JPH0511796A - Voice recognizer - Google Patents

Abstract

(57) [Abstract] [Purpose] To realize a voice recognition device capable of performing stable voice recognition processing for various voice input units without being influenced by the unit of input voice. A voice recognition device identifies a voice input unit in a voice input unit identification unit 12 based on voice input unit identification information for an input voice, and selects dictionary information suitable for the voice input unit identification result. Then, the voice recognition process is performed using the voice recognition process.

Description

Detailed Description of the Invention

[0001]

[Industrial application] The present invention relates to a voice recognition device.

[0002]

2. Description of the Related Art A conventional voice recognition device will be described with reference to FIG.

In the conventional voice recognition device, the voice input by the acoustic analysis unit 1 is analyzed and the characteristics are output.
Based on the output from the acoustic analysis unit 1, the voice recognition unit 2 matches the voice recognition dictionary to recognize the input voice. The result recognized by the voice recognition unit 2 is output so that the operator can confirm it. The result of the feature extraction that analyzes the input voice is stored in the storage unit 3.

[0004]

In voice input, data input can be performed without key input operation, and there is no need to know the key arrangement position of the key input device, the key operation method, etc. This is a convenient input method. However, the voice input method is different from the key input method, and may be erroneously recognized even if the operator speaks accurately.

Speech recognition errors in the speech recognition apparatus increase as the grammatical unit of speech to be recognized becomes complicated, such as "syllable", "word", "bunsetsu", and "sentence". Further, the background noise level increases in the environment in which the recognition device is used.

When the voice recognition unit is "syllable" or "word", the recognition process is considered to be a simple pattern matching process. On the other hand, when the speech recognition unit is "bunsetsu" or "sentence", this cannot be considered as merely pattern matching.

When a voice is input to a voice recognition device in a grammatical unit such as "bunsetsu" or "sentence", there is a problem of a particle as one of the causes of a decrease in the voice recognition rate. Although the voice recognition rate for simple phonemes, syllables, and words is high, if the voice recognition rate is extremely reduced when input is accepted for sentences or syllables,
It seems that the reason is the handling of particles. Particles are less pronounced clearly and clearly than other parts of speech, and particles are not accented. For this reason, semantic and grammatical knowledge for particles, which is different from the voice recognition processing for voice input in word units, is effective.

Another cause of the decrease in the voice recognition rate with respect to sentence-based input is that input speech in sentence units is not processed using grammatical knowledge or sentence semantic knowledge. Can be mentioned.

As described above, the recognition units are “bunsetsu” and “sentence”.
Then a more complex recognition process is required. Also, as the background noise level rises, the false recognition rate also rises sharply. However, if the conditions for speech recognition are good and sufficient recognition can be obtained even with “bunsetsu” and “sentence” as recognition units, a speech recognition device using “bunsetsu” and “sentence” as recognition units It is much easier to use than the one using syllables and words as recognition units.

In the speech recognition apparatus using the prior art, the grammatical unit of the input speech is "syllable", "word",
Even if it changes to "bunsetsu", "sentence", etc., the voice recognition processing is performed only by the acoustic characteristics of the voice without using the unique information based on the voice input unit. Moreover, such a voice recognition unit is fixed in use of the voice recognition device, and cannot be changed according to the environment. Therefore, in the voice recognition device using the conventional technology, the voice recognition rate may greatly change depending on the voice input unit, or the voice recognition may be performed as an input in a unit different from the unit in which the operator inputs the voice. There was a case where a very strange voice recognition result was output.

[0011]

A speech recognition apparatus according to the present invention comprises an acoustic analysis section for outputting the characteristics of an input speech, and a speech input unit identification section for determining a grammatical unit of the input speech. A voice recognition unit that converts the output of the acoustic analysis unit into a code string based on the output of the voice input unit identification unit; an acoustic analysis storage unit that stores the output of the acoustic analysis unit; and an output of the voice recognition unit And a voice recognition storage unit for storing

[0012]

EXAMPLES The present invention will be described in detail below based on examples.

(Embodiment 1) FIG. 1 is a block diagram showing the principle of a voice recognition apparatus according to the present invention.

FIG. 2 is a block diagram of the voice recognition apparatus of the present invention.

As grammatical units of speech that can be spoken by humans, "syllable", "word", "bunsetsu", "sentence", etc. can be considered. In the present invention, FIG.
The unit which is identified in (1) means these grammatical units.

The input voice is sampled as a digital signal by the voice input unit 21 shown in FIG. 2, which includes a microphone, a high-frequency emphasis filter, and an AD converter, which are components of the acoustic analysis unit 11 shown in FIG. Further, in the feature extraction circuit 22 of FIG. 2 which is also a component of the acoustic analysis unit 11 of FIG. 1, the voice signal converted into a digital signal is frequency-converted, the feature parameters in the frequency domain are extracted, and the feature of the uttered word is extracted. It is represented as a parameter string. The characteristic parameter sequence of the input voice extracted by the characteristic extraction circuit 22 of FIG. 2 is stored in the characteristic parameter sequence storage circuit 23 of FIG.

