JPH09114484A - Voice recognition device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To eliminate the ambiguity of modification relation in a syntax analysis by determining the modification relation of the syntax analysis as to each voice section divided with a voiceless section, etc., and recognizing a spoken voice. SOLUTION: A detection part 30 for a voiceless section, etc., detects the voiceless section, etc., including a section based upon a pause, a redundant word, rhythmical information, etc., according to a time series of feature parameters outputted from a buffer memory 3, and outputs its detection signal to an LR purser 5. The LR purser 5 reads in data in the speech section of a section unit indicated with the inputted detection signal and performs a section-limited HMM-LR process using an HMM-LR method for the speech section. Consequently, the modification relation in the speech section of each section unit is determined, and then modification relation between words, phrases, or clauses in the speech sections of different section units is determined.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a speech recognition apparatus, and more particularly to a speech recognition apparatus which detects a pause (silent section) in a uttered voice or a silent section such as a redundant word to continuously perform speech recognition. . In this specification,
Those that include pauses, redundant words, and breaks based on prosodic information are called silent intervals.

[0002]

2. Description of the Related Art In recent years, continuous speech recognition has been actively researched, and a sentence speech recognition system has been constructed by several research institutions. Many of these systems use carefully spoken speech as input targets. However, in human-to-human communication, redundant words such as "Ah" and "Eto", and stagnation, rewording, and rewording, which are poses such as silent intervals where there is no vocalization, are possible. Appears frequently.

FIG. 9 is a view showing a speech recognition operation in a stack form when the example sentence "I saw a girl with beautiful black hair" shown in FIG. 2 is subjected to speech recognition processing by a conventional continuous speech recognition apparatus. . The speech recognition operation of the conventional continuous speech recognition apparatus will be described with reference to FIG. First, as shown in the state stack 201 of FIG. 9, a sequence of uttered voices “clean” is recognized and stacked as a character. Next, since the character "beautiful" in the state stack 201 is included in the voice recognition dictionary, it is converted into the character "adj" representing an adjective phrase as shown in the state stack 202. Next, a sequence of utterances of "black" is recognized and stacked as characters as shown in the state stack 203. Since "black" in the state stack 203 is in the voice recognition dictionary, as shown in the state stack 204. It is converted to the character "adj" which represents an adjective phrase.

Next, a sequence of vocalized voice "hair no" is recognized and further stacked as characters as shown in the state stack 205, and the character "hair no" in the state stack 205 is listed in the voice recognition dictionary. So state stack 2
As shown in 06, it is converted into the character "NP" representing a noun phrase. Further, in the state stack 206, the adjective “adj” and the noun phrase “NP” are the noun phrase “N”.
The adjective phrase “adj” in which “black” has been converted and the noun phrase “NP” in which “hair” have been converted by applying the syntactic rule of becoming “P” are as shown in the state stack 207. Converted to the noun phrase "NP". That is, the noun phrase “NP” in the state stack 207 represents “black hair”.

In the state stack 207, whether or not the syntax rule that the adjective "adj" and the noun phrase "NP" become the noun phrase "NP" is applied or not.
There are two options. Here, when the syntax rule is applied, "clean" is related to "hair", and when the syntax rule is not applied, "clean" is the syntax structure not related to "hair" and the dependency relationship is determined. It will be carried over to the subsequent voice recognition processing. Therefore, in such a case, in the continuous speech recognition apparatus of the conventional example, the apparatus for accumulating characters is separated into two and the subsequent speech recognition is executed. That is, one device executes the following speech recognition processing in the state shown in the state stack 208 in which the syntax rules are applied to the state stack 207,
The other device executes the following voice recognition processing while the state stack 207 remains as it is. Here, the noun phrase "NP" in the state stack 208 of one device represents "clean black hair."

In one device, the state stack 209
As shown in, the sequence of the vocalized voice "girl" is recognized and piled up as a character on the noun phrase "NP" that displays "clean black hair", and the character "girl" is added in the state stack 209. Is listed in the voice recognition dictionary, so as shown in the state stack 210, "N
Is converted to the character "P". Then state stack 210
, The noun phrase “NP” and the noun phrase “NP” become the noun phrase “NP” by applying the syntactic rule to convert the “clean black hair” of the state stack 210 into the converted noun phrase “ The noun phrase “NP” obtained by converting “NP” and “girl” is converted into the noun phrase “NP” as shown in the state stack 211. Here, the noun phrase “NP” in the state stack 211 represents “a girl with beautiful black hair”.
Then, the sequence of the vocalized speech "saw" is recognized and stacked as characters as shown in the state stack 212, and "saw" in the state stack 212 is shown in the state stack 213 because it is in the voice recognition dictionary. Is converted into “VP” representing a verb phrase, and is recognized as one sentence as shown in the state stack 214. That is, a recognition result of a structure in which "clean" is related to "hair" is obtained.

