JPH0792671B2 - Voice recognizer - Google Patents

Description

Description: TECHNICAL FIELD The present invention relates to a voice recognition device that detects and recognizes a voice section using noise information.

(Prior Art) Conventionally, this type of voice recognition device measures noise caused by the ambient environment of the speaker or noise caused by input conditions in advance before the voice is uttered, and the result of the noise measurement is used as noise information for the noise information. Is used to detect and recognize the voice section. FIG. 5 is a block diagram showing the configuration of a conventional voice recognition device. In the figure, 1 is a voice analysis unit for analyzing input data, 2 is a switch for switching the input destination of the analysis data from the voice analysis unit 1, and 3 is noise information created from the analysis data input via the switch 2. A noise information creation unit 4, a noise information storage unit 4 for storing the noise information created by the noise information creation unit 3, and a switch 2
An analysis data storage unit for storing analysis data input via the, 6 sets a level threshold value using the noise information stored in the noise information storage unit 4, and sets the analysis data in the analysis data storage unit 5 and the level threshold value. A voice section detecting unit for detecting the voice section by comparing the two, a recognition processing section 7 for recognizing the input voice based on the analysis data and the voice section information from the voice section detecting section 6, and a controller 8 for controlling the entire apparatus. is there.

Next, the operation will be described. First, a noise measurement is performed before the voice is first uttered. The controller 8 switches the switch 2 to the noise information creation unit 3 side to activate the voice analysis unit 1 and the noise information creation unit 3. As a result, the input data is input to the voice analysis unit 1, where an analysis process for extracting the power information and frequency information of the input data is performed at regular time intervals (frame units). At, noise information is created and stored in the noise information storage unit 4. When the noise measurement is completed, the controller 8 stops the voice analysis unit 1 and the noise information creation unit 3, and then performs the voice section detection and recognition operation. In this case, the controller 8 switches the switch 2 to the voice section detection unit 6 side to activate the voice analysis unit 1 and the voice section detection unit 6. As a result, the input data is analyzed by the voice analysis unit 1 as in the noise measurement, and the analysis result is input to the analysis data storage unit 5 and the voice section detection unit 6 as voice analysis data. The voice section detection unit 6 sets a level threshold for detecting the input voice level by using the voice analysis result and the noise information stored in the noise information storage unit 4 in advance and using the noise information obtained by noise measurement, The first frame of the section larger than the level threshold is set as the start end and the last frame is set as the end, and the section from the start end to the end is determined as the voice section, and then the controller 8 activates the recognition processing unit 7. The recognition processing unit 7 performs recognition processing by generally known pattern matching based on the voice analysis data and the voice section information and noise information output from the voice section detection unit 5, and outputs the recognition result. FIG. 6 shows the relationship between noise measurement and voice section detection for input data.

(Problems to be Solved by the Invention) However, in the device having the above-described configuration, a change in ambient noise of a speaker when a voice is input through a telephone or a microphone having a weak directivity, and transmission to a voice recognition device after the input. When non-pulse noise uttered by noise mixed in during the measurement is accidentally picked up in the noise measurement, the noise information becomes peculiar and the noise state near the actual voice and the noise measurement were obtained. There is a problem in that the noise information is greatly different and an accurate voice section cannot be cut out, resulting in erroneous recognition.

SUMMARY OF THE INVENTION It is an object of the present invention to eliminate the above-mentioned problem of erroneous voice section detection due to peculiar noise information, and to provide a device that cuts out a more accurate voice section and has an excellent recognition rate.

(Means for Solving Problems) In order to solve the above problems, the present invention provides an analysis means for analyzing input data by voice, an analysis result storage means for storing the analysis result of the analysis means, and the analysis result. A noise information calculation unit that calculates noise information by measuring noise in a measurement section between a predetermined start frame and an arbitrary end frame of the analysis result stored in the storage unit, and an end frame in the noise information calculation unit. A stop frame setting means for changing the measurement section by increasing by a predetermined amount, a noise information storage means for storing noise information from the noise information calculation means, and a level threshold value using the noise information from the noise information storage means. Is set, and the voice section is detected by comparing the contents of the analysis result storage means with the level threshold at the timing when the setting is made. Detecting means, recognizing means for recognizing the voice based on the output signals of the analysis result storing means and detecting means, and the detecting means, until the detecting end detects the beginning of the voice section, the stop frame setting means is operated to end. A voice recognition device comprising: a control unit that increases a frame by a predetermined amount to update noise information, and detects a voice section based on the updated noise information.

