JP2001034288A - Voice recognition system and method for portable telephone system - Google Patents

Abstract

PROBLEM TO BE SOLVED: To reduce erroneous recognition when a voice is inputted to a voice recognition device in which many similar recognition words are registered under a noisy environment. SOLUTION: The voice recognition system of a portable telephone system to which a voice is to be inputted is provided with a voice recongizing part 5 which preserves a dictionary in which plural recognition words and plural noise words indicating voices are registered and which recognizes an inputted voice and outputs recognized results of plural candidates and a recognition deciding 4A performing a deciding processing of whether the inputted voice is a recognition word or a noise word with the ratio of occurrence of the candidates of the recognition word and the ratio of occurrence of candidates of the noise word recognized in the voice recognizing part 5.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a portable telephone device.
In particular, the present invention relates to a speech recognition system and method for a mobile phone device.

[0002]

2. Description of the Related Art As a prior art, Japanese Patent Laid-Open No. 61-142
No. 99 is disclosed. This publication describes an example of a method of improving the recognition rate of speech recognition. In the speech recognition, when the similarity value of the first candidate is smaller than a predetermined first threshold value, the second candidate of the second candidate is increased. The difference between the similarity of the first candidate and the similarity of the first candidate is determined, and if this difference is smaller than the second threshold, it is determined to be out of the recognition target, thereby reducing erroneous recognition.

[0003]

However, when speech recognition is used in a noisy environment, the similarity value of the first candidate increases even though the recognition result of the first candidate is not correct. There is a case where the condition of being smaller than the threshold value of 1 may not be satisfied, and there is a problem that a determination process for reducing false recognition cannot be performed.

[0004] Therefore, in view of the above-mentioned problems, the present invention provides a portable telephone capable of reducing erroneous recognition when a speech recognition device in which many similar recognition words are registered is inputted under a noisy environment. It is an object to provide a speech recognition system and method for a telephone device.

[0005]

According to the present invention, in order to solve the above-mentioned problems, a plurality of recognition words indicating a voice and a plurality of noise words are registered in a voice recognition system of a portable telephone device for inputting voice. A speech recognition unit that holds a dictionary, recognizes the speech input, and outputs recognition results of a plurality of candidates, and an appearance ratio of recognition word candidates recognized by the speech recognition unit and an appearance ratio of noise word candidates. A speech recognition system for a mobile phone device, comprising: a recognition unit that performs a process of determining whether the word is a recognition word or a noise word.

[0006] According to this means, it is possible to reduce erroneous recognition when a voice is input in a noise environment to a voice recognition device in which many similar recognition words are registered. Preferably,
When the first candidate is a recognized word and the appearance ratio of noise words after the second candidate is equal to or greater than a determination value, the recognition determination unit determines that the recognition result is a noise word, and the noise word after the second candidate is a noise word. If the appearance ratio of is less than the determination value, the recognition result is determined to be a recognized word. Preferably, when the first candidate is a noise word and the appearance ratio of the recognized words after the second candidate is equal to or more than a determination value, the recognition determination unit determines the recognition result as a recognized word, and
If the appearance rate of recognized words after the candidate is less than the judgment value,
The recognition result is determined to be a noise word. Preferably, the appearance ratio of the noise word and the appearance ratio of the recognition word are respectively weighted for a plurality of noise words and a plurality of recognition words, and each weight is added each time a noise word and a recognition word appear as candidates. Is calculated.

[0007] By this means, not only the first candidate of the recognition result but also a plurality of recognition results from the second candidate to the k-th candidate are used to determine whether the input during speech recognition is speech or noise. , The recognition result is output, erroneous recognition can be prevented, and accurate voice recognition can be performed. Preferably, among the candidates for the recognition word, the candidates for the recognition word having a smaller number of characters than a predetermined number are set as the candidates for the noise word.

