JPS62195700A - Continuous numerical voice recognition - 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 digit speech recognition method in speech recognition.

(Prior Art) Conventionally, continuous digit speech recognition methods for recognizing and determining input speech of consecutive digits uttered continuously have been developed and put into practical use. For example, the document “Unexamined Japanese Patent Publication No. 58-189
This type of recognition method disclosed in ``No. 895'' recognizes continuous input of reading numbers by using the DP matching method and limiting the number of digits, especially regarding numbers among continuous speech recognition methods. It is.

This conventionally proposed continuous digit speech recognition method will be briefly explained with reference to FIG.

In FIG. 2, l is an input terminal for continuous digit voice.

2 is an A/D converter, 3 is a buffer, 4 is a feature extraction unit, 5
is the standard button memory, and 6 is the distance calculation section.

7 is a continuous DP processing section, 8 is a logic@arithmetic processing section, 9 is a power analysis section, 10 is a limiter, and 11 is an output terminal.

This method calculates the short-term power from the input audio at regular intervals, and then expands the time-series value of the short-time power with an orthogonal function.
Mora (syllables) that compose input speech by orthogonal function expansion
Find the number. Then, the obtained mora number plus 1 is divided by 2, the decimal part is rounded down to an integer, and the number of digits when the input voice is continuous numbers is estimated and recognition processing is performed.

(Problems to be Solved by the Invention) However, in the conventional method described above,
Since the iDP matching method is used, the scale of the matching section etc. becomes extremely large, making it unrealistic from an economic point of view. Second, since there is a limit on the number of digits, it has the advantage that the number of matchings is reduced and the recognition processing time is shortened to some extent compared to a method without a limit on the number of digits. In this case, the number of digits is limited, and the range of use becomes extremely narrow when using standard pronunciation (for example, when inputting a telephone number "Zero Niigo Ohachi"). Roughly related to this second point, when uttering consecutive numbers in E33 with various units (for example, ``yen'' for monetary amounts, rKmJ for distance units, etc.), always Since it is uttered with the place value "O million O thousand O hundred O + - O", it is used not only for regular consecutive numbers, but also for other consecutive numbers. Naturally, it is difficult to recognize using this algorithm. Therefore, when considering continuous digit recognition methods that meet the needs of users in the future, it is thought that a system that has both functions and can immediately respond to input speech will be required.

In view of the above-mentioned problems, the purpose of this invention is to detect whether the input voice has a scale or a reading of the numbers, when a number of arbitrary digits is uttered continuously, and after detection, responds to each. It is an object of the present invention to provide a continuous number recognition method that enables highly accurate recognition processing.

(Means for solving the problem) In order to achieve this purpose, in the continuous digit speech recognition method of the present invention, especially when 8-digit numbers (N is any positive integer) are continuously uttered, , an input form recognition judgment is performed by detecting whether the input voice has a scale (00,000,000,000,000) or numbers are read in a straight line.This judgment is performed by an input form recognition judgment section.

Next, in accordance with the determination result obtained in this recognition determination, recognition processing is performed according to the input form of place value or stick reading, and continuous digit speech recognition processing is performed with high precision.This processing is performed by the recognition processing section.

(Function) According to such a configuration, when a continuous number voice with an arbitrary number of digits is input, the input form recognition determining section determines whether it is a digit with a place value or a digit with a 0 place value. Recognition processing is performed in the recognition processing section, and when it is determined that the number is a stick-reading number, recognition processing is performed on the voice in the stick-reading recognition processing section.

In this way, the input form of the input voice is determined, and the speech recognition process corresponding to the determined input form is performed, so the recognition accuracy is high, and since the recognition process can be performed regardless of the number of single digits, the range of application is extremely wide.

(Embodiments) Hereinafter, embodiments of the present invention will be described with reference to the drawings.

FIG. 1 is an explanatory diagram of the consecutive number recognition method of the present invention, and is a block diagram showing one embodiment of the configuration of an apparatus for carrying out this method.

