Voice recognizing device

Description: TECHNICAL FIELD [Detailed description of the invention]

【0001】

[Technical field to which the invention belongs]

The present invention relates to a voice recognition device capable of suppressing a decrease in recognition rate under noise or the like.

【0002】

[Conventional technology]

In the detection of a voice section (speech part during sound input) in a voice recognition device, consonants and vowels at the beginning of a word are often not detected, especially in a noisy state. As a result, the recognition rate drops. As a conventional speech recognition device that solves such a problem, there is one described in Japanese Patent Application Laid-Open No. 62-217298. In the conventional voice recognition device, a means for extracting a feature amount of voice input from a microphone, a means for detecting a voice section, a means for separating a sound section and a silent section in one word, and a silent section thereof. In a speech recognition device provided with a means for registering an addition dictionary in block units based on the above and a means for performing recognition processing based on a silent section, a word having a consonant block or the like that is relatively easy to be missing when creating a dictionary is previously added. It is characterized in that, at the time of recognition, the similarity between the case where the block is missing and the case where the block is not is calculated for the input pattern, and the higher one is used as the similarity of the word. ..

【0003】

[Problems to be Solved by the Invention]

However, in such a speech recognition device, when creating a dictionary, it is necessary to specify in advance a word having a consonant block or the like that is relatively easy to be missing. That is, there is a problem that it takes a lot of time and effort to create a word dictionary.

【0004】

The present invention has been made to solve the above-mentioned problems, and an object of the present invention is to provide a voice recognition device capable of suppressing a decrease in recognition rate at the time of noise with a simple configuration.

【0005】

[Means for solving problems]

In order to achieve the above object, the present invention is configured to be described in the claims. That is, in the invention according to claim 1, the voice section detecting means for detecting the voice section of the input signal wave, the voice dictionary in which the word to be recognized is recorded, the voice section of the input signal wave, and the above. In a voice recognition device that calculates the degree of matching with words in a voice dictionary and outputs the word with the highest degree of matching as a recognition result, half of the beginning of the word to be recognized (word stored in the voice dictionary). A word omitting a syllable is used as a word to be recognized, and the degree of coincidence with the voice section of the input signal wave is calculated.

【0006】

As described above, in the invention according to claim 1, the word to be recognized is configured to calculate the degree of coincidence by using a word without a half-syllable at the beginning of the word to be recognized as a word to be recognized. Therefore, it is possible to suppress a decrease in the recognition rate during noise without using a complicated word dictionary.

【0007】

When the half syllable at the beginning of the word is likely to be missing, as described in claim 2, when there is an unvoiced consonant at the beginning of the word, / p / or / t / at the beginning of the word as described in claim 3. Alternatively, there may be a half syllable of / k / or / s / or / h / or / f / or / m / or / n /, and in such a case, the word in which the first half syllable of the word is omitted is used. As the word to be recognized, the degree of coincidence with the voice section of the input signal wave is calculated.

【0008】

Next, in the invention according to claim 4, when the single syllable at the beginning of the word is line A or line C, the word omitting the single syllable at the beginning of the word is used as the word to be recognized. It is configured to calculate the degree of coincidence with the audio section of the input signal wave. As described above, in the invention according to claim 4, instead of omitting the half syllable as in claim 1, the single syllable is omitted. Even in the case of words in which syllables are likely to be missing, it is possible to suppress a decrease in the recognition rate during noise.

【0009】

When the single syllable at the beginning of the word is likely to be missing, as described in claim 5, the single syllable at the beginning of the word is line I or line U, and the consonant that suppresses the strength of the speech at the beginning of the word is If it continues, as in claim 6, the first single syllable is line I or c, and if it is followed by a plosive, then as in claim 7, the first single syllable is line I, Or, if it is a line and is followed by a sokuon, as described in claim 8, the first single syllable is the line i or line u, and / p / or / t / or / k / or / s. / Or / h / or / c / or / g / or / j / or / z / may follow.

