JP3340163B2 - Voice recognition device - Google Patents

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 suitable for controlling a device by recognizing a voice uttered by a human.

[0002]

2. Description of the Related Art In a speech recognition apparatus of this type, a speech uttered by a user is recognized by a predetermined recognition method, and the recognition result is displayed, for example, up to the third place as a selection candidate in descending order of similarity. Thus, the user can select one of them.

[0003]

However, conventionally, when selection candidates are displayed in the order of similarity according to a specific recognition method, candidates that are not similar even to the user's sense often appear together, There was a problem that a user could feel very uncomfortable.

[0004] The present invention has been made in view of the above circumstances, and its object is to display a candidate that is not extremely similar when the recognition candidate is externally output and selected by a user based on the result of speech recognition. It is an object of the present invention to provide a speech recognition device capable of preventing the voice recognition.

[0005]

SUMMARY OF THE INVENTION The present invention provides a voice input unit for inputting a voice uttered by a user, an analysis / feature extraction unit for analyzing the input voice and extracting a feature amount, and an extraction unit. a recognition means for obtaining a feature amount of one or more recognition candidates by recognizing speech using a listening erroneously
Group similar easy candidates in advance and register
Registration means are, in the first place obtained by the recognition means
Based on the recognition candidate, the candidate group registered in the registration means
Group and identify them with candidates and recognition means belonging to this group.
Candidates matching with the obtained second and subsequent recognition candidates
A selection candidate determining means for determining a first position determined by the recognition means
Of the candidate for recognition and the candidate obtained by the selection candidate determining means.
In addition, a candidate table is displayed so that these determined candidates can be selected.
And means shown in the drawings .

[0006]

In the above arrangement, the voice uttered by the user is input by the voice input means, analyzed by the analysis / feature extraction means, and the feature quantity is extracted. Thereafter, the feature quantity is used by the recognition means. A recognition process is performed to obtain a recognition candidate. The selection candidate determination unit refers to the registration unit based on the first-ranked recognition candidate obtained by the recognition unit. This registration means, by mistake for the user listen
Similar candidates that are easy to remove are grouped in advance and registered. This registration information can be obtained experimentally by summing up the results obtained by a plurality of listeners recognizing words uttered by a plurality of speakers one by one.

[0007] Therefore, the selection candidate deciding means, as described above,
A candidate group is specified by referring to the registration means based on the first-ranked recognition candidate, and candidates belonging to this group are identified.
Between the second and subsequent recognition candidates obtained by the recognition means
Find a matching candidate. The candidate display means is configured to recognize the first place.
Combine knowledge candidates and candidates obtained by the selection candidate determination means
So that the user can select these requested candidates.
Complementary display.

As described above, according to the above configuration, candidates recognized by the recognition means are sorted (for example, in descending order of similarity).
Not directly be selected by the user indicates the table, the second
The recognition candidates after the first rank are similar to the first recognition candidates.
Since the user selects only candidates that are easy to hear by mistake and makes the user select it, the user does not feel uncomfortable.

[0009]

DESCRIPTION OF THE PREFERRED EMBODIMENTS One embodiment of the present invention will be described below with reference to the drawings, taking as an example a case where the present invention is applied to a voice recognition device used in a ticket vending machine at a station. FIG. 1 is a block diagram schematically showing the configuration of the speech recognition apparatus in the embodiment.

In FIG. 1, reference numeral 1 denotes an A / D (analog / digital) audio signal (input audio) input to the apparatus.
An A / D converter for conversion. The A / D converter 1 quantizes the input voice at, for example, a sampling frequency of 12 kHz and 12 bits.

The input voice quantized by the A / D converter 1 is supplied to an analysis / feature extraction unit 2 for analyzing the voice and extracting a feature. The analysis / feature extraction unit 2
The voice power of the input voice quantized by the A / D converter 1 is calculated, and an LPC (Linear Predictive Coding) analysis is performed. This LPC analysis is performed, for example, for a frame length of 16
The analysis is performed using the LPC mel-cepstral of the 16th order with msec and a frame period of 8 msec as analysis parameters. The analysis and
The analysis in the feature extraction unit 2 is not limited to LPC analysis, but may be BPF (Band Pass Filter) analysis or the like.