[0007] In the other device, the sequence of the vocalized voice "girl" is recognized and stacked as a character as shown in the state stack 221, and the character "girl" is converted into a noun phrase as shown in the state stack 222. It is converted to the character "NP". Next, in the state stack 222,
By applying the syntax rules, the noun phrase "NP" in which "black hair" is converted and the noun phrase "N in which" girl "is converted
“P” is the noun phrase “N” as shown in the state stack 223.
P ". Here, the noun phrase “NP” in the state stack 223 represents “black hair girl”. Then, the syntax rules are further applied, and the adjective phrase “adj” converted from “pretty” and the noun phrase “NP” converted from “black hair girl” are converted into the nouns as shown in the state stack 224. Converted to the phrase "NP". That is, “pretty” is recognized as a structure related to “girl”. Next, the sequence of the vocalized speech "saw" is recognized and stacked as characters as shown in the state stack 225, and "saw" in the state stack 225 represents a verb phrase "VP" as shown in the state stack 226. , And is recognized as one sentence as shown in the state stack 227. That is, the recognition result of the structure in which "pretty" relates to "girl" is obtained.

[0008]

As described in detail above,
When the example sentence of FIG. 2 is recognized by the conventional continuous speech recognition apparatus,
The results of recognition of two different structures are obtained: the result of recognizing the structure that "clean" is related to "hair" and the result of recognition of the structure related to "pretty" is "girl". There was a problem that the sex could not be resolved. In addition, as a result, there is a problem that ambiguity increases when dealing with longer utterances.

An object of the present invention is to solve the above problems and to provide a speech recognition apparatus capable of eliminating the ambiguity of dependency relations in syntactic analysis.

[0010]

According to a first aspect of the present invention, there is provided a voice recognition device comprising voice recognition means for voice-recognizing an input voice and outputting a voice recognition result. The speech recognition means includes a detection means for detecting at least one of a pause, a redundant word, and a boundary of a phrase or a clause based on the inputted uttered voice and outputting a detection signal. It is characterized in that the dependency relation in the syntactic analysis is determined to recognize the uttered voice.

A speech recognition apparatus according to a second aspect is the speech recognition apparatus according to the first aspect, wherein the speech recognition means uses at least one of the pause, a redundant word, and a boundary of a phrase or a clause. After voice recognition processing is performed on each voice section of the input uttered voice composed of a plurality of divided voice sections, the dependency relation between words, phrases, or clauses belonging to different voice sections is determined, and the input is performed as described above. It is characterized by performing voice recognition of the vocalized voice.

Further, the voice recognition device according to claim 3 is
The voice recognition device according to claim 1 or 2, wherein the detecting means has a first condition that the power of the uttered voice is equal to or less than a predetermined threshold value within a predetermined time range, and a zero cross of the uttered voice. Is detected by detecting that at least one of the second conditions, which is equal to or larger than a predetermined threshold value, is satisfied during a predetermined time.

Furthermore, the speech recognition apparatus according to claim 4 is the speech recognition apparatus according to claim 1 or 2, wherein the detection means is at least one of the pause, the redundant word, and the boundary of the phrase or clause. Are detected by determining whether or not they match the respective predetermined language models.

