JPH0119598B2 - - Google Patents

Description

[Detailed Description of the Invention] (a) Technical Field of the Invention The present invention relates to a speech recognition device, and particularly to a monosyllabic speech recognition device that recognizes input speech of a speaker whose monosyllabic speech has been registered in advance in the speech recognition device with a high recognition rate. Regarding recognition method.

(b) Background of the Technology As speech recognition technology has improved in recent years, there has been a desire for a speech recognition device that makes fewer errors when recognizing a speaker's voice. Speech recognition methods mainly convert the speaker's monosyllabic speech into feature parameters and store them in advance, and then match the feature parameters of the unknown input monosyllabic speech with the pre-stored feature parameters to select the most similar one. However, even if the same monosyllabic voice is pronounced, the characteristic parameters change depending on the way it is uttered, so even if the same monosyllabic voice is registered several times with different utterance methods, there will be no errors. It is difficult to do so. In particular, for monosyllabic speech that has characteristic parameters that are likely to cause recognition errors, it is impossible to improve the recognition rate unless the matching method is taken into consideration.

(c) Purpose of the Invention Based on the above-mentioned request, the purpose of the present invention is to compare and match the feature parameters of input monosyllabic speech with pre-registered feature parameters in a monosyllabic speech recognition method to find the most similar feature parameter. Select multiple candidates sequentially from , change the re-matching parameters for each combination of the multiple candidates, perform re-matching with the feature parameters of the unknown input monosyllabic speech, and select the most similar one as the corresponding monosyllabic speech. To provide a speech recognition device with a high speech recognition rate. Also, the method of re-verification is the same as the initial verification, which is DP.
This is done through verification, which simplifies the configuration of the device.

(d) Structure of the Invention The structure of the present invention is to register monosyllabic speech in advance, and to calculate the characteristic parameters of the unknown input monosyllabic speech and the characteristic parameters of all monosyllabic speech registered in advance.
After DP matching, multiple rematching candidates are selected from the registered monosyllabic speech in descending order of similarity, and the unknown input monosyllabic speech is matched with the unknown input monosyllabic speech using the rematching parameter determined by the combination of the multiple rematching candidates. As a result, the most similar rematching candidate is recognized as the corresponding monosyllabic speech and sent out. However, since the rematching parameter has characteristics in the frequency spectrum, It uses a weighted frequency spectrum of . That is, components in a certain frequency band are emphasized and other frequency components are reduced. Furthermore, some consonants have different levels when uttered, and the recognition rate can be improved by changing the threshold for distinguishing between speech and noise, so it is better to lower the threshold even if there is a risk of picking up noise. There are some that are good, and others that require raising the threshold to avoid picking up noise. The above conditions are used differently depending on the selected candidates and re-verified.

(e) Embodiment of the invention The figure is a block diagram of a circuit showing an embodiment of the invention.

First, the speaker connects the switching section 3 to the parameter storage section 4 under the control of the control section 12 in order to register monosyllabic speech in advance, and inputs the monosyllabic speech. The preprocessing section 1 performs audio level adjustment, analog-to-digital conversion, etc., and sends it to the parameter extraction section 2, which extracts characteristic parameters and stores them in the parameter storage section 4. Next, in order to perform voice recognition, the switching unit 3
is connected to the storage unit 5 and uttered. All the feature parameters stored in the parameter storage section 4 are read out under the control of the control section 12 and are compared with the feature parameters from the storage section 5 in the comparison section 8 . The control unit 12 sequentially selects a plurality of monosyllabic speech candidates corresponding to the feature parameters read from the parameter storage unit 4 that are most similar to the feature parameters from the storage unit 5 as the first candidate, and The feature parameters corresponding to the candidates are stored in the parameter storage unit 4.
The first candidate is then input to the multiplier 9. At the same time, the feature parameters in the storage unit 5 are input to the multiplier 10,
From the frequency weight storage unit 6, the frequency weight determined by the combination of the plurality of candidates, that is, the frequency spectrum is weighted to emphasize the components of a certain frequency band, and the components of other frequency bands are reduced, is applied to the multiplier 9, Send to 10. Furthermore, if the threshold value needs to be changed due to a combination of the plurality of candidates, the optimal threshold value is sent from the threshold storage section 7 to the re-verification section 11.
The re-verification unit 11 performs re-verification using the outputs of the multipliers 9 and 10 and the threshold value from the threshold value storage unit 7. A plurality of candidates are re-verified in order from the first candidate, and the monosyllabic speech corresponding to the most similar feature parameters is output as a recognition result via the control unit 12. The above operation is repeated each time the speaker utters a monosyllabic voice, and the re-matching candidates are selected each time.Even if the same monosyllabic voice is used, the re-matching candidates do not necessarily have the same combination, so the re-matching parameters are also selected for each combination of re-verification candidates.

In this example, when discriminating between monosyllabic sounds such as sa where the consonant starts with /s/ and monosyllabic sounds such as ta where the consonant starts with /t/, how to reduce the discrimination error. , emphasizes components in the range of approximately 250Hz to 600Hz and 3400Hz to 4500Hz from frequency spectrum information of the monosyllabic speech, and reduces other frequency components,
In addition, excellent effects were obtained by lowering the threshold of the voice start detection level. In addition, monosyllabic speech whose consonants begin with /b/ and /p/, such as ba and pa, and monosyllabic speech whose consonants begin with /z/ and /s/, such as za and sa, have components in the range of approximately 250Hz to 1300Hz. By emphasizing
For monosyllabic speech starting with /d/ and /z/, such as da and za, the same excellent effect as above was obtained by increasing the threshold of the speech start detection level and removing the noise at the speech start.

(f) Effects of the Invention As explained above, the present invention extracts and stores the frequency spectrum features of the characteristic parameters unique to each monosyllabic voice, and extracts and stores the frequency spectrum characteristics of the characteristic parameters of the monosyllabic voice of the speaker. Re-verification can greatly reduce recognition errors, which has a great effect.
Furthermore, since the verification method is DP verification by the control section, the verification section and re-verification section are the same, and the configuration of the device is easy.

[Brief explanation of drawings]

The figure is a block diagram of a circuit showing one embodiment of the present invention. 1 is a preprocessing section, 2 is a parameter extraction section, 3 is a switching section, 4 is a parameter storage section, 5 is a storage section, 6 is a frequency weight storage section, 7 is a threshold storage section, 8 is a collation section, 9 and 10 are Multiplier, 11 is re-verification unit, 12
is the control section.

Claims (1)

[Claims] 1. In a speech recognition device that recognizes monosyllabic speech by comparing feature parameters of monosyllabic speech registered in advance with feature parameters of unknown input monosyllabic speech, all registered monosyllabic speech After matching the syllable speech with the unknown input monosyllabic speech, a plurality of re-matching candidates are selected from the registered monosyllabic sounds based on the matching results, and a re-matching parameter is selected for each combination of the re-matching candidates. A monosyllabic speech recognition method characterized by providing a re-verification means to recognize monosyllabic speech. 2. Re-matching at least one of the feature parameters obtained by multiplying the frequency spectrum of the monosyllabic speech by the frequency weight and the feature parameters obtained by varying the threshold of the speech detection level for determining the matching interval. The monosyllabic speech recognition method according to claim 1, wherein the monosyllabic speech recognition method is selected as a parameter.