The feature parameters analyzed by the analysis / feature extraction unit 2 are given to a continuous matching unit 3. The continuous matching unit 3 includes a speech segment (Phonetic Segment)
A matching process is continuously performed in the time axis direction with a predetermined PS unit recognition dictionary registered in a complex dictionary unit (hereinafter, referred to as a PS complex dictionary unit) 4 to perform the first to n-th ranks. To determine the label sequence (PS label sequence) and the similarity thereof. The above audio segment (PS) is described in detail in, for example, Japanese Patent Application No. 2-306061. In addition, the above recognition dictionary is used for each PS (PS label).
It consists of an identification dictionary created from a plurality of standard patterns for each. The continuous matching process by the PS in the continuous matching unit 3 is performed using a composite LPC mel-cepstral similarity scale shown in the following equation.

[0013]

(Equation 1)

In the equation (1), C is an LPC mel-cepstrum, W _m ^(Ki) and φ _m ^(ki) are PS names K, respectively.
The weight and eigenvector obtained from the eigenvalue of i.
Also, (•) indicates the inner product, and {} indicates the norm.

[0015] Of the PS label sequences obtained by the continuous matching unit 3, the first sequence is an HMM (hidden marco).
v model; Hidden Markov Model) is sent to the HMM recognition unit 5 for performing word matching.

The word matching in the HMM recognition unit 5 will be described. First, word matching is performed by passing the first-ranked PS label sequence sent from the continuous matching unit 3 through an HMM for each word (for each category).

Here, a general formulation of the HMM will be described. The HMM has N states S ₁ , S ₂ ,..., S _N , and the initial states are stochastically distributed among these N states. For speech, a model is used in which the state transitions at a certain probability (transition probability) every fixed frame period.
At the time of transition, a label is output with a certain probability (output probability), but a null transition that transitions between states without outputting a label may be introduced. Even if an output label sequence is given, a state transition sequence is not uniquely determined. Since only the label sequence can be observed, the hidden (hidden) marcov model
(HMM). The HMM model M is 6
It is defined from two parameters. N: number of states (states S ₁ , S ₂ ,..., _SN ) K: number of labels (labels R = 1, 2,..., K) p _ij : transition probability Probability q _ij (transition from S _i to S _j ) k): probability of outputting a label k at the time of transition to the S _j from S _i m _i: probability initial state probability initial state is S _i F: the set of final states

Next, a restriction on the transition reflecting the characteristics of the voice is added to the model M. In speech, generally state S _i
Loop transitions that return to the previously passed state (S _i−1 , S _i−2 ,...) Are not allowed because they disturb the temporal context.

The evaluation of the HMM is based on the probability Pr that the model M outputs the _first label sequence O ₁ = o ₁₁ , o ₂₁ ,..., O _T1.
(O / M). At the time of recognition,
Assuming each model in the HMM recognizing unit 5, the first label sequence (PS label sequence) sent from the continuous matching unit 3
Is used to find a model M1 that maximizes the probability Pr (O / M), and sets it as the first recognition candidate. Similarly,
The HMM recognizing unit 5 sets the model M2 having the next highest probability as the second recognition candidate, and sets the model M3 having the next highest probability as the third recognition candidate. However, when the probability is smaller than a predetermined value, it is not regarded as a recognition candidate. This H
(Parameters of) each model assumed by the MM recognition unit 5
Is obtained by learning the HMM.
It is stored in the buffer 6. As described above, by performing the recognition processing of the uttered input voice, the input voice, for example, the name of the station at the destination can be recognized with high accuracy.

Now, the speech recognition apparatus shown in FIG.
In addition to the D converter 1, analysis / feature extraction unit 2, continuous matching unit 3, PS compound dictionary unit 4, HMM recognition unit 5, and HMM buffer 6, a display unit 7, a selection unit 8, a candidate display restriction unit 9, and a control unit It has a part 10.

Under the control of the control unit 10, the display unit 7 displays, for example, an invitation to utter the destination station name (here, the display of the character string "Please utter"), and the recognition result of the HMM recognizing unit 5. (Recognition candidates) are displayed. As shown in FIGS. 3A and 3B, the display of the recognition result by the display unit 7 can be performed by displaying a plurality of candidates, for example, even if only the first-ranked recognition candidate (here, “Komono”) is displayed. As shown in FIG. 3 (c), the first
The position recognition candidate (here, “Komono”) and the second position recognition candidate (here, “Komano”) may be displayed.