[0014]

Embodiments of the present invention will be described below with reference to the drawings. <First Embodiment> FIG. 1 is a block diagram of a continuous speech recognition apparatus 81 according to a first embodiment of the present invention. The continuous speech recognition device 81 according to the first embodiment has an SSS (Succ
essive State Splitting) -LR (le
ft-to-right right most derivation type, that is, rightmost derivation type unspecified speaker continuous speech recognition apparatus, which is a hidden Markov model (hereinafter, referred to as HM network) stored in a memory 11 (hereinafter, referred to as HM network). HMM) network is used to execute the phoneme matching process in the phoneme matching unit 4, and the resulting speech recognition score is sent to the phoneme context dependent LR parser (hereinafter referred to as LR parser) 5 and responds to it. Then, the LR parser 5 executes continuous speech recognition on one sentence of the inputted speech and sends the phoneme prediction data to the phoneme matching unit 4 to perform the speech recognition process. First
In particular, the embodiment of the present invention detects a silent section including a pause, a redundant word, and a section based on prosodic information as a clue based on the time series of the characteristic parameters output from the buffer memory 3, and detects the detection signal by LR. The LR parser 5 is provided with a silent section detection unit 30 for outputting to the parser 5, and the LR parser 5 reads the data of the speech section of the delimiter unit indicated by the detection signal input from the silent section detection unit 30, On the other hand, HMM-L with interval restriction using HMM-LR method
The speech recognition result data is output by executing the R process and executing the HMM-LR process without section restriction for one sentence of the vocalized voice that is input when the end of the last division unit is reached. To do.

Here, in the above SSS, with respect to the stochastic model expressing the temporal transition of the speech parameter by the stochastic transition between the stochastic stationary signal sources (states) assigned in the feature space of the phoneme, Then, the model refinement is sequentially executed by repeating the operation of dividing each state into the context direction or the time direction based on the likelihood maximization criterion.

In FIG. 1, the uttered voice of the speaker is input to the microphone 1 and converted into a voice signal, and then input to the feature extraction unit 2. After performing A / D conversion on the input audio signal, the feature extraction unit 2 performs, for example, LPC analysis, and obtains a 34-dimensional feature parameter including logarithmic power, 16th cepstrum coefficient, Δlog power, and 16th Δcepstrum coefficient. Is extracted. The time series of the extracted feature parameters is input to the phoneme matching unit 4 via the buffer memory 3.

HM network memory 1 connected to the phoneme collation unit 4.
The HM network in 1 is represented as a plurality of networks in which each state is a node, and each state has the following information. (A) State number (b) Acceptable context class (c) List of preceding and succeeding states (d) Parameters of output probability density distribution (e) Self transition probability and transition probability to succeeding state

In the first embodiment, since the HM network needs to specify which speaker each distribution is derived from, a predetermined speaker mixed HM network is created by conversion. Here, the output probability density function is a Gaussian mixture distribution having a 34-dimensional diagonal covariance matrix, and each distribution is learned using a specific speaker sample.

The phoneme matching unit 4 executes a phoneme matching process in response to a phoneme matching request from the LR parser 5. At this time, the LR parser 5 passes phoneme context information including a phoneme matching section, a phoneme to be matched, and phonemes before and after the phoneme. The phoneme matching unit 4 connects the states on the HM network capable of accepting such a context based on the received phoneme context information within the constraints of the preceding state list and the following state list, thereby forming one model. Is selected. Then, the likelihood for the data in the phoneme matching section is calculated using this model, and the value of the likelihood is returned to the LR parser 5 as a phoneme matching score. Since the model used at this time is equivalent to HMM,
The forward pass algorithm used in the normal HMM is used as it is for the calculation of the likelihood.

On the other hand, the silent section detecting unit 30 detects a silent section including pauses, redundant words, and prosodic information as a clue based on the time series of the characteristic parameters output from the buffer memory 3. Then, the detection signal is output to the LR parser 5. Here, the silent section detection unit 30 recognizes the redundant word as a redundant word by comparing and collating with the phoneme model of the redundant word stored in the internal memory in advance, while recognizing the pause as a silent section as described in 2 below. A pose is detected when one of the two conditions is satisfied. (First Detection Condition) When the time t0 when the power is below a predetermined threshold level is a value in the following range, for example. Preferably, 50 milliseconds ≤ t0 ≤ 3 seconds. More preferably 50
Ms ≦ t0 ≦ 500 ms. (Second detection condition) When the time t1 when the number of zero crosses at which the input audio signal crosses the zero potential is equal to or more than a predetermined threshold value is, for example, a value in the following range. Preferably, 50
Milliseconds ≦ t1 ≦ 3 seconds. More preferably, 50 milliseconds ≤ t
1 ≦ 500 milliseconds. Further, the delimiter based on prosodic information or the like is presumed to be a phrase or clause boundary when the intonation sharply rises or falls. With respect to this, the boundary or boundary is determined by detecting that the fundamental frequency among the input characteristic parameters has rapidly increased or decreased at a predetermined degree of inclination or higher.

