Speech recognition apparatus

Claim 2

The speech recognition apparatus according to claim 1, characterized in that said speech analyzing means (1) includes converting means for converting the speech signal to a digital signal, subjecting the digital signal to analysis processing including FFT (Fast Fourier Transform), filter analysis, LPC (Linear Predictive Coding) analysis and cepstrum analysis, and generating the speech feature parameter series.

Claim 3

The speech recognition apparatus according to claim 1, characterized in that said speech recognition means is connected to said dictionary means (21) being storing speech feature patterns corresponding to the keywords, and comprises matching means for matching the speech feature patterns of the dictionary means (21) with the speech feature vector.

Claim 4

The speech recognition apparatus according to claim 3, characterized in that said speech recognition means (2) comprises a speech recognition unit which employs a word spotting method in which the speech feature parameter series generated by said speech analyzing means (1) is matched with the speech feature patterns of said dictionary means (21) at every predetermined frame cycle.

Claim 5

The speech recognition apparatus according to claim 4, characterized in that said speech recognition means (2) matches the speech feature parameter series with the speech feature patterns of said dictionary means (21) at every frame cycle of 8 msec.

Claim 6

The speech recognition apparatus according to claim 1, characterized in that said speech recognition means (2) shifts successively an end point of the input speech feature parameter series by a predetermined time, and matches the speech feature vector obtained in each of start point candidate periods determined with respect to each position of the shifted end point, i.e. in each of periods between the shifted end point and different start points with the speech feature patterns of said dictionary means (21), in order to perform word recognition and generate a plurality of word candidates.

Claim 7

The speech recognition apparatus according to claim 1, characterized in that said syntactic analysis means (3) includes means (31, 32, 33) for receiving the recognition result from said speech recognition means in the form of temporally discrete word series, performing syntactic analysis by using the likelihood of each word, and finding the sentence candidate likelihood by means of evaluation function.

Claim 8

The speech recognition apparatus according to claim 7, characterized in that said syntactic analysis means (3) includes means (32) for outputting preferentially the sentence candidate having a highest likelihood.

Claim 9

The speech recognition apparatus according to claim 1, characterized in that said syntactic analysis means (3) includes means (32) for analyzing a temporal connection relationship of temporally discrete keywords and a syntactic and/or semantic connection relationship thereof, and generating a semantic expression of the analyzed series of the keyword.

Claim 10

The speech recognition apparatus according to claim 1, characterized in that said syntactic analysis means (3) includes grammar/dictionary memory means (4) which stores grammar and dictionary data relating to syntactic/semantic data and temporal relationship of an input sentence, and means (32) for performing syntactic and semantic analysis with use of the grammar/dictionary data of said memory means.

Claim 11

The speech recognition apparatus according to claim 1, characterized in that said syntactic analysis means (3) has sub-sentence candidate buffer means (6) for storing at least one sub-sentence candidate which is a candidate of a sub-sentence produced during analysis.

Claim 12

The speech recognition apparatus according to claim 11, characterized in that said syntactic analysis means (3) includes means (34) for outputting the sub-sentence candidate as a sentence candidate.

Claim 13

The speech recognition apparatus according to claim 10, characterized in that said syntactic analysis means (3) comprises means (31) for determining whether the word candidate received from said speech recognition means is capable of being a top word of a sentence, by referring to the syntactic data stored in said grammar/dictionary memory means (4), means (34) for generating a semantic expression of the sub-sentence candidate as a sub-sentence candidate when the word candidate is capable of being the top word of the sentence, and sub-sentence candidate buffer means (6) for storing the sub-sentence candidate and its semantic expression.

Claim 14

The speech recognition apparatus according to claim 1, characterized in that said speech recognition means is connected to said syntactic analysis means (3) and includes means for outputting word candidates in the form of a word lattice and converting the same to temporally discrete keyword lattice expressions.

Claim 15

The speech recognition apparatus according to claim 14, characterized in that said syntactic analysis means (3) includes means for analyzing a temporary connection relationship of the temporal discrete keywords of the keyword lattice and a syntactic connection relationship thereof, to generate a semantic expression of the analyzed keyword lattice.

Claim 16

The speech recognition apparatus according to claim 11, characterized by further comprising means (135) for predicting a possible word candidate on the basis of syntactic limitations and the sub-sentence candidate under analysis, and using the predicted word candidate for the next input speech recognition processing.

Claim 17

The speech recognition apparatus according to claim 16, characterized in that said predicting means (135) includes means for detecting the word candidates which are connectable to each other from the sub-sentence candidate.

Claim 18

A speech recognition apparatus characterized by comprising: speech analyzing means (1) for receiving an external speech signal, analyzing the speech signal, and generating a speech feature parameter series; dictionary means (21) being storing a plurality of predetermined words; speech recognition means including means for converting the speech feature parameter series input from said speech analyzing means (1) into a speech feature vector, and means for matching the speech feature vector with the reference vectors of predetermined words in said dictionary means (21), and outputting a series of a plurality of word candidates used as keywords; and syntactic analysis means (3) for analyzing the series of the word candidates as the keywords according to syntactic limitations, said syntactic analysis means (3) including means for analyzing a temporal connection relationship of temporally discrete keywords and a syntactic connection relationship thereof, and generating a semantic expression of the analyzed series of the keywords.

Claim 19

A speech recognition apparatus characterized by comprising: speech analyzing means (1) for receiving an external speech signal, analyzing the speech signal and generating a speech feature parameter series; dictionary memory means (21) for storing a plurality of predetermined words; speech recognition means (2) including means for converting the speech feature parameter series input from said speech analyzing means (1) into a speech feature vector, matching the speech feature vector with the reference vectors of predetermined words in said dictionary memory means, and outputting a series of word candidates; means (31) for detecting whether or not a word candidate capable of being a top word of a sentence and a word candidate following thereto are connectable to each other in time and in grammar, and outputting a combination of the connectable word candidates as a sub-sentence; and means (34) for determining whether the sub-sentence candidate can be established as a sentence, and generating a sentence candidate.

Claim 20

A speech recognition apparatus according to claim 19, characterized in that said means for determining includes means (34) for establishing a sentence when the word candidate corresponding to the end of the sentence is detected.

Claim 21

A speech recognition apparatus characterized by comprising: speech recognition means (2) for matching a pattern of an input speech signal with a plurality of prepared dictionary words and outputting a plurality of temporally discrete words as a plurality of keyword candidates; and syntactic and/or semantic analysis means (3) for checking whether or not the keyword candidates are connectable to each other on the basis of a syntactic and/or semantic rule and outputting a combination of connectable keyword candidates as a sentence candidate.