The input voice converted into the characteristic parameter is
In the FIG. 2 matching circuit 25, the voice input unit is identified using the voice input unit identification information 24 of FIG. Figure 2
The voice input unit identification information 24 and the matching circuit 25 of FIG. 2 constitute the voice input unit identification unit 12 of FIG. The voice input unit determined by the matching circuit 25 of FIG. 2 is stored in the voice input unit storage circuit 26 of FIG. Then, regarding the next voice input, the voice input unit storage circuit 26 shown in FIG.
The voice recognition processing is performed on the basis of the above information, and the processing in the voice input unit identification unit 12 in FIG. 1 is not performed. When the result of the voice recognition processing based on the information of the voice input unit storage circuit 26 in FIG. 2 is unsuccessful, the voice input unit identification unit 12 in FIG.

The flow of this processing is similar to the case where a person makes a call in an environment with a high noise level. For example, if you say "I will arrive at Narita at 3 o'clock tomorrow" and the other party cannot hear you, say "Tomorrow", "3 o'clock", "To Narita", or "Arrived" in a phrase. .. Or, if you still can't hear, simply say "Tomorrow", "3
When the other person mistakenly hears "Narita" and "Haneda", he or she speaks "na", "ri", and "ta" in syllables.

The speech recognition apparatus of the present invention identifies these utterance units and performs speech recognition processing by an optimum method.

The voice recognition dictionary 28 shown in FIG. 2 is provided with information according to the result of the voice input unit identifying section 12 identifying the voice input unit. The voice recognition dictionary 28 shown in FIG. 2 includes the voice recognition information 32 shown in FIG.
3, the FIG. 2 semantic information 34 is provided. In the DP matching circuit 27 shown in FIG.
The result stored in FIG. 3 and the voice recognition dictionary 28 shown in FIG.
P matching is performed and voice recognition is performed as a code string. Figure 2DP
The matching circuit 27 constitutes the voice recognition unit 13 in FIG. The result of voice recognition in the DP matching 27 of FIG. 2 is stored in the voice recognition storage circuit 29 of FIG. The voice recognition storage circuit 29 shown in FIG. 2 constitutes the voice recognition storage unit 14 shown in FIG.

The result of the voice recognition processing is stored in the voice recognition storage circuit 29 shown in FIG.
It is output to the display unit 31 shown in FIG. 2 under the control of 0 and can be confirmed by the operator.

The voice input unit identification unit 12 shown in FIG. 1, which is a constituent element of the present invention, will be described in more detail.

As described above, the voice input unit identification section 12 of FIG. 1 comprises the voice input unit identification information 24 of FIG. 2 and the matching circuit 25 of FIG. The information used in the voice input unit identification information 24 in FIG. 2 is time length information of the input voice, phoneme number information included in the input voice, and the like.

If the length information of the input voice time is less than a certain threshold value by using the voice input unit identification information 24 shown in FIG. 2, the input is made by a "word" or a unit smaller than that. It is determined that If the length information of the input voice time is longer than a certain threshold value, it is determined that the input is in a unit larger than the "word".

Furthermore, if the input voice unit is difficult to identify by the input voice time length, the voice input unit can also be identified by the number of phonemes included in the input voice. The input unit is further limited by the number of phonemes so that it can be determined as "word" if the number of phonemes existing in the input speech is within a certain number range and "bunsetsu" if it is within another range. Can be identified. In this way, the unit of the input voice can be identified easily and accurately.

Based on the information of the input unit identified by the voice input unit identification unit 12 of FIG. 1, the voice recognition unit 13 of FIG.
In, the knowledge used for the voice recognition process is selected. That is, when the voice input unit identification unit 12 in FIG. 2 determines that the voice input unit is a “word”, the information regarding the “word” stored in the voice recognition dictionary 28 in FIG. FIG. 2 The grammatical information 3 shown in FIG.
3, the semantic information 34 of FIG. 2 is not used. Also, if the voice input unit is identified as "bunsetsu", information about "bunsetsu"
That is, FIG. 2 voice recognition information 32, FIG. 2 grammar information 33, FIG.
2D using the information about the “bunsetsu” of the semantic information 34.
A voice recognition process is performed in the P matching circuit 27.