Then, the LR parser 5 reads the data of the voice section in the delimiter unit indicated by the detection signal input from the silent section detection unit 30 and uses the HMM-LR method for the voice section. HMM-LR processing with section restriction is executed, and when the end of the last division unit is reached, no section restriction HMM is applied to one sentence of the uttered speech input.
-Output the voice recognition result data by executing the LR process. Here, the section-restricted HMM-LR processing is the H that is executed only within the voice section of one division unit.
This is a speech recognition process by the LR parser 5 using MM, and the section-unlimited HMM-LR processing does not limit sections and a speech of different delimiter units is applied to one sentence of the inputted uttered speech. An HMM that executes by applying the syntax rules in the LR table memory 13 to words, phrases, or clauses belonging to an interval
Is a voice recognition process by the LR parser 5 using. Here, the voice section is one section divided by a silent section or the like (shown in parentheses in FIG. 5) in one sentence of the vocalized voice input as shown in FIG. The term “delimiter unit” means one unit consisting of a voice section and a silent section after the voice section as shown in parentheses in FIG. Further, in the present specification, a silent section or the like means a pause and a section including a redundant word and prosodic information as a clue, and a pause unit is a section divided by a pose as shown in FIG. Refers to a unit.

A predetermined context-free grammar (CFG) in the context-free grammar database memory 20 is automatically converted in advance to create an LR table and stored in the LR table memory 13, as is known. For example, the LR parser 5 refers to the speaker model memory 12 including the phoneme duration model and the LR table, and processes the input phoneme prediction data from left to right without backtracking. If there is syntactic ambiguity, the stack is split and the analysis of all candidates is processed in parallel. LR parser 5 is LR
The next phoneme is predicted from the LR table in the table memory 13 and the phoneme prediction data is output to the phoneme collation unit 4.
In response to this, the phoneme collation unit 4 collates by referring to the information in the HM network memory 11 corresponding to the phoneme, returns the likelihood to the LR parser 5 as a speech recognition score, and sequentially connects the phonemes. By doing so, continuous voice recognition is performed.

In the continuous speech recognition apparatus 81 of the first embodiment configured as described above, the feature extraction unit 2, the phoneme matching unit 4, and the LR parser 5 are composed of, for example, a digital computer.

FIG. 6 shows L of the continuous speech recognition device 81 of FIG.
6 is a flowchart showing a voice recognition process executed in the R parser 5. The voice recognition process will be described below with reference to FIG.

As shown in FIG. 6, in step S1, initialization of the HMM work area and initialization of the LR parser 5 are executed. Specifically, the state stack 0 cell is set to 1
Create pieces. Here, the cell used in the continuous speech recognition device 81 is a data structure that holds information necessary for analyzing speech recognition of the conventional HMM-LR method, that is, an LR work area having a state stack, a speech recognition score, and a probability table. And an HMM work area consisting of.

Then, in step S2, the data of the voice section in the delimiter unit indicated by the detection signal input from the silent section detection unit 30 is read. Furthermore, in step S3, section-limited HMM-LR processing using the HMM-LR method is executed for the voice section in units of delimiters in which the voice data is read. In step S4,
It is determined whether or not the end of the last division unit among the plurality of division units has been reached. If the end of the last division unit has not been reached (NO in step S4), the process proceeds to step S2, and steps S2 and S3. The process of is repeated. On the other hand, in step S4, when the end of the last division unit is reached (Y in step S4)
ES) Proceed to step S5, and HMM-LR without section restriction
The process is executed and the voice recognition process is ended.

Next, the voice recognition operation of the continuous voice recognition apparatus 81 of the first embodiment of FIG. 1 will be described using the example sentence shown in FIG. FIG. 2 is an example sentence including ambiguity of dependency relations in sentence structure analysis, that is, syntactic analysis. When recognizing only the character string in the example sentence of FIG. 2 and trying to analyze, the first dependency relation shown by the arrow above the example sentence of FIG. 2 and the second dependency relation shown by the arrow below the example sentence are shown. At least two dependency ambiguities remain. That is, "clean"
It is unclear whether "is" the "clean girl" of the first dependency relationship relating to "girl" or "clean" is the "clean hair" of the second dependency relationship relating to "hair". The present inventors utilize one of the above-mentioned two dependency relationships by utilizing a pause which is a silent section.
I found that I could decide one day. That is, if there is a pause in a silent section between "clean" and "black" (indicated by "△" between "clean" and "black" in FIG. 2), "clean" Is a first dependency relationship for “girl”, and poses between “hair” and “girl” (in FIG. 2, “hair” and “girl”). If there is an intervening “Δ”), it can be determined that “pretty” is the second dependency relationship for “hair”. The present invention removes the ambiguity of the dependency relation in the syntactic analysis and executes the voice recognition process by using the rule between the pause and the dependency relation described above.