According to this means, when the recognition result candidate includes a recognition word, if the number of characters of the recognition word is small, there is a high possibility that the input at the time of voice recognition is noise, and if the number of characters is long, the input is speech. Can be determined to be high. In this way, when the recognition result candidate includes a recognition word in addition to the determination of whether the recognition result candidate is speech or noise, the number of characters is added to the determination value to output a more accurate voice recognition result. It becomes possible to do.

Further, according to the present invention, in a voice recognition method of a portable telephone device for performing voice input, a dictionary in which a plurality of recognition words indicating voice and a plurality of noise words are registered is stored, and the voice input is recognized. Outputting a recognition result of the plurality of candidates, and performing a process of determining whether the recognition word is a recognition word or a noise word based on the appearance ratio of the recognized recognition word candidate and the appearance ratio of the noise word candidate. A feature of the present invention is to provide a voice recognition method for a mobile phone device.

[0010] According to this means, similarly to the above-mentioned invention, it is possible to reduce erroneous recognition when a voice is input in a noise environment to a voice recognition device in which many similar recognition words are registered.

[0011]

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram showing a schematic configuration of a voice recognition system for a mobile phone device according to the present invention. As shown in FIG.
Is provided, and the radio unit 1 transmits and receives a radio signal to and from a base station (not shown).

Further, an operation unit 2 is provided in the portable telephone device, and the operation unit 2 operates the portable telephone, starts voice recognition, and the like. Further, a display unit 3 is provided in the mobile phone device,
The display unit 3 displays numbers, characters, and the like. The wireless unit 1, the operation unit 2, and the display unit 3 have a main CPU (Central Pr)
A processing unit (central processing unit) 4 is connected, and the main CPU 4 controls transmission and reception of the wireless unit 1, receives a control instruction from the operation unit 2, and performs display control on the display unit 3.

A speech recognition unit 5 is connected to the main CPU 4, and the speech recognition unit 5 includes a speech recognition LSI (Large S).
and a voice integrated circuit, which performs a speech recognition process. When the speech recognition start key is pressed by the operation unit 2, the main CPU 4 detects the depression of the speech recognition start key, determines the start of speech recognition, and transmits a speech recognition start command to the speech recognition unit 5. You.

The main CPU 4 and the voice recognition unit 5 have an A / D
(Analog to Digital; analog
A (digital) converter 6 and a speech synthesis unit 10 are connected, and the speech synthesis unit 10 synthesizes the recognition result from the speech recognition unit 5 into speech. D / A (Digital)
to Analog (digital / analog) converter 7 is connected.

A microphone 8 is connected to the A / D converter 6, and the microphone 8 inputs voice. A / D
The converter 6 converts an analog audio signal into digital audio data. The voice input to the microphone 8 is processed as a transmission sound of the mobile phone.

Further, the voice recognition unit 5 includes a main CPU
When a start command is received from 4, the recognition process is performed on the voice input from the microphone 8. D / A
A speaker 9 is connected to the converter 7, and the speaker 9 outputs sound. The D / A converter 7 converts digital voice data into a voice analog signal.

The sound output to the speaker 9 is processed as the sound received by the mobile phone. Furthermore, when the start sound sounds at the start of voice recognition and the recognition result is determined by the voice recognition unit 5, the voice corresponding to the recognition result is synthesized by the voice synthesis unit 10 and D / A
Output from the speaker 9 via the converter 7. In addition, the recognition result determined by the voice recognition unit 5 is displayed on the display unit 3.

Next, the speech recognition unit 5 recognizes
Calling the memory dial and functions of the mobile phone
The plurality of recognition words and the n number of noise words to be registered in order to prevent malfunction due to noise are registered in the dictionary. A noise word is a dictionary for voice recognition that is registered to prevent malfunctions of recognition words for memory dials and function calls when noise is input during voice recognition. Noise is registered.

According to this registration, even when sudden noise is input during voice recognition, when pattern matching with the voice recognition dictionary is performed, the input word of the noise word is better than the recognized word of the memory dial or function call. Is similar to.