Preferably, the control means controls each means so that the latest contents of the noise information storage means are noise information of the first noise measurement section at the next voice recognition.

(Operation) According to the present invention, since the voice recognition device is configured as described above, the technical means operates as follows. The voice input data including the non-voice (noise measurement) section is subjected to voice analysis by the analysis means and stored in the analysis result storage means. The control means instructs the noise information calculating means to calculate the noise information by instructing the noise measurement section of the first frame and the arbitrary end frame of the analysis data of the analysis result storage means. Next, the detection means sets a level threshold value using the noise information input via the noise information storage means, and at this timing (end frame + level threshold setting time), compares the level of the analysis data with the level threshold value. Then, the voice section from the beginning to the end of the input voice is detected. When the detection means does not detect the start end of the voice section, the control means instructs the stop frame setting means to update the noise information by increasing the end frame by a predetermined amount until the start end is detected. The detection means sets the level threshold value using the updated noise information, and similarly detects the voice section at this timing (update end frame + update level threshold setting time). The recognizing means recognizes the input voice based on the voice section information obtained in this way. Therefore, the voice section detection is performed using the latest noise information until the voice section is detected, so that the problems of the conventional technique can be solved.

(Embodiment) FIG. 1 is a block diagram showing a first embodiment of the present invention. In the figure, the same reference numerals as those in FIG. 5 denote the same components. 30 is the noise information creation unit 3 in FIG.
The noise information creation unit corresponding to the above is composed of a noise information calculation unit 31 and a stop frame setting unit 32. The noise information calculation unit 31 inputs the voice analysis data (including the non-voice section) stored in the analysis data storage unit 5 and creates noise information. The stop frame setting unit 32 is the controller 8
In accordance with this instruction, the end frame number of noise measurement is stored and given to the noise information calculation unit 31.

Next, an operation example will be described. First, before speech is input, noise measurement is performed as primary noise measurement as in the conventional case. All the voice analysis data analyzed by the voice analysis unit 1 in the conventional manner are sequentially stored once in the analysis data storage unit 5 from the predetermined start frame number, and the controller 8 sets the end frame number of the section in which the primary noise measurement is performed. Set in the stop frame setting unit 32. Next, the noise information creation unit 30 is activated, and the noise information calculation unit 31 stores the voice analysis data in the section pointed by the start frame number and the end frame number set in the stop frame setting unit 32 in the analysis data storage unit. 5, noise information is created and stored in the noise information storage unit 4 in the same manner as in the related art. This completes the primary noise measurement. Next, the controller 8 activates the voice section detection unit 6, and the voice section detection unit 6 detects the voice section in the same manner as the conventional one based on the noise information obtained by the primary noise measurement, and further stops when the start end of the voice section is not detected. A frame number obtained by adding a fixed number of frames to the end frame number preset in the frame setting unit 32 is reset in the stop frame setting unit 32. Next, the noise information creating unit 30 is activated to read the voice analysis data of the section pointed by the first frame and the reset end frame number from the analysis data storage unit 5 in the same manner as the primary noise measurement.
Noise information is created and stored in the noise information storage unit 4,
The voice section detection unit 6 detects the start of the voice section. By repeating the above operation until the start of the voice section is detected, the noise measurement section is expanded and the latest noise information is given to the voice section detection unit 6. As a result, the voice section detection unit 6 detects the voice section in the same manner as the conventional one based on the latest noise information stored in the noise information storage unit 4.

FIG. 2 shows the relationship between the noise measurement and the voice section detection for the input data. In the section, the noise information obtained in the section [1] and the noise information obtained in the section [2] are shown. , ... represents the timing of detecting the voice section using the noise information obtained in the section [n], and the section of the primary noise measurement corresponds to the first noise measurement section [1], nth noise measurement section [n]
Indicates that the beginning of the voice section has been detected.

Since the operation after the detection of the voice section is the same as the conventional one, its explanation is omitted. In the first embodiment, the case where the section for detecting the speech section of the primary noise measurement is continuous has been described, but even if the section is discontinuous, the speech analysis data is treated as if it is continuous. Noise measurement may be performed as in the case.

Next, a second embodiment will be described. FIG. 3 is a block diagram showing a second embodiment of the present invention. In addition to the components of the first embodiment, the average value of the voice analysis data of the section in which the noise information used for the previous recognition is measured is stored. The average noise storage unit 41 is provided. Next, an operation example will be described. First, in the first recognition operation for input data, similar to the first embodiment, the primary noise measurement is performed and the noise measurement is repeated until the start end of the sequential voice section is determined, and finally the voice section is detected. After the processing, the average value of the voice analysis data of the section where the noise information used for the voice section detection is measured is obtained and stored in the average noise storage unit 41. Further, regarding the recognition of the input data from the second time onward, the primary noise measurement is not performed, and the average value of the voice analysis data stored in the average noise storage unit 41 according to the instruction of the controller 8 is used as the voice section in the analysis data storage unit 5. It is stored in the analysis data storage area for primary noise measurement corresponding to each frame in the detection section, and is treated as if the primary noise measurement was completed. The following operation is similar to that of the first embodiment,
Noise measurement is repeated until the start end of the voice section is detected for the section indicated by the start frame number and the end frame number of the stop frame setting unit 32 given by the controller 8, and thereafter the same as in the first embodiment. Take action. The other operations are the same as those in the first embodiment. Further, after the recognition processing, the average value of the voice analysis data of the section in which the noise information used for the voice section detection and the recognition processing is finally measured is the average noise storage unit 41. To store. Similar to FIG. 2 in the first embodiment, FIG. 4 shows the relationship between the noise measurement and the voice section detection with respect to the input data. The section of the noise information obtained in the section [1] In the section, the noise information obtained in the section [2] is used, and in the section [...], the noise information obtained in the section [n] is used to represent the timing of detecting the voice section. The section for storing the average value of the voice analysis data corresponds to the first noise measurement section [1], and indicates that the beginning of the voice section is detected by the nth noise measurement section [n]. .

In the second embodiment, the average value of the voice analysis data of the section in which the noise information finally used for the voice section detection and recognition processing is measured is stored in the average noise storage unit 41 in the device and used for the next recognition. However, the average value of the voice analysis data may be output to the outside of the device and received from the outside in the next recognition.

As described above, according to the present embodiment, the noise information is repeatedly measured until the beginning of the voice section is detected, and the voice section is detected by the noise information that is extremely close to the voice. It is possible to cut out a more accurate voice segment even when there is a change in the above, and to provide a voice recognition device with an excellent recognition rate. Further, as in the second embodiment, the primary noise measurement is not performed in the recognition of the input data for the second time and thereafter, so that the continuous measurement does not take time for the noise measurement and the speaker does not need time. The vocalization can be done naturally.

(Effects of the Invention) As described above, according to the present invention, a voice segment can be accurately cut out, and thus a voice recognition device having an excellent confirmation rate can be provided.

[Brief description of drawings]

FIG. 1 is a block diagram showing a first embodiment of the speech recognition apparatus of the present invention, and FIG. 2 is a diagram explaining the operation of the first embodiment,
FIG. 3 is a block diagram showing a second embodiment of the voice recognition device of the present invention, and FIG. 4 is a diagram explaining the operation of the second embodiment,
FIG. 5 is a block diagram showing a conventional voice recognition device, and FIG. 6 is a diagram for explaining a conventional voice section detection operation. 1 ... Voice analysis unit, 4 ... Noise information storage unit, 5 ... Analysis data storage unit, 6 ... Voice section detection unit, 7 ... Recognition processing unit, 8 ... Controller, 30 ... Noise information creation unit ,
31: Noise information calculation unit, 32: Stop frame setting unit, 41: Average noise storage unit.

Claims (2)

[Claims] Claims: (a) analysis means for voice analysis of input data; (b) analysis result storage means for storing analysis results of the analysis means; (c) predetermined analysis results stored in the analysis result storage means. Information measuring means for calculating noise information by measuring noise in a measurement section between the first frame and an arbitrary end frame, (d) increasing the end frame in the noise information calculating means by a predetermined amount and performing the measurement Stop frame setting means for changing the section, (e) noise information storage means for storing the noise information from the noise information calculation means, (f) setting a level threshold value using the noise information from the noise information storage means, Detection means for detecting a voice section by comparing the content of the analysis result storage means with the level threshold value at the timing when the setting is made; Recognizing means for performing voice recognition based on the output signals of the result storing means and the detecting means, (h) until the detecting means detects the start end of the voice section, the stop frame setting means is operated to set the end frame by a predetermined amount. The noise information is updated in increments of
A voice recognition device comprising: a control unit for detecting a voice section based on the updated noise information. 2. The control means controls each means so that the latest contents of the noise information storage means are noise information of the first noise measurement section at the time of the next speech recognition. A speech recognition apparatus according to claim 1.