【0010】

Next, in the invention according to claim 9, only when the noise is equal to or higher than a predetermined value, the word without the beginning of the word is set as the word to be recognized, and the degree of coincidence with the voice section of the input signal wave is determined. It is configured to calculate. With the above configuration, the process of omitting the half syllable or single syllable is performed only when the first half syllable or single syllable is likely to be missing, and the above process is not performed in the case of low noise. Since the match is judged for the entire word with, the recognition rate can be improved.

【0011】

Further, in the invention according to claim 10, when the noise is equal to or more than the first predetermined value and less than the second predetermined value, the word in which the first half syllable of the word to be recognized is omitted is recognized. When the noise is equal to or higher than the second predetermined value, the word without the single syllable at the beginning of the word is used as the word to be recognized, and the degree of coincidence with the voice section of the input signal wave is calculated. ing. As described above, by distinguishing between omission of half syllables and omission of single syllables according to the magnitude of noise, it is possible to match the magnitude of noise with the syllables that are easily missing, and further recognize during noise. In addition to being able to suppress a decrease in the rate, in the case of normal low noise, the recognition rate can be improved because the matching judgment is performed for the entire word without performing the above processing.

【0012】

[Effect of the invention]

In the inventions according to claims 1 to 3, the word that omits the half-syllable that is relatively easy to be missing at the beginning of the word is set as the word to be recognized, and the degree of coincidence with the detected speech section is calculated. Therefore, it is possible to suppress a decrease in the recognition rate at the time of noise without using a complicated word dictionary.

【0013】

Further, in the inventions according to claims 4 to 8, by omitting the single syllable at the beginning of the word, the single syllable at the beginning of the word is likely to be missing even when the word is noisy. It is possible to suppress a decrease in the recognition rate in.

【0014】

Further, in the invention according to claim 9, the process of omitting the half syllable or the single syllable is performed only when the noise is equal to or more than a predetermined value and the half syllable or the single syllable at the beginning of the word is actually likely to be missing. In the case of low noise, the recognition rate can be improved because the matching judgment is performed for the entire word without performing the above processing.

【0015】

Further, in the invention according to claim 10, by distinguishing between omission of half syllables and omission of single syllables according to the magnitude of noise, it is possible to match the magnitude of noise with the syllables that are easily missing. Further, it is possible to suppress a decrease in the recognition rate at the time of noise, and in the case of low noise, the recognition rate can be improved because the matching judgment is performed for the entire word without performing the above processing. The effect is obtained.

【0016】

BEST MODE FOR CARRYING OUT THE INVENTION

Hereinafter, the present invention will be described with reference to the drawings. 1 to 4 are diagrams showing the first embodiment of the present invention, FIG. 1 is a block diagram showing a configuration, FIG. 2 is a flowchart showing the entire arithmetic processing, and FIG. A figure showing a part, FIG. 4 is a sub-flow chart of the degree of matching calculation.

【0017】

First, in FIG. 1, the voice recognition device performs recognition processing with a microphone 101 for inputting voice, a filter 102, and an A / D converter 103 for converting an input signal into a digital signal, and a signal processing for outputting a recognition result. It consists of a device 104. The signal processing device 104 includes a CPU 104a, a program, data, a word dictionary, and a memory 104b that stores a condition for determining whether to omit a half syllable or a single syllable at the beginning of a word.

【0018】

Next, the operation will be described. First, in the flowchart shown in FIG. 2, the contents of each step are as follows. Step 201 ... The signal processor 104 stores the input signal in the memory 104b through the microphone 101, the filter 102, and the A / D converter 103. The band of the filter 102 is set to about 100 Hz to 7 kHz according to the frequency band of human voice.

【0019】

Step 202 ... The signal processing device 104 cuts out a voice section from the input signal. The cut out voice section is stored in the memory 104b. The audio section can be cut out by, for example, the following method. That is, the spoken voice is input from the microphone 101, and the amplitude (may be power), spectrum, etc. of the voice signal wave changed into an electric signal are detected to detect the voice. Specifically, the points where the amplitude level exceeds and falls below a certain threshold are set as the start point and the end point of the voice section, respectively, or the points where the spectrum suddenly changes in the vicinity of the above points are set as the start point and the end point. Cut out the audio section.

