DP path shown in FIG. 2 has slid in the direction of the horizontal axis i.

Next, referring to FIG. 4 and FIG. 5, the recognition result determination unit 4
Will be described. Since the evaluation distance 13 is calculated for each input frame, it becomes time-series data for the input frame as shown in FIGS. 4 (a) and 4 (b). The horizontal axis of FIG. 4 is the input frame, and the vertical axis is the evaluation distance. FIG. 4 (a) is a graph of the evaluation distance for the correct standard pattern, and FIG. 4 (b) is a graph of the evaluation distance for the error standard pattern.

Since the power 15 is calculated for each input frame, it becomes a time series data for the input frame as shown in FIG. Fifth
The horizontal axis of the figure is the input frame, and the vertical axis is the power.

In the recognition result judging unit 4, the evaluation distance threshold DT as shown in FIG.
H and the evaluation distance duration threshold LDTH are set, and the evaluation distance is D
It is determined whether the time of the section that is continuously below TH is equal to or longer than the LDTH frame, the section that is equal to or greater than LDTH is the valley section, and the existence of the valley section is a condition related to the evaluation distance. Also, a power threshold PTH and a power duration threshold LPTH as shown in FIG.
The power condition is that PTH frames fall below the PTH frame, and when the two conditions are satisfied, the recognition determination is ended and the recognition result 14 is output.

The example shown in FIGS. 4 and 5 is a case where noise is added before the voice. In the case of the evaluation distance with the error standard pattern in FIG. 4B, matching with noise causes the evaluation distance to fall below DTH and its duration LD2 to be LDTH or more, which results in a valley section and spotting error occurs. The lowest point of the valley section is V2. If only the condition related to the evaluation distance is set as the end condition, the recognition determination is ended at this point and the recognition result 14 is output, resulting in erroneous recognition.

However, in the present invention, the recognition determination ends when the two conditions of the evaluation distance and the power are satisfied, so that the power of the section corresponding to the valley section whose lowest point is V2 is
Although it is less than PTH, the duration LP1 is less than or equal to LPTH, so the condition regarding power is not satisfied, the termination condition is not satisfied, and erroneous recognition is prevented.

Next, in the case of the evaluation distance with respect to the correct standard pattern of FIG. 4 (a), the evaluation distance is less than DTH and the duration LD1 thereof is LDTH or more, which is a valley section and is the lowest point V1. In addition, the power of the section corresponding to the valley section whose lowest point is V1 is
When Pen is below PTH and the duration is above LPTH Ien
In d, the condition regarding the power and the condition regarding the evaluation distance are satisfied, and the ending condition is satisfied.

On the other hand, as for the evaluation distance with the error standard pattern in FIG. 4 (b), the evaluation distance again falls below DTH, and its duration LD3 is LDTH.
Because of the above, a valley section is formed, and the lowest point is V3. In this embodiment, the evaluation distance at the lowest point of the valley section is used as the judgment score. However, what is used as the determination score is not particularly limited.

In this way, the judgment score with all the standard patterns is obtained, and the standard pattern with the smallest judgment score is selected and output as the recognition result 14. In the case of the example shown in FIGS. 4 (a) and 4 (b), since V1 and V3 are the judgment scores and V1 <V3, the correct answer standard pattern is selected and output as the recognition result 14.

As is clear from the examples shown in FIGS. 4 (a), 4 (b), and 5, when only the condition related to the evaluation distance is used as the end condition, when spotting error occurs, determination of recognition at an appropriate position is made. Can not be ended, which leads to misrecognition. on the other hand,
As shown in the above-described embodiment, by using the condition regarding power and the condition regarding evaluation distance as the ending condition, the recognition determination can be ended at an appropriate position even if a spotting error occurs, and the conventional method is used. It is possible to obtain the correct recognition result by removing the incorrect recognition.

EFFECTS OF THE INVENTION As described above, according to the present invention, the judgment of the recognition result is not completed only by the condition for the evaluation distance, but the condition for the evaluation distance according to the present invention and the condition for the power of the input voice are satisfied. By using the method of ending the determination of the recognition result and outputting the recognition result at the time of performing, there is an effect that an accurate recognition result can be obtained and erroneous recognition can be reduced.

[Brief description of drawings]

FIG. 1 is a block diagram of an embodiment of the present invention, FIG. 2 is a diagram showing a permissible area of a DP path of start-end free DP matching, FIG. 3 is a diagram showing a DP path of start-end free DP matching, and FIG. a) is a diagram showing the relationship between the correct standard pattern and the evaluation distance, FIG. 4 (b) is a diagram showing the relationship between the error standard pattern and the evaluation distance, and FIG. 5 is a diagram showing an example of a power graph of the input voice. Is. Description of symbols of main parts 1 ... Voice analysis unit 2 ... Standard pattern unit 3 ... Pattern matching unit 4 ... Recognition result determination unit 5 ... Power calculation unit

Claims (1)

[Claims] 1. A standard pattern portion for storing a characteristic pattern of a word or a phoneme to be recognized as a standard pattern, and an input pattern which is a characteristic pattern of the input speech by analyzing the input speech at a predetermined frame period. A voice analysis unit that outputs, a pattern matching unit that calculates an evaluation distance by matching a standard pattern and an input pattern, a recognition result determination unit that determines a recognition result based on the evaluation distance,
And a power calculation unit that calculates the power of the input voice,
The pattern matching unit shifts the end point of the input pattern by one frame for each frame of the input pattern to perform pattern matching, calculates the inter-pattern distance as an evaluation distance for each frame, and recognizes the evaluation distance as time-series data. Output to the determination unit, the recognition result determination unit is a voice recognition method for determining the recognition result by providing an evaluation distance threshold and an evaluation distance duration threshold for the evaluation distance that is time-series data. A power threshold and a power duration threshold are provided for the power of a certain input voice, and the evaluation distance of the input voice satisfies the condition defined by the evaluation distance threshold and the evaluation distance duration threshold, and the power of the input voice Ends the recognition result determination when the condition defined by the power threshold value and the power duration threshold value is satisfied, Speech recognition system, characterized in that the force.