FIG. 3 is a diagram showing the speech recognition operation of the continuous speech recognition device 81 in the case of having the first dependency relationship in the example sentence of FIG. 2 in a stack form. The voice recognition operation in the case of having the first dependency relationship will be described below with reference to FIG. First, as shown in the state stack 51 of FIG. 3, the LR parser 5 recognizes a sequence of vocal sounds of “clean” and accumulates them as characters, and then immediately after the recognition process of “clean”, the pause is a silent section or the like. The detection signal is detected by the detection unit 30, and the detection signal is input from the detection unit 30 to the LR parser 5 and accumulated as “Δ” indicating a pose on the character “pretty”. Then state stack 5
Since the character "Beautiful" in 1 is included in the voice recognition dictionary, it is converted to the character "adj" representing an adjective phrase as shown in the state stack 52. Next, L
The R parser 5 recognizes a series of vocalized voices called "black" and stacks them as characters on "△" indicating a pose as shown in the state stack 53, and "black" in the state stack 53 is stored in the voice recognition dictionary. Since it is listed, it is converted into the character "adj" representing the adjective phrase as shown in the state stack 54. Here, in the state stack 54, between the adjective phrase “adj” in which “pretty” is converted and the adjective phrase “adj” in which “black” is converted, “Δ” indicating a pause is stacked. Syntax rules do not apply.

Next, the LR parser 5 recognizes the vocalized sequence "hairy", and as shown in the state stack 55, it is stacked as a character on the converted adjective phrase "adj". , The character "hair no" in the state stack 55 is in the voice recognition dictionary, so the state stack 5
As shown in 6, it is converted into the character "NP" representing a noun phrase. Further, in the state stack 56, the adjective phrase “adj” and the noun phrase “NP” are the noun phrase “NP”.
The adjective phrase “adj” in which “black” is transformed and the noun phrase “NP” in which “hair” is transformed are applied to the noun phrase Converted to "NP". That is, the state stack 57
The noun phrase "NP" in represents "black hair". Next, a sequence of vocalized voices "girl" is recognized and stacked as characters on the noun phrase "NP" representing "black hair" as shown in the state stack 58, and the state stack 58
Since the character "girl" in is included in the voice recognition dictionary, "N" represents the noun phrase as shown in the state stack 59.
Is converted to the character "P".

Next, in the state stack 59, the syntactic rule that the noun phrase “NP” and the noun phrase “NP” become the noun phrase “NP” is applied, and the “black hair” of the state stack 59 is changed. The converted noun phrase "NP" and "girl"
The converted noun phrase “NP” is converted to the noun phrase “NP” as shown in the state stack 60. Here, the noun phrase "NP" in the state stack 60 is "black hair girl"
Represents Then, the LR parser 5 recognizes the sequence of the vocalized voice "saw" and is stacked as a character on the noun phrase "NP" representing "black hair girl" as shown in the state stack 61, and the state stack Since "see" in 61 is included in the voice recognition dictionary, it is converted into "VP" representing a verb phrase as shown in the state stack 62.

Then, it is determined that the end of the last pose unit has been reached, and the adjectives "adj" and "black haired girl" representing "pretty" located before and after "△" indicating the pose are displayed. As shown in the state stack 63, the syntactic rule that the adjective phrase “adj” and the noun phrase “NP” become the noun phrase “NP” is applied to the noun phrase “NP” representing “”. Converted to the noun phrase "NP". Here, the noun phrase "NP" of the state stack 63 represents "a pretty dark-haired girl" having a structure in which "pretty" refers to "a girl". Further, as shown in the state stack 64, it is converted into “S” representing a sentence, and only the speech recognition result having a structure in which “pretty” relates to “girl” is output. As described above, the syntax rule between the adjective phrase “adj” that represents “pretty” and the noun phrase “NP” that represents “black hair girl” that is positioned before and after “△” that displays a pose Since the application is executed after reaching the end of the last pose unit, only the speech recognition result of the structure in which "pretty" relates to "girl" can be output. As described above, in the first embodiment, after the voice recognition of each voice section of one sentence of the input uttered voice including a plurality of voice sections is executed, the section of the input uttered voice is not limited without limiting the section. By applying the syntax rules in the LR table memory 13 between words, phrases or clauses belonging to different speech intervals for one sentence, a dependency relation between words, phrases or clauses belonging to different speech intervals is determined. ing.

FIG. 4 is a diagram showing in stack form the speech recognition operation of the continuous speech recognition apparatus 81 in the case of having the second dependency relationship in the example sentence of FIG. The voice recognition operation in the case of having the second dependency relationship will be described below with reference to FIG. First, as shown in the state stack 151 of FIG. 4, the LR parser 5 recognizes a sequence of vocal sounds "clean" and stacks them as characters. Next, since the character "beautiful" in the state stack 151 is included in the voice recognition dictionary, it is converted into the character "adj" representing an adjective phrase as shown in the state stack 152. Next, the LR parser 5 recognizes a sequence of vocalized voices of "black", and as shown in the state stack 153, "clean".
Stacked as a character on the adjective adjective
Since "black" in the state stack 153 is included in the voice recognition dictionary, it is converted to the character "adj" representing an adjective phrase as shown in the state stack 154.

Next, the LR parser 5 recognizes the vocalized sequence "hairy" and stacks it as a character on the adjective phrase "adj" representing "black" as shown in the state stack 155. Immediately after the “hair” recognition process, a pause is detected by the silent section detection unit 30, and a detection signal is input from the detection unit 30 to the LR parser 5 to indicate the pose as “△” and “hair”. It is piled up on the character of ". Since the character "hair" in the state stack 155 is included in the voice recognition dictionary, it is converted into the character "NP" representing the noun phrase as shown in the state stack 156.

Further, in the state stack 156, the syntax rule that the adjective “adj” and the noun phrase “NP” become the noun phrase “NP” is applied, resulting in “black”.
The converted adjective phrase “adj” and the converted “hair” noun phrase “NP” are converted to the noun phrase “NP” as shown in the state stack 157. That is, the noun phrase “NP” in the state stack 157 is “black hair”.
Represents Further, in the state stack 157, the adjective phrase “adj” and the noun phrase “NP” are the noun phrase “N”.
The syntax rule of becoming "P" is applied, and "clean"
The converted adjective phrase “adj” and the converted noun phrase “NP” representing “black hair” are converted to the noun phrase “NP” as shown in the state stack 158. This allows "clean" to be perceived as a "hairy" structure.

Next, a sequence of vocalized voices "girl" is recognized and stacked as characters on "△" indicating a pose as shown in the state stack 159, and "girl" is displayed in the state stack 159. Since the character is in the voice recognition dictionary, it is converted to the character "NP" representing a noun phrase as shown in state stack 160. Here, in the state stack 160, the noun phrase “NP” representing “beautiful black hair” and the noun phrase “N” converted from “girl” are converted.
The syntax rule is not applied to “P” because “Δ” indicating a pause is stacked between them. Then, the sequence of the vocalized speech "saw" is recognized, and as shown in the state stack 161, "girl" is stacked as a character on the converted noun phrase "NP", and the "saw" in the state stack 161 is read. Since "ta" is included in the voice recognition dictionary, it is converted into "VP" representing a verb phrase as shown in the state stack 162.

Then, the LR parser 5 determines that the end of the last pause unit has been reached, and the state stack 1
The syntax in the LR table memory 13 is the noun phrase "NP" representing "beautiful black hair" and the noun phrase "NP" representing "girl" located before and after "△" that displays the pose in 62. The rules are applied to convert the noun phrase to “NP” as shown in the state stack 163, and further to the letter “S” representing a sentence as shown in the state stack 164, so that “pretty” becomes “ Only the voice recognition result having the structure related to "black hair" is output.

