JPS61148496A - Continuous voice recognition equipment - Google Patents

Abstract

(57) [Summary] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

DETAILED DESCRIPTION OF THE INVENTION (Field of Industrial Application) The present invention relates to a continuous speech recognition device, and particularly relates to an improvement of a device that recognizes sentence speech continuously uttered according to grammar.

(Prior art) Among speech recognition devices, devices that recognize sentence sounds uttered according to grammar can recognize computer programs, limited business texts, air traffic control and control commands for various equipment, etc., and have a wide range of applications. have. It is known in principle that when grammatical constraints are given, misrecognition can be prevented by using the grammatical rules. In particular, when there is a restriction on the number of digits in input speech in continuous number recognition, the recognition rate can be improved by regularizing the restriction.

A method of recognizing sentence sounds continuously uttered according to such a grammar was proposed in Japanese Patent Application No. 59-68015 by the inventor of the present application.
``Continuous Speech Recognition Device'' is described in the specification of No. K.

The principle of this diagonal blockwise DP matching method is as follows. Expressing the grammar with automaton α,
The automaton α is defined as follows.

α−<K, Σ, ΔtPo, F> ・・・・・・m1・
(1) Here, K: set of states p (plp-1, 2°・
...π) Σ: Set of input words n (rlIn"1゜2,...N) Δ: State transition rule [(p, q, n)] where,
(p, q, n) is p! −q means a state transition.

Po: initial state, hereinafter indicated as p-0.

F: Final state set FcK Next, the speech pattern A obtained by continuously uttering the word 1EE-Σ according to the automaton α is A z al ,
az, ・-al+ ・++ az “・” (
2), and this is called an (unknown) input pattern. Weight 1
The standard pattern for lner is B"-b:, b star,...b"...b=...
・・・・・・・・・(3)°j°.

is prepared and called a word standard pattern. Continuous speech standard patterns C-B", B"2. which are obtained by connecting this word standard pattern B" according to the automaton α.
...Perform DP matching between B' and the incoming person,
A quantity representing the degree of mutual difference between two patterns (hereinafter referred to as a degree of dissimilarity) is calculated, and a word sequence that provides the minimum degree of dissimilarity is taken as a recognition result.

Here, the minimum degree of dissimilarity is found using the following dynamic programming method. The initial condition is T(o,o)−.

T(m+q)=”, m〆0eQ−〇−(4)G(p
.
``Based on the boundary conditions of

n) Calculate for all pairs (p, n) such that εΔ.

That is, the boundary condition is set, and msj=ms for each standard pattern time j of J-1e...Jo. 10 (j - a) m@j ``m 穆 j + IL - Matadashiku Then, the boundary values g(mej*j)-11(mej+j) are stored in the table storage G(p, n, D, H(p, n, j), respectively).

After the calculation of equation (7) is completed at standard pattern time j−Jζ, as described above, (7
) is calculated by dividing the input pattern into blocks for each elm frame as shown in FIG. 2, and executing the calculation for each diagonal block that is further tilted diagonally.

Finally, the recognition result of the input pattern is determined by the following procedure as a determination process.

Initial condition q=argmin ・・・・・・
(9) Find qEF Q”q, m-I (10).

if? > o, then q-6+ In -'j (11
) repeat the formula. l! If it is -0, it ends.

(Problems with the Prior Art) In the method of the above-mentioned Japanese Patent Application No. 59-68015, the standard pattern is converted into G(p, n, j) which is an intermediate result of calculation.
, H(p.

n+J) read and write times are block width I
Inversely proportional to LK, the larger the block width IL, the shorter the memory access time.

On the other hand, for this block width IL, if %aBL is the block slope, then ml n (J"'J / aBL 4 IL
..." (12) The maximum block width depends on the minimum number of frames of the standard pattern. For example,
aBL is equal to the maximum slope of the DP matching bus, usually 2
Therefore, if the minimum number of frames of the standard pattern is 10, it can be set to IL-5. However, with IL-5, calculations for standard patterns smaller than 10 frames cannot be performed.

In this way, in the conventional method, if there is a standard pattern that is shorter than the ILxaBL frame, the block width IL must be changed to a smaller value, which increases memory access time and reduces the response time until recognition results are obtained. It had the disadvantage of being large.

In addition, the block width IL must be determined by the control section of the recurrence formula calculation, and if the user cannot change it, the block width IL can be determined using the standard pattern or ILxa.
There was a drawback that the registration had to be repeated again until the BL frame was longer. Furthermore, when the device is constructed so that the block width IL can be changed, the circuit becomes complicated.

(Objective of the Invention) The object of the present invention is to expand the short standard pattern and replace it with a standard pattern whose length is longer than the allowable pattern length, if there is a standard pattern that is shorter than the allowable pattern length when registering the standard pattern. By solving the above drawback, the block width I
To provide a continuous speech recognition device that can always output recognition results within a short constant response time without changing L.

(Structure of the Invention) The structure of the continuous speech recognition device of the present invention includes a pattern length verification section that verifies whether or not the standard pattern length is longer than the allowable pattern length when registering a standard pattern; an extension pattern section that extends the standard pattern to a length greater than the allowable pattern length when the standard pattern length is determined to be short; and a continuous standard that connects word standard patterns for all combinations specified by the finite state automaton. A diagonal blockwise DP matching unit that calculates the minimum distance between a pattern and an input pattern using dynamic programming for each diagonal block that has a predetermined time width on the input pattern and is tilted with respect to the standard pattern axis; The present invention is characterized by comprising a recognition output unit that outputs a combination of words that yields the minimum distance determined by the block quiz DP matching unit as a recognition result.

(Example) Next, the present invention will be explained in detail with reference to the drawings.

FIG. 1 is a block diagram showing one embodiment of the present invention. When a voice is input from the microphone 1, the voice analysis unit 2 performs voice analysis, for example, frequency analysis, converts it into a vector time series indicating characteristics, and at the same time performs voice detection to determine the time interval in which the voice exists. is required.

The user first utters the voice of the set of words Σ-(n) and registers it in the standard pattern memory 6 as a standard pattern. When performing this registration, switch S1 is pushed down. The speech segment length l obtained from the speech analysis section 2 is sent to the pattern length verification section 3, where it is compared with the allowable pattern length/mt by a comparator. Also, in the case of /:hemt, the switch S2. S3 is turned to the A side, and the time series of feature vectors obtained by the speech analysis section 2 is stored in the standard meta/memory 6 as a standard pattern.

On the other hand, if l<1H1t, the switch s2. s3 is moved to the B side, and the feature vector time series obtained by the speech analysis section 2 is sent to the pattern expansion section 4. In this pattern expansion section 4, 1mt-7 feature vectors are inserted at equal intervals, and l! The frame pattern is linearly expanded to become a 1mt7 frame pattern. In this case, the feature vector to be inserted is the same as the adjacent feature vector.

The pattern expanded by the pattern expansion section 4 is stored in the standard pattern memory 6.

After all the words are registered as described above, the switch S1 is turned to the R side and recognition is started. The unknown input voice is converted into a time series of feature vectors by the voice analysis unit 2 and stored in the input pattern memory 5, as in the case of registration.

Next, in the diagonal blockwise DP matching section 7, DP matching is performed between the input pattern and the standard pattern.
(5) and (6) using equation (4) as the initial value.

Equations (7) and (8) are calculated. As this diagonal blockwise DP matching section 7, the embodiment described in Japanese Patent Application No. 59-68015 can be used.

Finally, as an output of the recognition result, the determination unit 8 outputs (9), (
10) and (11) are calculated and the recognition result n is output. This determining section 8 can also use the determining section described in Japanese Patent Application No. 59-68015.

Although the present invention has been described above based on examples, these descriptions do not limit the scope of the rights of the present invention. Although the pattern expansion section of this embodiment linearly expands the pattern, a method of non-linearly expanding the pattern is also conceivable. For example, perform DP matching with a pattern already registered in the same category, use the resulting DP matching path to find the time correspondence between the pattern you want to expand and the already registered pattern, and follow that time correspondence. A method of elongation is also considered.

(Effects of the Invention) As explained above, in the continuous speech NR device of the present invention,
The standard pattern can always be made longer than the allowable pattern length by verifying the pattern length at the time of registration and expanding the pattern if it is shorter than the allowable pattern length. This eliminates the need to reduce the block width IL, which is the calculation unit of the diagonal blockwise Dr matching section, and it is possible to always obtain recognition results within a short constant response time. Also,
In the conventional method, if the standard pattern was short, it was necessary to re-register it until the pattern became long, but with the present invention, there is no need to re-register it, thereby improving usability.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing an embodiment of the present invention, and FIG. 2 is a diagram showing a calculation procedure for diagonal blockwise DP matching. In the figure, 1...Microphone, 2...Speech analysis section, 3...Pattern length verification section, 4...
Pattern extension section, 5... Input pattern memo 1c 6
...Standard pattern memo 1c 7...Diagonal blockwise DP matching section, 8...Judgment section,

Claims (1)

[Claims] In a continuous speech recognition device that recognizes speech in which a string of words specified by a finite state automaton is continuously uttered by performing DP matching with a standard pattern, when registering the standard pattern, is the standard pattern length shorter than the allowable pattern length? a pattern length testing section that tests whether or not the standard pattern length is short; an extension pattern section that extends the standard pattern beyond the allowable pattern length when the pattern length testing section determines that the standard pattern length is short; and the finite state automaton. For all specified combinations, the minimum distance between the input pattern and a continuous standard pattern in which word standard patterns are connected is calculated by moving each diagonal block with a predetermined time width on the input pattern and tilted with respect to the standard pattern axis. The present invention is characterized by comprising a diagonal blockwise DP matching section that calculates using a logical programming method, and a recognition output section that outputs a combination of words that yields the minimum distance determined by the diagonal blockwise DP matching section as a recognition result. Continuous sound recognition device.