【0020】

Step 203 ... The signal processing device 104 calculates the degree of matching with the words in the speech dictionary for the cut-out speech section (details will be described later). The signal processing device 104 has a voice dictionary in the memory 104b, and the CPU 104a calculates the degree of coincidence between all the words in the voice dictionary and the voice section. FIG. 3 shows a part of the contents of the voice dictionary.

【0021】

Step 204 ... The signal processor outputs the word with the highest degree of matching.

【0022】

Hereinafter, the calculation content of the degree of coincidence in step 203 will be described using the subflow shown in FIG. Here, consider the case of calculating the degree of coincidence with the word "fujisawa".

【0023】

In FIG. 4, the processing contents of each step are as follows. Step 401 ... CPU104a determines whether to omit the first half syllable. In this case, when the memory 104b of the signal processing device 104 has / p / or / t / or / k / or / s / or / h / or / f / or / m / or / n / at the beginning of the word. I remember the condition that the half syllable is omitted. With reference to this condition, it is determined that "fujisawa" is a word that omits the beginning of the word.

【0024】

Step 402 ... Omit the first half syllable "f" of "fujisawa" and convert it to "ujisawa". Step 403 ... Calculate the degree of agreement between the above "ujisawa" and the detected voice section. For this calculation, for example, HMM (Hidden Markov Model) generally used in speech processing technology is used.

【0025】

As described above, in the first embodiment, the half syllable at the beginning of the word to be recognized is automatically determined, and the word without the half syllable is detected as the word to be recognized. Since the degree of coincidence with the syllable is calculated, it is possible to suppress a decrease in the recognition rate even when the consonant or vowel at the beginning of the word is not detected during noise without using a complicated word dictionary. Next, FIGS. 5 to 7 are diagrams showing a second embodiment of the present invention, FIG. 5 is a block diagram showing a configuration, FIG. 6 is a flowchart showing the entire arithmetic processing, and FIG. 7 is a degree of agreement. It is a sub-flow chart of the calculation. First, in the configuration of FIG. 5, the signal processing device 104 in the configuration of FIG. 1 is replaced with the signal processing device 504. The signal processing device 504 includes a CPU 504a and a memory 504b.

【0026】

Next, the action will be described. The flowchart of FIG. 6 is obtained by replacing step 203 of the flowchart shown in FIG. 2 with step 603. FIG. 7 is a subflow showing this step 603. Hereinafter, a case of calculating the degree of coincidence with the word "fujisawa" will be described based on FIG. 7.

【0027】

Step 701 ... CPU504a determines whether to omit the first single syllable. The memory 504b of the signal processing device 504 has a condition that the single syllable at the beginning of the word is omitted when the single syllable at the beginning of the word is the i-line or the c-line. With reference to this condition, it is determined that "fujisawa" is a word that omits the beginning of the word.

【0028】

Step 702 ... Omit the single syllable "fu" at the beginning of "fujisawa" and convert it to "jisawa". Step 703 ... Calculate the degree of coincidence between the above-mentioned "jisawa" and the detected voice section. For this calculation, for example, HMM is used as described above.

【0029】

As described above, in the second embodiment, the single syllable that is relatively easy to be missing at the beginning of the word to be recognized is automatically determined, and the word without the single syllable is matched as the word to be recognized. Since the degree is calculated, it is possible to suppress a decrease in the recognition rate even when the consonants and vowels at the beginning of the word are not detected during noise without using a complicated word dictionary.

【0030】

As a condition for omitting the above-mentioned single syllable, if the single syllable at the beginning of the word is line I or line U, followed by a consonant that suppresses the strength of the single syllable at the beginning of the word, the single syllable at the beginning of the word is used. If the syllable is line i or u, followed by a plosive, the first single syllable is line i or u, and / p / or / t / or / k / or / s / or / If a sokuon such as h / or / c / or / g / or / j / or / z / continues, a single syllable may be omitted.