The present invention will be further described based on (Example 1) of the present invention. Processing when inputting in units of words in the voice recognition device that is an embodiment of the present invention will be described. The operator inputs a voice into the voice recognition device in units of words "Hanagasha Ondo". The input voice is received from the voice input unit 21 of FIG. 2, the feature of the voice is extracted by the feature extraction circuit 22 of FIG. 2, and is stored in the feature parameter string storage circuit 23 of FIG. The voice input time length, which is one of the voice characteristics, can be easily known by measuring the time when the voice power obviously different from noise is observed by using the voice power information. The voice input time length of the voice input here is 0.8 seconds. The voice input time information identifies the voice input unit as a word or a unit smaller than the word.

Next, the number of phonemes present in the input voice, which is one of the voice input unit identification information, is checked. It turns out that seven phonemes exist. Also from this information, it is determined that the voice input unit is a word or a unit smaller than the word.

Therefore, based on the input unit information obtained from the voice input unit identifying section 12 of FIG. 1, the voice recognition dictionary 2 of FIG.
The information corresponding to the voice input unit is selected from 8, and the voice recognition processing is performed in the voice recognition unit 13 in FIG. here,
Since the result identified as the voice input unit is a “word”, the voice recognition information 3 of FIG. 2 in the voice recognition dictionary 28 of FIG.
The DP matching circuit 27 in FIG. 2 performs DP matching with the feature parameter sequence stored in the feature parameter sequence storage circuit 23 using the information about the two words, and the input speech is recognized as a code sequence. At this time, it is effective only when the voice input unit is a “bunsetsu” or “sentence” unit. The voice recognition process can proceed without using any information, which is very efficient.

The input voice is voice-recognized through the processing as described above and is output from the display unit 31 of FIG. 2 in a form that can be confirmed by the operator under the control of the display unit control circuit 30 of FIG.

When "Hanakasa Ondo" is also input to the conventional voice recognition device which does not identify the voice input unit, a non-sentence meaning "flower is on the left side" is not recognized at all as a voice recognition result. It may be output. It is considered that the cause of outputting the erroneous voice recognition result is that the voice recognition processing is advanced due to the use of an inappropriate dictionary or knowledge for the voice input unit.

In the first embodiment, as the voice input section,
A microphone, a high-frequency emphasis filter, and an AD converter, which are sampled as a digital signal, are used, but any other structure may be used as long as the input voice can be sampled quickly. Further, in the feature extraction circuit, the method of frequency-converting the voice signal converted into the digital signal, extracting the feature parameter in the frequency domain, and expressing it as the feature parameter string of the uttered word is used.
Any other method may be used as long as it can accurately extract the features. Although the method of using the voice input time information and the phoneme number information as the information for identifying the voice input unit is shown, any other method may be used as long as it is a method capable of quickly and accurately identifying the voice input unit. I don't care.
Also, as a means of notifying the operator of the voice recognition result,
Although the method of displaying the voice recognition result on the display unit is used in the first embodiment, any method other than this may be used as long as it can promptly notify the operator of the voice recognition result.

[0033]

As described above, the voice recognition apparatus of the present invention recognizes the input unit of the input voice, and uses only the correct information that is in the voice input unit, so that the voice recognition process can be performed very quickly. I can do it now. As a result, it is possible to input voice to the voice recognition device in different input units, and it is very flexible even in situations where the voice input unit changes frequently due to changes in the usage environment or voice input data. Now, it is possible to greatly expand the use of voice input devices.

Further, since the voice input in every unit can be quickly used, the time required for the operator using the voice recognition device to master the voice input operation method can be extremely reduced. Therefore, it has become easier for many people to create an environment in which the voice recognition device is used.

[Brief description of drawings]

FIG. 1 is a principle block diagram of a voice recognition device of the present invention.

FIG. 2 is a block diagram of an embodiment of the present invention.

FIG. 3 is a block diagram of a conventional voice recognition correction device.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 acoustic analysis unit 2 speech recognition unit 3 storage unit 11 acoustic analysis unit 12 speech input unit identification unit 13 speech recognition unit 14 speech recognition storage unit 21 speech input unit 22 feature extraction circuit 23 characteristic parameter sequence storage circuit 24 speech input unit identification information 25 Matching Circuit 26 Voice Input Unit Storage Circuit 27 DP Matching Circuit 28 Voice Recognition Dictionary 29 Voice Recognition Storage Circuit 30 Display Control Circuit 31 Display 32 Voice Recognition Information 33 Grammar Information 34 Semantic Information

Claims (1)

Claim: What is claimed is: 1. An acoustic analysis unit that outputs characteristics of an input voice, a voice input unit identification unit that determines a grammatical unit of the input voice, and the voice input unit identification. A speech recognition unit that converts the output of the acoustic analysis unit into a code string based on the output of the unit, an acoustic analysis storage unit that stores the output of the acoustic analysis unit, and a speech recognition storage that stores the output of the speech recognition unit. A voice recognition device comprising: