JPS6239900A - 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 speech recognition device, and more particularly to a speech recognition device that reduces the reading time of a speech standard pattern and the speech matching processing time.

[Prior Art J] This type of conventional speech recognition device sequentially superimposes the input speech and all speech standard patterns registered in advance, calculates the distance between the two each time, and selects the one with the minimum distance. The patterns were extracted and used as speech recognition results. Therefore, as the number of recognizable words increases, the number of registered words increases, the time required for recognition increases significantly, and the recognition rate decreases.

In order to avoid this, for example, standard spoken words can be classified and registered according to concepts such as words, numbers, monosyllables, etc., and when recognition is performed, it is possible to select the memory section in which the word group is stored, and then A method that performs matching processing is considered effective.

There are two ways to select and change the word group memory, such as by key operation or by voice.

[Problems to be solved by the invention] However, although the key operation allows the selection and change of the word group memory section to be performed reliably, the operation becomes complicated as manual key input and voice input must be performed at the same time. This places a heavy burden on people.

In addition, in addition to the original registered voice pattern group, the voice type requires a command to select and change the memory of the registered voice pattern group, and for this purpose, a name representative of each word group memory is required. A separate storage unit must be provided for the modified vocabulary group consisting of voice patterns.

In other words, the original speech pattern is divided into several word group memory sections depending on the characteristics of the word, and each one is given a change command such as "Henkora J." When selecting or changing the word group memory section, , First, when you input "Henkora J" by voice, it is recognized in the word group memory that was selected at that time, and the memory of the change term group is selected.Next, if you wish to change the selected word, Selection/change is possible by inputting the name of the word group memory unit by voice.However, this method does not allow selection/change.
It is time-consuming because it requires voice input twice to make changes.

Furthermore, since the voice patterns of the above-mentioned change commands are registered for each word group storage unit, voice patterns of the same word but with different voice volume levels and voice time lengths are registered. Therefore, even when the same selection or change is made, the recognition is different, and in the worst case, it may become impossible to select a changed word group from a certain word group storage unit.

[Means for Solving the Problems] The present invention has been made for the purpose of solving the above-mentioned drawbacks of the prior art. A voice pattern storage means for storing a plurality of standard voice patterns, a voice input means for inputting voice information, a voice length detection means for detecting a voice length of voice input from the voice input means, and a voice length detection means for detecting the voice length of the voice input from the voice input means. voice pattern reading means for reading out a corresponding standard voice pattern from the voice pattern storage means according to the utterance length detected by the detection means; and voice recognition means for recognizing the voice.

[Function] In this configuration, by incorporating the utterance time information of the input speech into the speech recognition method, it is possible to read and write to the speech pattern storage means at high speed, shorten the matching processing time for speech recognition, and achieve a high recognition rate. can get.

[Embodiments] Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings.

FIG. 1 is a block diagram of a speech recognition device according to an embodiment of the present invention.

In the figure, 1 is a microphone that converts audio into electrical signals, and 2 is a 8-30 channel with a frequency range of 200-6000Hz.
3 is an A/D that samples and quantizes the extracted features every 5 to 10 mS. It is a converter. Further, 4 and 14 are registration/recognition changeover switch means for switching the signal path between standard voice registration and input voice recognition, and 5 and 12 are for calculating the utterance duration of input voice at the time of registration or recognition. 6 is a buffer memory for storing the input voice feature amount until the input voice, and 6 is a start/Φ end detection circuit that detects points corresponding to the start and end of a word syllable from the power signal of the input voice.

7 is a utterance length measurement circuit that measures the time from the start to the end of the input voice based on the detection point information of the start/end detection circuit 6;
9 is a utterance length selection circuit that generates a selection signal for various quantity group storage units 101 to Ion in accordance with the utterance time length detected and measured by the utterance length measurement circuit 7; is stored in the collective storage unit 10 during voice registration.
A switch 11 switches between 1 and 1on, and 11 is a switch that switches between the vocabulary group storage units 101 and 101 during speech recognition. Reference numeral 13 denotes a pattern matching unit that compares the input speech pattern with the registered speech pattern read out by switching the vocabulary group storage units 101 to Ion during speech recognition, and 15 a general-purpose central processing unit that processes the recognition results. (CPU), 16 is an operation keyboard, 17 is a display section for displaying recognition results, etc., and 18 is a card writer for storing the voice standard pattern stored in the memory 10 on a recording card. Reference numeral 19 denotes a card reader that loads into the memory 10 the voice standard pattern registered in the recording card previously stored by the card writer 18 when the device is used.

