JP2979999B2 - 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, and more particularly to a word for calculating the likelihood of a specific word at regular time intervals (frames) during utterance and detecting the specific word based on the calculated likelihood. The present invention relates to a spotting device.

[0002]

2. Description of the Related Art Conventionally, as a technique for detecting a specific word from an utterance, the Transactions of the Institute of Electronics and Communication Engineers, '84 / 10, Vo
l. J67-D No. 10 p. 1242-1249
"Continuous Speech Recognition Algorithm Using Extended Continuous DP Method"
(Seiichi Nakagawa) is known as a word spotting device.

FIG. 8 is a block diagram showing an outline of the word spotting apparatus.
1, the voice analysis is performed, and the likelihood of the specific word for each frame is used as it is as the detection reference value, and the result output unit 103 performs detection by comparing with the detection threshold.
Note that the likelihood calculation unit 102 is configured to execute the phoneme dictionary 104 or the word dictionary 10.
5 and the like are calculated for each frame.

In this method, the likelihood calculated independently for each frame is used as it is as a detection reference value, and detection is performed by comparison with a predetermined detection threshold.

[0005]

In such a conventional method, the likelihood in a frame unit is used as a detection value as it is, so that a word whose likelihood is unstable and changes with time is unstable. In the case where the likelihood repeats up and down around the detection threshold, the same word may be detected multiple times consecutively, and there is also a problem that an instantaneous increase in the likelihood causes false detection. .

SUMMARY OF THE INVENTION It is an object of the present invention to reduce the harm caused by using the likelihood of a frame unit as a detection reference value as it is, to reduce the instability of the likelihood of a candidate word with time, and to achieve more robust word detection. An object of the present invention is to provide a voice recognition device that can perform the voice recognition.

[0007]

According to the present invention, speech recognition for calculating the likelihood of a specific word at regular time intervals (frames) during utterance and detecting the specific word based on the calculated likelihood. The apparatus includes a likelihood correction unit that corrects and derives the likelihood of each frame based on the likelihood of a current frame and the likelihood of at least one past frame.
Only, the likelihood correction means outputs the current frame and the immediately preceding frame.
If the likelihood of each frame is equal to or greater than a predetermined threshold, the threshold
Is derived as the modified likelihood, otherwise
Derives the likelihood of the current frame as the modified likelihood
Thus , a speech recognition device characterized by performing the above is obtained.

[0008]

According to the present invention, the likelihood of each frame is not independently used, but the likelihood of a current frame and at least one past frame are referred to reduce the instability of the temporal change of the likelihood. It is.

[0009]

Embodiments of the present invention will be described below with reference to the drawings.

FIG. 1 is a schematic block diagram of an embodiment of the present invention, and the same parts as those in FIG. 8 are denoted by the same reference numerals. FIG.
In, the voice analysis unit 101 performs frequency analysis of the input voice, and extracts a feature vector for each fixed time (frame synchronization). Likelihood calculation section 102 is speech analysis section 1
By performing matching between the time series of the feature vector from 01 and the phoneme dictionary 104 and the word dictionary 105, the likelihood of each candidate word in each frame period is calculated and output to the automaton unit 106.

The automaton section 106 is a likelihood calculation section 102
Then, the likelihood is output, the likelihood of each candidate word in each frame period is corrected, and the result is output to the output unit 103. The output unit 103 receives the likelihood of each candidate word, compares the likelihood with a predetermined threshold, and outputs the word as a detected word having a likelihood exceeding the threshold in each frame period.

The difference from the conventional FIG. 8 is that the automaton unit 1
This is because the likelihood calculated by the likelihood calculating unit 102 is corrected by using the same method as in 06. An example of the likelihood correction performed by the automaton unit 106 will be described below. The automaton unit 106 shown here receives the likelihood of each frame output from the likelihood calculation unit 102 as input likelihood, performs state transition by each case determination every time the likelihood is received, and then performs state transition when each case is established. Processing shall be performed.