The continuous speech recognition apparatus 81 of the first embodiment described above is provided with the silent section and the like detecting section 30 which detects the silent section and outputs the detection signal, and the LR parser 5 is the silent section and the like detecting section 30. HMM-LR with section restriction using the HMM-LR method by reading the data of the section of speech indicated by the detection signal input from
When the processing is executed and the end of the last division unit is reached, there is no section restriction H for one sentence of the voiced speech input.
The voice recognition result data is output by executing the MM-LR process. With this, after determining the dependency relation in the speech section of each delimiter unit, it is possible to determine the dependency relation between words, phrases, or clauses belonging to the speech sections of different delimiter units. The ambiguity can be resolved.

<Second Embodiment> FIG. 7 is a block diagram of a continuous speech recognition apparatus 82 according to a second embodiment of the present invention. Continuous speech recognition device 8 of the second embodiment of FIG.
2 is configured by including a hidden Markov network memory 11a in place of the hidden Markov network memory 11 of the continuous speech recognition apparatus 81 of the first embodiment of FIG. 1 and excluding the silent section detection unit 30. In the continuous speech recognition device 82 of the second embodiment, a model in which silent sections such as pauses, redundant words, and breaks based on prosodic information and the like are modeled by HMM is stored in the hidden Markov network memory 11a. The phoneme matching unit 4 detects a silent section or the like using the model.

FIG. 8 is a flowchart showing a voice recognition process executed in the continuous voice recognition device 82 of FIG. Hereinafter, the voice recognition process of the continuous voice recognition device 82 according to the second embodiment will be described with reference to FIG. First, in step S10, initialization of the HMM work area and initialization of the LR parser 5 are executed. Specifically, one cell of state stack 0 is created. And step S
In 11, for example, the voice data is read for each voice frame (for example, 20 milliseconds) which is the processing unit of the characteristic parameter, and the section-limited HMM-LR process is executed in step S12. Next, in step S13, it is determined whether or not a silent section or the like is detected. If the silent section or the like is not detected, the process proceeds to step S11 and the processes of steps S11 and S12 are repeated. It proceeds to step S14.

In step S14, it is determined whether or not the voice recognition processing for all the voice sections has been completed. If the processing for all the voice sections has not been completed (step S14).
14), the process proceeds to step S11, step S
When the processing of 11, S12, and S13 is repeated and it is determined that the processing of all the voice sections is completed (step S14)
(YES in step S15), the process proceeds to step S15, the section-unrestricted HMM-LR process is executed for one sentence of the input uttered voice, and the voice recognition process ends.

The continuous speech recognizer 82 of the second embodiment described above detects the detection of a silent section or the like by the hidden Markov network memory 11.
The phoneme collation unit 4 uses a model of the silent section or the like modeled by the HMM stored in a, and the LR parser 5 reads the voice data and sets the HMM for one voice section.
-A section-restricted HMM-LR processing using the LR method is executed, and when processing for each speech section is completed, section-unrestricted HMM-LR for one sentence of the input uttered speech
The voice recognition result data is output by executing the processing. With this, after determining the dependency relations within the speech segment of each break unit, it is possible to determine the dependency relations between words, phrases, or clauses belonging to different speech segments, so that the ambiguity of the dependency relations in the syntactic analysis can be determined. It can be resolved.

In the above first and second embodiments,
There is no section restriction for one sentence of the input voice
The voice recognition result data is output by executing the MM-LR process. However, the present invention is not limited to this, and one phrase such as one phrase or clause of the input uttered speech is
HMM without section restriction for vocalization of one sequence
-LR processing may be executed, or there is no section restriction HMM-L for voiced speech input from when the switch of the continuous speech recognition device is turned on to when it is turned off.
R processing may be executed. Even with the above configuration, the same operation and the same effects as those of the first and second embodiments are achieved.

In the above first and second embodiments,
Although the speech recognition apparatus using the HMM-LR method has been described, the present invention is not limited to this, and can be applied to other types of speech recognition apparatuses such as a speech recognition apparatus using a neural network.

[0044]

According to the first aspect of the present invention, the speech recognition apparatus detects at least one of a pause, a redundant word, and a boundary of a phrase or a clause based on the input uttered voice and outputs a detection signal. The voice recognition means determines the dependency relation in the syntactic analysis based on the detection signal and performs voice recognition of the uttered voice. As a result, the ambiguity of the dependency relation in the syntactic analysis can be resolved.