The selection unit 8 is configured so that a user can select a recognition candidate displayed on the display unit 7.
The selection unit 8 and the display unit 7 are configured using a touch panel in which a pressure-sensitive transparent tablet is combined on a liquid crystal display (a display monitor such as a CRT display may be used).

When displaying the plurality of recognition candidates obtained by the HMM recognizing unit 5 on the display unit 7, the candidate display restricting unit 9 may, for example, place the second and lower candidates in a position extremely different from the first candidate. This is to restrict things that are not similar from being displayed. As shown in FIG. 4, the candidate display restriction unit 9 determines whether the first candidate is allowed to be displayed in the second place or less, specifically, when the first candidate is uttered. Is a table (database) that registers (accumulates) words that can be easily heard by humans for each candidate that can be the first place. The registration information in this table can be obtained experimentally by summing up results obtained by a plurality of listeners recognizing words (station names in this example) spoken by a plurality of speakers one by one.

The control unit 10 performs display control on the display unit 7 in accordance with the recognition result of the HMM recognizing unit 5, the registered contents of the candidate display restricting unit 9 and the like, and according to the selection instruction information from the selection unit 8, an external device (here, , Ticket vending machines). FIG. 2 shows a display unit 7 and a selection unit 8 in the voice recognition device of FIG.
Is shown.

As shown in FIG. 2, the display unit 7 comprises a liquid crystal display 71 and a display memory 72 for storing display information to be displayed on the display 71.
The display information in the display memory 72 is written by the control unit 10 shown in FIG. The contents displayed on the liquid crystal display 71 include, in addition to the display information for solicitation invitation and the display information of the recognition candidates, confirmation of the displayed recognition candidates (in the case of displaying only the first candidate). Display information for inviting (character string "please confirm"), display information of the area (key area) used for user's confirmation operation on the screen (item key of [Confirm]), (displayed Display information of an area (key area) for accepting re-speaking by the user when the recognized recognition candidate is an error, that is, when a candidate intended by the user is not displayed (item key of [rephrase]) And display information for inviting a selection operation of the recognition result (character string “please select”).

On the other hand, the selection unit 8 includes a pressure-sensitive sheet type transparent tablet 81 laminated on the display screen of the display unit 7, that is, on the liquid crystal display 71 and integrally formed with the display 71, and a transparent tablet 81. Is constituted by an indicated coordinate detecting unit 82 for detecting a coordinate position on the transparent tablet 81 surface when the user is pressed by a user's finger or the like, and an indicated information determining unit 83. The instruction information determination unit 83 determines which item key (display information) on the screen has been selected and designated by the user based on the coordinates detected by the instruction coordinate detection unit 82 and the contents of the display memory 72. The determination result is sent to the control unit 10 as selection instruction information.

In such a configuration, while the user utters “komono” as the destination station name in accordance with the invitation to utterance under the control of the control unit 10, the HMM recognition unit 5
It is assumed that a recognition process for the voice is performed, and a plurality of recognition candidates are sent to the control unit 10 as a recognition result.
It is assumed that the recognition result includes the first-ranked recognition candidate “Kono” and the second-ranked recognition candidate “Komano”.

The control unit 10 sets a fixed period (or a period until the user selects a candidate) starting from a certain time before the vocal invitation solicitation timing and sandwiching the same timing.
The HMM recognition unit 5 is configured to receive all the recognition results. By doing so, the restriction on the timing of the utterance of the user is relaxed. Instead of setting the receiving period of the control unit 10 for the recognition result of the HMM recognizing unit 5, the operating period of the HMM recognizing unit 5 may be set. In addition, a switch for setting the input period of the voice to be recognized is provided, and the voice uttered by the user is input to the apparatus only while the user turns on this switch, and all the input voices are input. HMM recognition unit 5
, And the control unit 10 may receive the recognition result.

The control unit 10 receives a plurality of recognition candidates including the first-ranked recognition candidate “Komono” sent from the HMM recognition unit 5 and displays the plurality of recognition candidates on the display screen of the display unit 7 (the liquid crystal display 71 thereof). For example, as shown in FIG. 3A, only the first-ranked recognition candidate “Kono” is displayed. At the same time, on this screen, for example, an invitation message of "Please confirm" is displayed at the top of the screen, an item key [Confirm] is displayed at the lower right of the screen, and an item key [Restate] is displayed at the lower left of the screen.