FIG. 2 is a diagram showing the first embodiment. The automaton in this example is an automaton whose initial state is s1, and the likelihood input in the state s1 is output as it is. At this time, if the likelihood is equal to or more than B (detection threshold), the state transits to the state s2. Otherwise, the state stays in the state s1. The state s2 is a state in which the likelihood of the immediately preceding frame is equal to or greater than B. Here, while the input likelihood is equal to or greater than B, the state stays in the state s2 and continues to output a modified likelihood smaller than B. If not, the input likelihood is output as modified likelihood as it is, and the process returns to the state s1.

In this automaton, only when both the likelihood in the current frame and the likelihood in the immediately preceding frame exceed the detection threshold, a certain value less than the detection threshold is output as the modified likelihood. In other cases, the likelihood of the current frame is output as it is. The modified likelihood output using this automaton is equal to or greater than the detection threshold only for frames whose input likelihood has increased from less than the detection threshold to greater than or equal to the detection threshold.
If the likelihood of the first frame is equal to or greater than the detection threshold,
This likelihood is output.

By using such a modified likelihood, it is possible to prevent the same word from being continuously detected for several consecutive frames.

Incidentally, the value less than B as the modified likelihood in transition 2 in FIG. 2 may be extremely-値 if it is a value smaller than B. However, it is not practical. −
A value of 1,0 "may be output as the modified likelihood.

When the automaton shown in FIG. 2 is used, the detection threshold is increased from B to B '(B'> B) as shown in FIG. 3 in order to reduce the number of detections. However, in the case of B ', the number of times of detection may increase contrary to t2 to t4. Then, the second shown in FIG.
Is considered.

The operation of the automaton shown in FIG. 4 in the state s2 of the automaton shown in FIG. 2 is different.
In the state s2 of this automaton, the input likelihood is B2 (B2
<B) At the time, the value remains at s2, and a value less than B is output as the modified likelihood. Otherwise, the input likelihood is output as modified likelihood as it is, and the process returns to the state s1. B2 is a detection validation threshold.

In this automaton, in addition to the detection threshold, a detection enable threshold smaller than the detection threshold is determined, and even if the detection threshold is once exceeded, the likelihood exceeding the detection threshold is not output until the detection enable threshold is once exceeded. . As a result, the detection is not performed a plurality of times while the likelihood remains at or above the detection validation threshold. Therefore, even if the detection threshold is increased as shown in FIG. 5, by setting the detection validation threshold to the same value as the original detection threshold, it is possible to prevent a phenomenon in which the number of detections increases.

In other words, when the calculated likelihood of the current frame is equal to or greater than the detection threshold B, and the likelihood of B or more is not output in the past frame, the detection enabling threshold B2 smaller than B is set. Unless there is a frame that outputs a lower likelihood before the immediately preceding frame and thereafter there is no frame that outputs a likelihood of B or more up to the immediately preceding frame, B
The above value is corrected likelihood (eg, likelihood Lf of the current frame)
In other cases, a value less than B is derived as the modified likelihood.

FIG. 6 shows a third embodiment. The automaton shown in FIG. 6 is an automaton whose initial state is s1, and when the input likelihood is smaller than B, the state stays in the state s1 and the value less than the detection threshold value. (Lf in this example) is used as the modified likelihood and a value smaller than the detection threshold is output.

In the state s2, when the input likelihood is equal to or larger than B, the state remains. At this time, the past n is displayed in state s2.
Only when the number of frames remains equal to or longer than the frame, a value equal to or larger than the detection threshold (in this example, Lf) is output as the modified likelihood. Otherwise, a value smaller than the detection threshold is output as the modified likelihood. If the input likelihood is less than B in the state s2, the process returns to the state s1 and outputs a value less than the detection threshold as the modified likelihood.

This automaton outputs a value equal to or greater than the detection threshold only when all the likelihoods in the past n frames are equal to or greater than the detection threshold. With this automaton, erroneous detection due to an instantaneous increase in likelihood of a word that should not be detected can be reduced.