The speech recognition apparatus according to claim 2 is the speech recognition apparatus according to claim 1, wherein the speech recognition means uses at least one of the pause, a redundant word, and a boundary of a phrase or a clause. After voice recognition processing is performed on each voice section of the input uttered voice composed of a plurality of divided voice sections, the dependency relation between words, phrases, or clauses belonging to different voice sections is determined, and the input is performed as described above. The voice recognition of the vocalized voice. As a result, the ambiguity of the dependency relation in the syntactic analysis can be resolved.

Further, the voice recognition device according to claim 3 is:
The voice recognition device according to claim 1 or 2, wherein the detecting means has a first condition that the power of the uttered voice is equal to or less than a predetermined threshold value within a predetermined time range, and a zero cross of the uttered voice. The above pose is detected by detecting that at least one of the second conditions, which is equal to or larger than a predetermined threshold value, is satisfied during a predetermined time. Thereby, the dependency relation in the syntactic analysis can be determined based on the pose, and the ambiguity of the dependency relation in the syntactic analysis can be resolved.

Furthermore, the speech recognition apparatus according to claim 4 is the speech recognition apparatus according to claim 1 or 2, wherein the detection means is at least one of the pause, the redundant word, and the boundary of the phrase or the clause. Are detected by determining whether or not they match the respective predetermined language models. Accordingly, at least one of the pause, the redundant word, and the boundary of the phrase or the clause can be detected in the voice recognition process, and the ambiguity of the dependency relation in the syntactic analysis can be resolved.

[Brief description of the drawings]

FIG. 1 is a block diagram of a continuous voice recognition device according to a first embodiment of the present invention.

2 is a diagram showing an example sentence having first and second dependency relationships used to describe a voice recognition operation of the continuous voice recognition device 81 of FIG. 1. FIG.

FIG. 3 is a diagram showing an example of a voice recognition operation of a continuous voice recognition device 81 of FIG. 1 in a stack format.

4 is a diagram showing an example of a voice recognition operation of the continuous voice recognition device 81 of FIG. 1 different from that of FIG. 3 in a stack format.

5 is a diagram showing a voice section, a pause (silent section, etc.), and a pause unit (separation unit) of the example sentence of FIG.

6 is an LR parser 5 of the continuous speech recognition device 81 of FIG.
It is a flowchart which shows the voice recognition process performed by.

FIG. 7 is a block diagram of a continuous voice recognition device 82 according to a second embodiment of the present invention.

8 is an LR parser 5 of the continuous speech recognizer 82 of FIG.
It is a flowchart which shows the voice recognition process performed by.

FIG. 9 is a diagram showing a speech recognition operation of a continuous speech recognition apparatus of a conventional example in a stack format.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 ... Microphone, 2 ... Feature extraction part, 3 ... Buffer memory, 4 ... Phoneme matching part, 5 ... LR parser, 11, 11a ... Hidden Markov network memory, 12 ... Speaker model memory, 13 ... LR table memory, 20 ... Context-free grammar database memory, 30 ... Silent section detection unit.

Claims (4)

[Claims] 1. A voice recognition device comprising voice recognition means for voice-recognizing an input uttered voice and outputting a voice recognition result, wherein a pause, a redundant word, and a boundary between a phrase or a clause based on the input uttered voice. And a detection unit that outputs a detection signal by detecting at least one of the above, and the voice recognition unit determines the dependency relation in the syntactic analysis based on the detection signal to recognize the voice of the uttered voice. A voice recognition device characterized by: 2. The voice recognizing means outputs a voice for each voice section of an input uttered voice, which is composed of a plurality of voice sections divided by at least one of the pause, a redundant word, and a boundary of a phrase or a clause. The speech recognition according to claim 1, wherein after the recognition processing, a dependency relation between words, phrases, or clauses belonging to different speech sections is determined, and speech recognition of the inputted uttered speech is performed. apparatus. 3. The detecting means comprises a first condition that the power of the uttered voice is equal to or less than a predetermined threshold value within a predetermined time range, and the number of zero crosses of the uttered voice is a predetermined time. 3. The voice according to claim 1, wherein the pause is detected by detecting that at least one of the second condition that is equal to or more than a predetermined threshold is satisfied during Recognition device. 4. The detection means detects at least one of the pose, the redundant word, and the boundary of a phrase or a clause by determining whether or not the pose matches a predetermined language model. The method according to claim 1 or 2, wherein
The speech recognition device according to the above.