Here, the recognition candidate "Komono" displayed on the screen is displayed.
If the is correct, the user follows the request of "Please confirm", and the item key of [Confirmation] at the bottom right of the display screen
Is touched on the transparent tablet 81 with a finger.

Then, the coordinates of the item key of “confirmation” are detected by the designated coordinate detecting section 82. The instruction information determination unit 83 determines that the item key indicated by the display information displayed at the detected coordinate position has been selected from the coordinates detected by the instruction coordinate detection unit 82 and the contents of the display memory 72, namely, ] Is selected, and selection instruction information indicating the selection is sent to the control unit 10. Thereby, the control unit 10
Controls the external device (ticket vending machine) assuming that the first candidate "Kono" has been confirmed.

Instead of displaying the item key of [confirmation] as shown in FIG. 3 (a), the item key of [confirmation] is added to the first recognition candidate itself as shown in FIG. 3 (b). And the confirmation area may be input by touching the display area of “Kono” with a finger.

When the user utters “komono”, as shown in FIG. 3C, for example, the first candidate “Komono” and the second candidate “Komano” are displayed as “confirmed”. When the user touches the display area of “Komono” with a finger according to the request of “Please select”, the instruction information determination unit 83 displays “ “Kono” is selected, and selection instruction information indicating that is selected is sent to the control unit 10. If the candidate intended by the user is not displayed, the user selects the [Rephrase] item key (area).
Is touched on the transparent tablet 81 with a finger.

Then, in the same manner as when the (confirmation) item key (area) is touched with a finger, the instruction information determination unit 83 in the selection unit 8 selects [rephrase]. Is determined, and the selection instruction information indicating that is
Sent to 0. Thereby, the control unit 10 [restate]
(Respeak) is determined to have been requested, the mode is set to the rephrasing mode (replay mode), and solicitation for resound is performed by, for example, a “p” sound. According to the invitation, the user can re-utter the destination station name that has not been correctly recognized.

In this embodiment, only the first candidate is displayed as shown in FIG. 3 (a) or (b), or only the second candidate is displayed as shown in FIG. 3 (c). Whether to do so is determined by the control unit 10. There are two determination conditions by the control unit 10.

The first condition is whether the difference between the first and second similarities exceeds a first predetermined value, or if the second similarity value is less than a second predetermined value. It is or not. The second condition is whether or not the second candidate is registered in the candidate display restriction unit 9 in a pair with the first candidate.

When the first condition is satisfied, only the first candidate is displayed regardless of whether the second condition is satisfied or not. If the first condition is not satisfied, the second candidate is displayed if the second condition is satisfied, and only the first candidate is displayed if the second condition is not satisfied.

As described above, in this embodiment, when the probability of the second candidate is low, only the first candidate is displayed so that the user can easily select the candidate. Further, even when the probability of the second candidate is high, if the second candidate is not registered in the candidate display restriction unit 9 in a pair with the first candidate, the second candidate is not registered. Is not displayed as not extremely similar to the first candidate, so that the user does not feel uncomfortable.

Here, because the first condition is not satisfied, it is determined whether to display the candidates up to the second place or only the first place candidates according to whether the second condition is satisfied or not. A specific example to be determined will be described.

FIG. 5 shows the first to third candidates of the recognition result of the HMM recognizing unit 5 in response to the utterance input of a certain user (speaker). However, only the case where the uttered word is not the first candidate is shown.

In FIG. 5, for example, in response to an utterance input of “Niigami”, the first place of the recognition candidate is “Jujo” (Jujou), the second place is “Niigami”, and the third place is “Nishi no Kyo”. is there. On the other hand, as shown in FIG. 4, the candidate display restriction unit 9 allows only “Shinjo” and “Nigami” to be displayed (below the second place) for the first place candidate “Jujo”. ing. Therefore, when the second condition is to be displayed because the first condition is not satisfied, the first position candidate “Jujo” is displayed.
In addition, the second place candidate “Niigami” is also displayed on the display unit 7.