FIG. 7 is a diagram showing a fourth embodiment. The automaton shown in FIG. 7 is an automaton whose initial state is s1, stays at s1 until the n-th frame is reached, and includes the current frame. The average value of the likelihood of all the past frames is continuously output. When the frame reaches the n-th frame, the average value of the likelihood of the past n frames including the current frame is output, and the process transits to s2. Thereafter, the average value of the likelihood of the past n frames including the current frame is output for each frame.

In this way, the output modified likelihood becomes a weighted average value of the likelihoods in the past n frames including the current frame, so that the likelihood of the candidate word is smoothed, and the instantaneous word that should not be detected is instantaneous. , It is possible to reduce the probability of erroneous detection due to the increase in the likelihood.

[0026]

As described above, according to the present invention, the calculation likelihood of the current frame is modified and output according to the likelihood of the current frame and at least one past frame. Reduce the instability of the word likelihood over time,
There is an effect that a word spotting apparatus capable of performing more robust word detection can be configured.

[Brief description of the drawings]

FIG. 1 is a schematic system diagram of word spotting according to an embodiment of the present invention.

FIG. 2 is a diagram illustrating an embodiment of the automaton unit 106 of FIG.

FIG. 3 is a diagram for explaining a problem of the embodiment of FIG. 2;

FIG. 4 is a diagram illustrating another embodiment of the automaton unit 106 of FIG.

FIG. 5 is a diagram for explaining an effect of the embodiment in FIG. 4;

FIG. 6 is a diagram illustrating still another embodiment of the automaton unit 106 of FIG.

FIG. 7 is a diagram illustrating another embodiment of the automaton unit 106 of FIG.

FIG. 8 is a schematic system diagram of a conventional word spotting apparatus.

[Explanation of symbols]

 Reference Signs List 101 voice analysis unit 102 likelihood calculation unit 103 result output unit 104 phonemic dictionary 105 word dictionary 106 automaton unit

────────────────────────────────────────────────── ─── Continuation of the front page (56) References JP-A-6-110488 (JP, A) JP-A-7-281694 (JP, A) JP-A-60-202493 (JP, A) Proceedings of the Acoustical Society of Japan Collection March 1994 2-P-13 p. 183-184 Proceedings of the Acoustical Society of Japan March 1993 1-4-5 p. 9-10 Proceedings of the Acoustical Society of Japan March 1993 1-4-4. 7-8 Proceedings of the Acoustical Society of Japan March 1992 1-1-22 p. 43-44 (58) Field surveyed (Int.Cl. ⁶ , DB name) G10L 3/00 531 JICST file (JOIS)

Claims (3)

(57) [Claims] 1. A speech recognition apparatus for calculating the likelihood of a specific word for each fixed time (frame) during utterance and detecting the specific word on the basis of the calculated likelihood. degrees and viewed including the likelihood correction means for deriving and correct based on the likelihood of at least one frame likelihood and the past in the current frame, the likelihood correcting means, the current frame
Frame and the frame immediately before it both have a certain threshold or more
In the case of, the likelihood below the threshold is derived as the modified likelihood.
Otherwise, modify the likelihood of the current frame
A speech recognition device characterized in that it is derived as: 2. The method according to claim 1, wherein the likelihood of the first frame is greater than or equal to the threshold.
The speech recognition apparatus according to claim 1, wherein in such a case, the likelihood is derived as a modified likelihood . 3. A feature for every fixed time (frame) from utterance.
The likelihood of a fixed word is calculated, and the calculated likelihood is used as a reference.
A speech recognition device that detects a specific word.
The likelihood of each frame is calculated as
At least one frame is modified and derived based on the likelihood.
Means for correcting the likelihood of the current frame.
The calculation likelihood of the game is greater than or equal to a first threshold, and the first threshold
Output a likelihood that falls below a second threshold that is smaller than
Frame exists before the previous frame, and
Up to the frame having output the likelihood greater than or equal to the first threshold value.
Modify values above the first threshold unless there is a frame
The likelihood, otherwise a value less than the first threshold
A speech recognition device characterized in that it is derived as a modified likelihood .