【0031】

Next, FIGS. 8 to 10 are diagrams showing a third embodiment of the present invention, FIG. 8 is a block diagram showing a configuration, FIG. 9 is a flowchart showing the entire arithmetic processing, and FIG. 10 is a degree of agreement. It is a sub-flow chart of the calculation.

【0032】

First, in the configuration of FIG. 8, the signal processing device 504 of FIG. 5 is replaced with the signal processing device 804, and a microphone 805 for noise input, a filter 806, and an A / D converter 807 are added. The noise input microphone is placed in a position where it is difficult for voice to enter (for example, a position far from the speaker's mouth). The signal processing device 804 has a CPU 804a and a memory 804b.

【0033】

Next, the action will be described. The flowchart of FIG. 9 is obtained by replacing step 603 of the flowchart shown in FIG. 6 with step 903. FIG. 10 is a sub-flow chart showing the details of step 903. Hereinafter, a case where the degree of coincidence with the word fujisawa is calculated will be described with reference to FIG.

【0034】

The CPU804a determines in steps 1001 and 1002 whether or not to omit the single syllable at the beginning of the word. Step 1001 ... CPU804a captures the noise around the speaker through the microphone 805, filter 806, and A / D converter 807 and calculates its power. If the calculation result is equal to or greater than the predetermined value, the process proceeds to step 1002. If the calculation result is less than the predetermined value, go directly to 1004.

【0035】

Step 1002 ... In the memory 804b of the signal processor 804, the first single syllable is line I or line C, and / p / or / t / or / k / or / s / or / h /. Alternatively, when the sokuon of / c / or / g / or / j / or / z / continues, there is a condition that the single syllable at the beginning of the word is omitted. With reference to this condition, it is determined that "fujisawa" is a word that omits the beginning of the word.

【0036】

Step 1003 ... Omit the "fu" at the beginning of " fujisawa" and convert it to "ji sawa". Step 1004 ... Calculate the degree of coincidence between the above "jisawa" and the detected voice section. HMM is used for this calculation as described above.

【0037】

As described above, in the third embodiment, only when the ambient noise level is equal to or higher than a predetermined value and the single syllable at the beginning of the word is likely to be missing, the beginning of the word to be recognized is relatively missing. Easy single syllables are automatically identified, and the word that omits that single syllable is used as the word to be recognized and the degree of matching is calculated. Therefore, consonants and vowels at the beginning of a word can be detected during noise without using a complicated word dictionary. Even if it is not done, the decrease in the recognition rate can be suppressed. In addition, only when the single syllable at the beginning of a word is likely to be missing, the process of omitting the single syllable is performed, and in the case of normal low noise, the matching judgment is performed for the entire word without performing the above process. The rate can be improved.

【0038】

In the third embodiment, the case where the noise level determination configuration is added to the second embodiment is illustrated, but it is added to the first embodiment (the half syllable is omitted). Of course it is possible to do so.

【0039】

Next, FIGS. 11 to 13 are diagrams showing a fourth embodiment of the present invention, FIG. 11 is a block diagram showing a configuration, FIG. 12 is a flowchart showing the entire arithmetic processing, and FIG. 13 is a degree of agreement. It is a sub-flow chart of the calculation.

【0040】

First, in the configuration of FIG. 11, the signal processing device 804 of FIG. 8 is replaced with the signal processing device 1104. The signal processing device 1104 has CPUs 1104a and 1104b.

【0041】

Next, the action will be described. The flowchart shown in FIG. 12 is obtained by replacing step 903 of the flowchart shown in FIG. 9 with step 1203. The details of step 1203 are shown in the sub-flow chart of FIG. Hereinafter, a case of calculating a match with the word "fujisawa" will be described with reference to FIG.