For this reason, the recognition result is a noise word, and it is possible to prevent malfunction due to noise input.
Further, the main CPU 4 is provided with a recognition determining unit 4A, and the recognition determining unit 4A performs a process of determining a recognition word and a noise word, which are the recognition results of the speech recognition unit 5.

FIG. 2 is a flowchart for explaining the outline of the operation of the main CPU 4, the voice recognition unit 5, and the recognition determination unit 4A of FIG. In step S1, a voice recognition start key is pressed from the operation unit 2. In step S2, the main CPU 4 detects pressing of the voice recognition start key.
In step S3, the main CPU 4 determines activation of voice recognition.

In step S4, the main CPU 4 sends a voice recognition start command to the voice recognition unit 5. In step S5, the voice recognition unit 5 starts voice recognition. In step S6, the voice recognition unit 5 performs voice recognition on the voice input from the microphone 8.

In step S7, the voice recognition unit 5 determines the result of voice recognition. In step S8, the recognition results of the first to k-th candidates are output to the recognition determination unit 4A as the recognition results. In step S9, the recognition determination unit 4A adds the second candidate to the k-th candidate in addition to the first candidate of the recognition result.
Perform determination processing up to candidates.

In step S10, the main CPU 4
When the recognition result is determined to be a recognized word by the recognition determining unit 4A, maintenance corresponding to the recognized word is displayed on the display unit 3,
The corresponding sound is output from the speaker 9. Step S1
In 1, when the recognition determination unit 4A of the main CPU 4 determines that the recognition result is a noise word, it is determined that noise has been input during speech recognition and a malfunction has occurred. In step S12, a message notifying that a malfunction has occurred with respect to the determined noise is displayed on the display unit 3, and a corresponding sound is output from the speaker 9.

FIGS. 3 and 4 are flowcharts for explaining the detailed processing of determining the recognition result in the recognition determining section 4A of the main CPU 4 in FIG. As shown in FIG. 3, when the voice recognition start key is pressed from the operation unit 2 in step S21, the recognition determination unit 4A starts voice recognition, and the voice recognition unit 5 responds to the voice input from the microphone 8. Wait for the recognition process to be performed and the recognition result to be determined.

In step S22, when the recognition result is determined and the k candidates from the first candidate to the k-th candidate are output as the recognition result in the voice recognition, the recognition determining unit 4A recognizes the k candidates. The results are obtained and used to make a determination as follows. In step S23, first, it is determined whether the first candidate is a recognized word. In step S24, if the first candidate is a recognized word, it is determined whether the second candidate is a noise word. If it is not a noise word, the process proceeds to step S26.

In step S25, when the second candidate is a noise word, weight w1 [k] is added to the judgment value. In step S26, it is next determined whether or not the third candidate is a noise word. If not a noise word, step S
Proceed to 28. In step S27, when the third candidate is a noise word, weight w1 [k-1] is added to the determination value.

In step S28, a process of determining whether each candidate up to the k-th candidate is a noise word is performed. If it is not a noise word, the process proceeds to step S30. In step S29, when the k-th candidate is a noise word, weight w1 [1] is added to the determination value. In step S30, when the determination processing up to the k-th candidate is completed, the cumulative addition value of the determination values is compared with the threshold Th1.

In step S31, the judgment value is Th1
If it is larger than the first word, the speech input during speech recognition is determined to be noise even if the first candidate of the recognition result is a recognized word. In step S32, the recognition result for the noise input is displayed, and the process ends. In step S33, when the determination value is smaller than Th1 in step S30, it is determined that the input at the time of voice recognition is a voice.

In step S34, the recognition result for the first candidate recognition word is displayed, and the process ends. For this reason, even if erroneous recognition in which the first candidate is a recognition word occurs despite noise input during voice recognition in voice recognition, the second candidate is used by using the results of the second to k-th candidates. When the k-th candidate includes many noise words and the determination value is equal to or greater than Th1, it is determined that the input during voice recognition is not a voice but a noise input.