Note that in this embodiment, a magnetic card is used as the recording card. Therefore, it is smaller and easier to handle than magnetic flexible disk devices, etc.
It is extremely easy to use. In addition, optical card, I
Needless to say, it may be a C card.

The operation of this embodiment having the above configuration will be explained in detail below.

First, the utterance time length of the input voice input from the microphone 1 is determined by the time difference between the start and end of the input voice. Various methods can be considered for detecting the start and end points of audio, but in this embodiment, power information P after A/D conversion by the A/D converter 3 is used.

Figure 2 shows the power information P of the input audio output from the A/D converter 3 every 5 to 1.0 mS, and the vertical axis represents the power amount P.
, in which the horizontal axis is expressed as a time axis.

In FIG. 2, first, in order to remove background noise mixed into the input audio, the average value of noise power in the laboratory is calculated in advance, and this is set as the threshold value PN.

Further, a threshold value for the level of a word-initial consonant that is easily devoiced or a word-initial consonant with low power is set as Pc, and the average value of the two thresholds PN and Pc is set as PM. Also, let Tp be the minimum pause time from when there is an input voice until the next input voice, and let Tw be the minimum utterance time recognized as input voice.

[Detection of Start Point SO] First, the first point where the power signal P output from the A/D converter 3 every 5 to 10 m5 becomes 222M is found. After this point, if the state of 222M continues for more than Tw time, the first point that becomes 222M is set as the starting point So. T
If it ends in less than w time, it is considered as noise,
Find the next point 222M and perform the same operation as above.

[Detection of end EO] Find the first point where the power signal P satisfies P < P s after the start end So is detected. After this point P < P M
If the state continues after Tp time, then P < P
Let the first point that becomes M be the terminal point Eo.

In this way, the start and end of the input audio are detected.

The utterance length measurement circuit 7 starts a timer when the start edge SO is detected by the start edge/end edge detection circuit 6, and stops the timer when the end edge Eo is detected, calculates the utterance time length, and uses that value as the utterance length. It is sent to the sorting circuit 8.

Incidentally, the above-mentioned operation can be realized by a microprocessor incorporating the control program shown in FIG.

The details of the utterance duration detection control will be explained below according to the flowchart 1 in FIG.

First, in step S1, a timer t is initialized to °O".

Then, in the following step S2, it waits for the power signal P to go from PM to -L. If 222M is satisfied in the determination in step S2, the process proceeds to step S3, and the contents of timer t at that time are saved in the start register So. Then, in steps S4 and S5, it waits until the state of 222M continues for Tw time or more. If 222M is not satisfied halfway through, the process returns to step S1, and the portion up to that point is treated as noise.

222M is longer than Tw time" - If it continues, the process advances to step S6, the contents of the start register So are ensured, and further P < P
Wait until it becomes M. Then, in the determination in step S6, P<
When PM is satisfied, the process proceeds to step S7, and the contents of timer t at that time are saved in the rear end register Eo. Then, in subsequent steps S8 and S9, it is waited for the state of PM to continue for a period of time TP or more. If P<PM is not satisfied during the process, the process returns to step S6, the power signal P up to that point is valid, and the input audio is treated as being continuous.

If the process continues for more than TP time, the process proceeds to step SIO, where it is determined that the input audio is at the end, and the contents of the trailing end register Eo are finalized. Then, in step S10, the section from time So to Eo is determined as the utterance length V1. Through the above processing, the utterance length of the input voice is measured.

Next, the storage structure of the memory 10 that stores standard voice patterns and the like will be explained. A specific example of the storage configuration of the memory 10 employed in this embodiment is shown in the table.

The memory 10 is a collective memory unit 1 for various quantities classified into utterance duration length side.
01 to Ion. Then, as shown in the table, words having a utterance time length of 0.4S to 3S are used as the utterance time length of the word to be recognized, and each quantity group storage unit 10. ~l
On stores the various quantities corresponding to each increase in utterance time length by 0.2S from 0.43.