In FIG. 1, 100 is an input terminal, 200 is a frequency analysis section, and 300 is a logarithmic conversion section. 400 is a voice section determining section, 500 is a spectrum converting section, and 600 is a resampling section. 700 is an input form recognition determination unit, which includes a blocking unit 710, a distance calculation unit 720, a “place” standard slam memory 730, a form determination unit 740, and each of these units 710,
It consists of a control section 780 that instructs and controls the processing at 720 , 730 , and 740 . 900 is a scale recognition processing unit, a distance calculation unit 910, a standard button memory 9
20, a determination section 930, and a rule table 940. 1
000 is the stick reading recognition processing unit, distance calculation unit +010
．． It consists of a standard button memory 1020 and a determining section 1030, and 810 is an output terminal.

In such a configuration, continuously uttered numeric sounds are input to the input terminal 100, and the input audio signal input from the input terminal 100 is input to the single frequency analysis section 200 to generate a quantum signal corresponding to a plurality of frequency bands. The zero-frequency analyzed signal is sent to the logarithmic conversion unit 300 to obtain logarithmic spectrum information and full range power information.

The spectral information and power information are sent to the voice section determination section 400, and only the spectral information is sent to the spectral conversion section 500. The spectrum conversion unit 500 normalizes the utterance intensity and sound source characteristics by subtracting the least squares approximation straight line of the voice spectrum from the analysis data that varies depending on the speaker.

The voice section [■I determination unit 400 detects the start frame (S T F' R) and end frame (E D F R) of the continuous numeric voice signal issued by jpH. The determined start frame (STFR) and end frame (E
The audio power data of the audio section determined between DFR) is sent to the resampling unit 600 at the same time as the information sent from the spectrum converting unit 500. The resampling unit 600 normalizes the time axis of the audio power data. The method of normalizing the time axis is a conventionally known technique, and the linear matching method is a method of dividing the speech section into a fixed number of equal intervals in time determined by the conditions of the recognition device and resampling the same.

Judgment of Input Form First, the voice power data resampled by the resampling unit 600 is processed by the bronc section in the input form recognition determining unit 700. Send to 10. Input form recognition stamp? A700 is the format in which N-digit consecutive numbers (N is any positive integer) are input as audio power data, that is, "O"fO100l-OJ. This is the part that determines the type of input, such as whether it is included in a series of digits that are uttered, or whether it is a digit that is read in a straight line, such as ``ichi ni san yeon.'' This determination process will be explained below.

Figure 3 is a voice power diagram showing, as an example, the voice power when uttering 4-digit numbers with scale and in stick reading.The horizontal axis shows the time axis, and the vertical axis shows the magnitude of the voice power. It is. In the same figure, the upper figure has a scale and is ``Rokusengohyakunanajuuni'' (6,572
) and the lower figure shows the power when pronouncing [Rokugo (I) Nanani H) J (6572), respectively.

Since the input form is determined using voice power data in units of one divided block, the blocking unit 710 performs blocking by dividing the input form at a place that is likely to be a word section.

The broken lines in FIG. 3 are called pointers, and are the positions of boundaries between word sections after block formation, and the space between the pointers is the sound of word candidates such as numbers and places. Further, the arrows between the voice power of the scale digit and the voice power of the scale digit indicate the position relationship between the respective pointers.

Then, the distance calculation unit 710 calculates the sound bang (!1
1 word candidate voice button) and “place” standard button memory 7
The distance RF+ calculation is performed with the "place" e quasi-button stored in 30. In this case, the "place" standard button memory 730 contains "
Standard stamps for ``place'' such as φ・100 million, 10,000, 1,000, 111, etc. are stored.

Also, depending on the place, for example, the number that comes before the place, such as the ``hundred'' place, the way ``hyaku'' is pronounced changes like ``byaku'' or ``pyaku'', so such modified bangs are also used. Stored.

In order to match these standard bangs and word candidate voice bangs, first, the blocking unit 710 extracts the end frame (EDFR) of blocks of voice power data delimited by pointers with small power values. To detect. By detecting this end frame, the end of one continuous utterance can be confirmed.