For this reason, it is possible to avoid erroneous recognition by displaying the recognition result for the noise input. As shown in FIG. 4, in step S35, when the first candidate is determined to be a noise word in step S23, a process of determining whether the second candidate is a recognized word is performed. If not, the process proceeds to step S37.

In step S36, when the k-th candidate is a recognized word, a weight w2 [k] is added to the determination value.
In step S37, it is next determined whether or not the third candidate is a recognized word. If not, the process proceeds to step S39. In step S38, when the third candidate is a recognized word, weight w2 [k-1] is added to the determination value.

In step S39, a process of determining whether each candidate up to the k-th candidate is a recognized word is performed. If not, the process proceeds to step S41. In step S40, when the k-th candidate is a recognition word, the determination value is weighted w2
[1] is added. In step S41, when the determination processing up to the k-th candidate is completed, the cumulative addition value of the determination values is compared with the threshold Th2.

In step S42, the judgment value is Th2
If it is larger than the first candidate of the recognition result, it is determined that the voice input during the voice recognition is not a noise but a voice even if the first candidate is a noise word. In step S43, the recognition result for the recognized word of the second candidate is displayed, and the process ends.

In step S44, step S41
When the determination value is smaller than Th2, the input at the time of speech recognition is determined to be noise. In step S45, the recognition result when the recognition result is a noise word is displayed, and the process ends.

FIG. 5 is a diagram for explaining a specific example. As shown in the configuration of the recognition dictionary in the voice recognition unit 5 in FIG.
Five noise words (n) are registered as the word (m),
It is assumed that up to the third candidate (k) is output as a recognition result.
Further, as shown in the recognition result example 1 of FIG. 3D, when noise is input during speech recognition, the first candidate is a recognition word Suzuki, and the second and third candidates are noise words. explain.

When the first candidate is a recognized word, the second candidate and the third
It is determined whether or not the candidate is a noise word, and a determination value is calculated. In Example 1 described above, since both the second candidate and the third candidate are noise words, the judgment values include the judgment weight w1 [2] (= 2) and the value w1 [1] (= The judgment value becomes 3 by adding 1).

Since the judgment value Th1 (= 3) or more, the first candidate of the recognition result is the recognition word Suzuki, but the input at the time of speech recognition is determined to be noise, and the recognition result is the recognition for the noise word. The result is displayed to prevent erroneous recognition. Next, as shown in the example (e) of the recognition result, when Suzuki is uttered during speech recognition, an example will be described in which the first candidate is a noise word, the second candidate is Suzuki, and the third candidate is Sato. I do.

When the first candidate is a noise word, a process of determining whether or not the second and third candidates are recognized words is performed, and a determination value is calculated. In the above example 2, since both the second candidate and the third candidate are recognized words, the judgment weight is set to the judgment weight w2 in FIG.
By adding [2] (= 2) and w2 [1] (= 1), the determination value becomes 3.

Since the determination value is a value equal to or greater than Th2 (= 3), the first candidate of the recognition result is a noise word, but the input at the time of voice recognition is determined to be voice, and the second candidate is Suzuki as the recognition result. Is output. As described above, since the process of determining whether the recognition word is a recognition word or a noise word is performed based on the appearance ratio of the noise word candidate and the recognition word candidate appearance ratio, it is possible to reduce erroneous recognition.

Although the speech recognition system of the portable telephone device according to the embodiment of the present invention has been described above, the present invention is not limited to this embodiment, and various changes can be made according to the gist of the invention. . Therefore, according to the present invention, it is possible to prevent erroneous recognition and output a more accurate voice recognition result. As a result, the recognition performance of speech recognition can be improved.