When registering the table standard voice, each contact C of the switch 4.14 is connected to the contact 41 side and the contact 141 side, respectively, as shown in FIG. The audio signal to be registered inputted from the microphone 1 is set in the buffer memory 5 via the feature extractor 2 and the A/D converter 3 under the same control as described above. At the same time, the output from the A/D converter 3 is also sent to the start/end detection circuit 6, and the output is input to the utterance length detection circuit 7. Then, when the utterance length v1 of the input voice detected by the utterance length detection circuit 7 is sent to the utterance length selection circuit 8, it is converted into a selection signal for the various quantity group storage sections 101 to 10n classified by utterance length. The selection signal is applied to the contact 14 of the switch 14.
1 to the terminology group storage unit changeover switch 9 for registration, and the corresponding quantity group storage unit is selected there. The word aggregation memory unit selected in this way is stored in the voice characteristic pattern (for example, from So to E) stored in the buffer memory 5.
(up to o) is stored as a standard pattern. In this way, voice patterns of various utterance lengths are stored in the word group memory section determined for each utterance length.

The voice standard patterns registered by each individual in this way are sent to the cart writer 18 and stored on the recording card.

When the device is used next time, the recording card on which the individual's standard voice pattern is recorded is directly loaded from the card reader 19 into each word group storage section of the memory 10, thereby saving the trouble of registering the voice standard pattern.

Further, during voice recognition, the contacts C of the switches 4 and 14 in FIG. 1 are connected to the contact 42 side and the contact 142 side, respectively. Therefore, the output from A/D converter 3 is set in buffer memory 12. Further, the selection signal outputted from the utterance length selection circuit 8 is sent to the miscellaneous cost group storage section changeover switch 11 for recognition via the contact 142 of the switch 14, and the word millet group storage section corresponding to the detected utterance length v1 is selected. be done. Next, the selected standard patterns in the word collection storage section are sent one by one to the pattern matching section 13, and the buffer memory 12
Pattern matching is performed between the standard pattern and the characteristic pattern of the input voice stored in , the standard pattern with the maximum similarity between the two is extracted, and its corresponding code is output to the CPU 15 as a recognition result.

Next, the above-mentioned operation will be explained according to a specific example.

First, at the time of registration, for example, a certain word A is input, its voice feature amount is stored in the buffer memory 5, and the utterance time length is determined to be 0.85 by the start/end detection circuit 6 and the utterance length measurement circuit 7.
Suppose that S is calculated. The utterance length selection circuit 8 selects the word group storage unit 103 based on the time information 0.853 according to the table, and the characteristic pattern of word A in the Yatsufa memory 5 is registered in the storage unit 103. In a similar operation, the storage section 103 is selected by the switch 11, and the plurality of standard patterns in the storage section 103 and the characteristic pattern of word A stored in the buffer memory 12 are matched in order.

By the way, even if the word is the same, if the utterance time at the time of standard pattern registration and the utterance time at the time of voice recognition are different, the desired quantity group storage unit may not be selected at the time of voice recognition.

For example, in the table, the utterance length at the time of registration of word B is 0.79
53, and the utterance length at the time of recognition is 0.8S, word B is registered in the storage unit 102, and recognition matching is performed with the standard pattern in the storage unit 103, so that the word B is will no longer be recognized. In this embodiment, in order to avoid problems caused by variations in utterance length, the various quantity collective storage units are selected using utterance time information that takes into account a predetermined variation range in the true utterance length for shortening the recognition time. For example, word B
If we consider the fluctuation range of ±o and ois and add it to the true utterance length Q, 8S during recognition, the utterance length of word B will be 0.7H~
It can be 0.801S. Since this value spans the storage unit 102 and the storage unit 10a, matching is first performed using the standard pattern in the storage unit 102, and then matching is performed using the standard pattern in the storage unit 103.

On the other hand, the utterance length at the time of registration for word C is 1.053,
In the case where the true utterance length at the time of recognition is 1.1O3, the utterance length at the time of recognition considering the fluctuation value of ±0.013 is also within the same word group in the storage unit 104 as at the time of registration. All you have to do is perform matching. As described above, according to this embodiment, it is possible to provide a speech recognition device that is resistant to variations in utterance length.

Comparing the results of recognizing 500 words using the Kinomi stream side device with those using the conventional method, the recognition processing time is 10%.
The recognition rate was improved by more than 20%, with an average recognition processing time of 280 m5 and a recognition rate of 98.5%.