【0042】

CPU1104a determines in steps 1301, 1302, and 1304 whether or not to omit the single syllable at the beginning of a word. Step 1301 ... The signal processor 1104 takes in ambient noise and calculates its power through the microphone 1105, filter 1106, and A / D converter 1107. The calculation result Pi is compared with two thresholds k1 and k2 in which k1 <k2. Then, if k1 Pi <k2, the process proceeds to step 1302, and if k2 Pi <k2, the process proceeds to step 1303. If Pi <k1, go directly to step 1306. The values of k1 and k2 are determined based on experiments.

【0043】

Step 1302 ... CPU1104a determines whether to omit the first half syllable. When the memory 1104b of the signal processor 1104 has / p / or / t / or / k / or / s / or / h / or / f / or / m / or / n / at the beginning of the word, half of it. It has the condition that syllables are omitted. With reference to this condition, it is determined that "fujisawa" is a word that omits the beginning of the word.

【0044】

Step 1303 ... Omit the first half syllable "f" of "fujisawa" and convert it to "ujisawa". Step 1304 ... In the memory 1104b of the signal processor 1104, the first single syllable is line I or line C, and / p / or / t / or / k / or / s / or / h /. Alternatively, it has a condition that the single syllable at the beginning of the word is omitted when a sokuon such as / c / or / g / or / j / or / z / continues. With reference to this condition, it is determined that "fujisawa" is a word that omits the beginning of the word.

【0045】

Step 1305 ... Omit the single syllable "fu" at the beginning of "fujisawa" and convert it to "jisawa". Step 1306 ... Calculate the degree of match. HMM is used for this calculation as described above.

【0046】

As described above, in the fourth embodiment, when the noise is equal to or more than the first predetermined value k1 and less than the second predetermined value k2, the word in which the first half syllable of the word to be recognized is omitted is recognized. The target word is set, and when the noise is equal to or higher than the second predetermined value, the word without the single syllable at the beginning of the word is set as the target word, and the degree of matching with the word in the speech dictionary is calculated. ..

【0047】

As described above, by distinguishing between omission of half syllables and omission of single syllables according to the magnitude of noise, it is possible to match the magnitude of noise with the syllables that are easily missing, and further recognize during noise. In addition to being able to suppress a decrease in the rate, in the case of normal low noise, the recognition rate can be improved because the matching judgment is performed for the entire word without performing the above processing.

[Simple explanation of drawings]

[Figure 1]

The block diagram which shows the structure of the 1st Embodiment of this invention.

[Figure 2]

The flowchart which shows the whole arithmetic processing in embodiment of FIG.

[Fig. 3]

The figure which shows a part of the contents of the speech dictionary in embodiment of FIG.

[Fig. 4]

A sub-flow chart showing the details of the matching degree calculation in the flowchart of FIG.

[Fig. 5]

The block diagram which shows the structure of the 2nd Embodiment of this invention.

[Fig. 6]

The flowchart which shows the whole arithmetic processing in embodiment of FIG.

[Fig. 7]

A sub-flow chart showing the details of the matching degree calculation in the flowchart of FIG.

[Fig. 8]

The block diagram which shows the structure of the 3rd Embodiment of this invention.

[Fig. 9]

The flowchart which shows the whole arithmetic processing in embodiment of FIG.

[Fig. 10]

A sub-flow chart showing the details of the matching degree calculation in the flowchart of FIG.

[Fig. 11]

The block diagram which shows the structure of the 4th Embodiment of this invention.

[Fig. 12]

The flowchart which shows the whole arithmetic processing in embodiment of FIG.

[Fig. 13]

A sub-flow chart showing the details of the matching degree calculation in the flowchart of FIG.

[Explanation of symbols]

101 ... Microphone 804 ... Signal processor 102 ... Filter 804a ... CPU 103 ... A / D converter 804b ... Memory 104 ... Signal processor 805 ... Microphone 104a ... CPU 806 ... filter 104b ... Memory 807 ... A / D converter 504 ... Signal processing device 1104 ... Signal processing device 504a ... CPU 1104a ... CPU 504b ... memory 1104b ... memory