In the above example, even if the third place candidate is to be displayed, if the first place candidate is "Jujo", the third place candidate is displayed.
Since the candidate "Nishi no Kyo" is not allowed to be displayed by the candidate display restriction unit 9, only the first place "Jujo" and the second place "Nigami" are not displayed. That is, in the present embodiment, the third place candidate "Nishi no Kyo", which is felt to be completely dissimilar to "Jujo" or "Nikami" for the user, is not displayed even when trying to display the third place candidate. You don't feel uncomfortable.

For example, in response to the utterance input of "Shiraki" (Shiraki), the first place of the recognition candidate is "Shiraki" and the second place is "Chiyozaki" (Chiyozaki). On the other hand, as shown in FIG. 4, the candidate display restricting unit 9
No words are allowed to be displayed. Therefore, even when the first condition is not satisfied, even if the second candidate is to be displayed, only the first candidate “Shiroki” is displayed, and the user is not completely similar to “Shiroki”. The second-place candidate "Chiyozaki" that can be felt is not displayed.

In the above embodiment, the candidate display restricting section 9
Are included in the table (database) that constitutes
Although it has been described as being registered (stored) for each candidate that can be the first place, the present invention is not limited to this. For example, candidates may be registered in groups, and if any of the candidates in the group is the first candidate, the display of the candidates not included in the group may not be permitted.

In the above-described embodiment, the case where the selection candidates are displayed on the screen has been described. However, the present invention can be similarly applied to the case where the first candidates are sequentially output as voices and presented to the user. Further, the present invention is not limited to a speech recognition device used for a ticket vending machine at a station, and is applicable to all speech recognition devices.

[0046]

As described above, according to the speech recognition apparatus of the present invention , the second and subsequent recognition candidates among the recognition candidates inputted by the speech input means and recognized by the recognition means.
Was mistakenly heard similar to the first-ranked recognition candidate.
Only the candidates that are easy to understand, that is,
Exclude dissimilar candidates and match them with the top recognition candidate.
In addition , since it is configured to display candidates so that the user can select it, even if it is an error specific to the applied recognition method,
Even if the candidate recognized by the recognition means includes a candidate that does not resemble the user's sense, this type of candidate is prevented from being selected, and the user does not feel uncomfortable. Only

[Brief description of the drawings]

FIG. 1 is a block diagram showing a basic configuration of a speech recognition device according to one embodiment of the present invention.

FIG. 2 is a block diagram showing a configuration of a display unit 7 and a selection unit 8 in FIG.

FIG. 3 is an exemplary view showing an example of a display screen for explaining an operation in the embodiment.

FIG. 4 is a view showing an example of the contents of a table constituting a candidate display restriction unit 9 in FIG. 1;

FIG. 5 is an HMM in FIG. 1 for a user's utterance input;
The figure which shows an example of the recognition result by the recognition part 5.

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 1 ... A / D converter (voice input means), 2 ... Analysis and feature extraction part, 3 ... Continuous matching part, 4 ... PS compound dictionary part, 5
... HMM recognition section, 6 ... HMM buffer, 7 ... Display section, 8
... selection unit, 9 ... candidate display restriction unit (registration unit), 10 ... control unit (selection candidate determination unit), 71 ... liquid crystal display,
72: display memory; 81: transparent tablet; 82: designated coordinate detecting unit; 83: designated information determining unit.

──────────────────────────────────────────────────続 き Continued on the front page (58) Field surveyed (Int.Cl. ⁷ , DB name) G10L 15/22

Claims (1)

(57) [Claims] An audio input means for 1. A user inputs a speech uttered, analyzes the voice inputted by said voice input means, and analyzing and feature extraction means for extracting a feature value, the analysis and feature Recognition means for recognizing speech using the feature amount extracted by the extraction means to obtain one or more recognition candidates, and similar candidates which are apt to be heard by mistake and are grouped in advance.
A registration unit that has been converted and registered, and a first recognition candidate obtained by the recognition unit.
The candidate group registered in the registration means
And the candidates belonging to this group and the
Matching candidates with the second and subsequent recognition candidates
A selection candidate determining means that, first of the recognition candidates with the selected candidate which has been determined by the recognition means
Together with the candidates determined by the decision means,
Means for displaying candidates so that the candidates can be selected .