In this embodiment, the PN was added using background noise in the laboratory, but the value of PN can be freely varied in any noise environment depending on the purpose of the speech recognition device. Also, the number of classifications in the word search group memory,
The capacity of each storage section, the length of utterance time, the variation range of utterance time taken into account during recognition, etc. can be freely varied depending on the application so as to always obtain the best recognition results.

Furthermore, by applying this embodiment to a typewriter,
A high-speed, highly reliable voice typewriter can be constructed.

In the above explanation, an example has been described in which a magnetic recording card is used as the recording card, and a magnetic card writer and a magnetic card reader are used as the card writer 18 and the card writer 19. A RAM) pack may be used to store the registered standard voice pattern information in the memory 10. By doing so, it is possible to make the device compact and require almost no time for reading and writing.

Furthermore, it goes without saying that a magnetic bubble card capable of storing a large amount of memory may be used, or an optical card may be used.

As described above, according to the present embodiment, it is possible to provide a speech recognition device that achieves shortening of the recognition time of speech recognition processing by adding utterance time information to the speech feature amount. That is,
Since pattern matching candidates during speech recognition are narrowed down to small groups based on utterance time information, the recognition processing time is shortened even when the total number of registered words is large.In addition, utterance time information is also important for speech recognition. By using this information in recognition processing, it is effective to improve the recognition rate.

Furthermore, by storing the standard voice patterns on a recording card or the like, it is smaller and simpler, and easier to maintain and manage than when storing them on a floppy disk or the like. Therefore, the standard speech patterns can be easily saved, and by saving the standard speech patterns for each individual, many people can use one recognition device. Furthermore, reading out the audio standard pattern can be performed easily and at high speed.

As described above, according to this embodiment, it is possible to obtain a device that is easy to handle and has a high recognition rate, and it is possible to achieve particularly great effects by widely applying the wood flow side device to consumer use.

[Effects of the Invention] As explained above, according to the present invention, by performing speech recognition according to the utterance time length information of input speech, in a short time,
In addition, speech recognition with a high recognition rate can be performed.

Furthermore, by recording the voice standard pattern information on a magnetic card, an IC card, etc. and making it readable, it is possible to provide a voice recognition device such as a voice typewriter, word processor, etc. that is easy to maintain and manage and is small.

[Brief explanation of the drawing]

Fig. 1 is a block diagram of a speech recognition device according to an embodiment of the present invention, Fig. 2 is a diagram showing the power information P of input speech of this embodiment on a time axis, and Fig. 3 is a diagram of a speech recognition device according to an embodiment of the present invention. 3 is a flowchart showing a utterance length measurement circuit. Here, 1...microphone, 2...feature extraction unit,
3... A/D converter, 4... Registration/recognition changeover switch means, 5... Registration buffer memory, 6... Start/end detection circuit, 7... Utterance length measurement circuit, 8 . . . Utterance length selection circuit, 9 . . . Registered term millet group changeover switch means, 10 .
... Recognition buffer memory, 13... Pattern matching unit, 14... Registration/recognition changeover switch means, 1
5...CPU, 16...Keyboard, 17...Display section, 18...Card writer, 19...Card league.

Claims (5)

[Claims] (1) A voice pattern storage device that stores a plurality of standard voice patterns grouped according to utterance length, a voice input device that inputs voice information, and detects the utterance length of the voice input from the voice input device. utterance length detection means; voice pattern reading means for reading out a corresponding standard voice pattern from the voice pattern storage means according to the utterance length detected by the utterance length detection means; and a voice standard pattern read out by the voice pattern reading means; 1. A speech recognition device comprising: speech recognition means for sequentially comparing input speech patterns and recognizing speech. (2) The voice pattern reading means determines a corresponding group of voice patterns to be read based on the utterance length detected by the utterance length detection means and the value obtained by adding a predetermined fluctuation range to the utterance length,
The speech recognition device according to claim 1, characterized in that the speech standard pattern for each group is read out. (3) The speech recognition device according to claim 1 or 2, characterized in that a magnetic card is used as the speech pattern storage means. (4) The speech recognition device according to claim 1 or 2, characterized in that an IC card is used as the speech pattern storage means. (5) The speech recognition device according to claim 1 or 2, characterized in that an optical card is used as the speech pattern storage means.