Not only the first candidate of the recognition result but also the second candidate
This is because, by using a plurality of recognition results up to the k-th candidate, a process of determining whether the input during speech recognition is speech or noise is performed, and the recognition result is output. Next, another embodiment of the present invention will be described.
In the present embodiment, in addition to the processing of determining whether the second to k-th candidates are noise or voice, the second candidate is used as a processing unit for determining whether the input during voice recognition is voice or noise. When the recognized word is included in the first to k-th candidates, whether the number of characters of the recognized word is long or short is added as a determination value.

For example, when the recognition result candidate includes a recognition word, if the number of characters of the recognition word is small, the input during speech recognition is likely to be noise, and if the number of characters is long, the input is likely to be speech. It can be determined to be high. In this way, when the recognition result candidate includes a recognition word in addition to the determination of whether the recognition result candidate is speech or noise, the number of characters is added to the determination value to output a more accurate voice recognition result. It becomes possible to do.

[0044]

As described above, according to the present invention,
By using not only the first candidate of the recognition result but also a plurality of recognition results from the second candidate to the k-th candidate, a process of determining whether the input during the voice recognition is a voice or noise is performed, and the recognition result is output. As a result, erroneous recognition can be prevented, and accurate voice recognition can be performed.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a schematic configuration of a voice recognition system for a mobile phone device according to the present invention.

FIG. 2 is a flowchart illustrating an outline of operations of a main CPU 4, a voice recognition unit 5, and a recognition determination unit 4A in FIG.

FIG. 3 is a flowchart illustrating a detailed process of determining a recognition result in a recognition determining unit 4A of a main CPU 4 in FIG. 1;

4 is a flowchart illustrating a detailed process of determining a recognition result in a recognition determining unit 4A of a main CPU 4 in FIG. 1;

FIG. 5 is a diagram illustrating a specific example.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 ... Radio | wireless part 2 ... Operation part 3 ... Display part 4 ... Main CPU 4A ... Recognition judgment part 5 ... Voice recognition part 6 ... A / D converter 7 ... D / A converter 8 ... Microphone 9 ... Speaker 10 ... Voice synthesis Department

Claims (6)

[Claims] 1. A speech recognition system for a mobile phone device for performing speech input, comprising: a dictionary in which a plurality of recognition words indicating speech and a plurality of noise words are registered; A speech recognition unit that outputs a recognition result, and the appearance ratio of recognition word candidates recognized by the speech recognition unit,
A speech recognition system for a mobile phone device, comprising: a recognition determination unit that performs a process of determining whether a recognition word is a recognition word or a noise word based on the appearance ratio of noise word candidates. 2. The method according to claim 1, wherein the first candidate is a recognition word and the appearance ratio of noise words after the second candidate is equal to or greater than a determination value, and the recognition result is determined as a noise word. The recognition result is determined as a recognized word when the appearance ratio of the noise word after the candidate is less than the determination value.
A speech recognition system for a mobile phone device according to claim 1. 3. When the first candidate is a noise word and the appearance ratio of the recognized words after the second candidate is equal to or greater than a determination value, the recognition determination unit determines the recognition result as a recognized word. The speech recognition system according to claim 1, wherein the recognition result is determined to be a noise word when the appearance ratio of the recognized word after the candidate is less than the determination value. 4. The appearance rate of the noise word and the appearance rate of the recognition word are respectively weighted to a plurality of noise words and a plurality of recognition words, and each time a noise word or a recognition word appears as a candidate, 4. The speech recognition system for a mobile phone device according to claim 2, wherein the value is calculated by adding 5. The mobile phone according to claim 1, wherein, among the candidates for the recognition word, the candidates for the recognition word having a smaller number of characters than a predetermined number are set as the candidates for the noise word. Equipment speech recognition system. 6. A voice recognition method for a mobile phone device for performing voice input, comprising: a dictionary in which a plurality of recognition words indicating voice and a plurality of noise words are registered; Outputting a recognition result; and performing a process of determining whether the recognition word is a recognition word or a noise word based on the appearance ratio of the recognized recognition word candidate and the appearance ratio of the noise word candidate. Voice recognition method for telephone equipment.