Next, the pointer is searched from the point in time when this end frame is detected, going back in the direction of the start end frame (STFR). This search starts from the end frame side and then all the blocks' audio bangs, that is, between all the pointers! This is done by matching the phonetic bang of the li word candidate with the ``place'' standard bang.

Then, the minimum value of the sum of each alternating layer obtained by distance calculation with the standard baton and the voice power data of the voice section for which the distance was calculated are output to the form determining section 740, where this total distance #minimum value Based on the value of , it is determined whether the input consecutive numbers have a scale or are in stick reading.

In this way, in this method, the starting frame (STFR)
The distance from the "place" standard button is calculated for all the blocks of the pointer 11H from to the end frame (EDFR).

However, ``I'' is the first bronc or second block counting from the end frame (EDF R) which is required when uttering the ``Ichi'' (for example, ni → jiyu → nana → Focusing on the fact that hiichi < ---) comes, it is possible to distinguish between scale digits and stick digits by checking only the first digit from the end frame side, so there is a method to further reduce the matching time. It is also conceivable to match only the first and second blocks counted from the end frame (EDFR) of the block with the "place" standard button and output the determination result.

Incidentally, the control section 760 performs, for example, the detection of the end frame, the search for the pointer, the instruction and control of specifying blocks to be matched to shorten the matching time, etc. in the input form recognition and determination section 700.

In this judgment unit 740, if the judgment result is a scale number, it sets a flag (FLAG) = 1, adds this flag -1 to the voice power data of the voice section that is sent, and also determines whether the judgment result is a scale figure or not. If it is a flag (FLA
G) A flag -O is added to the voice power data sent as =O, and the result is sent to the recognition processing section 800 at the next stage.

Recognition Processing In the recognition processing unit 800, the voice power data for which the flag is 1, that is, it is determined that the input voice is scaled, is sent to the scale recognition processing unit 900.

The distance calculation unit 910 performs distance calculation with the standard slam memo 1J920. This distance @ calculation will be explained below.

First, FIG. 4 shows the contents of the standard button memory 920. The content shown is an example of the standard button required to correspond to the transformation of the reading according to the combination of each digit and number. In addition to the standard button for ``place'' in the standard button memory 730, it also has the standard button for each number from 1 to 9 as shown in FIG. Furthermore, as is clear from Figure 4, ``kuchi ()'' only exists in the hundreds place, and ``hakufu'' only exists in the thousand or hundred places, etc. There is a distinctive type of bang that is different from the number sound type of bang. Furthermore, in the case of continuous numbers, in addition to the characteristic bang mentioned above, as shown in Figure 5,
There are certain specific rules.

FIG. 5 shows an example of the relationship between numbers and scale readings from the thousands digit to the - digit. In this figure, the readings of the numbers are shown in Roman letters. In addition, the place within the loco is the place, and the reading of the transformed place is prescribed. Also, the solid line connected to 0 is the number connected to the reading of each digit. For example, the thousands place is “SEN”.
There is a way to read it as ``ZEN'', but the number in the thousandth place is read as ``ZEN'', and the number is r3'' (SAN)L.

Or it doesn't exist. In addition, the hundreds digit is further classified, and it can be seen that there are some special types of number bangs themselves.

The rule table 940 in the scale recognition processing section 300 stores the feature rules for reading continuous numbers with scale shown in FIG. Therefore, the distance calculation unit 910 matches the word candidates already blocked by the blocking unit 710, sends the distance calculation result to the determination unit 930, and sends the distance calculation result to the determination unit 930.
30 refers to the rules stored in the rule table 940 during calculation and separates digits and numbers to shorten distance calculation time and judgment time and improve recognition accuracy.

As described above, the total distance is determined using only the limited combinations, and the category with the smallest distance ra value is output to the output terminal 810 as the recognition result.

-10,000, flag 20, that is, the voice power data determined that the input voice is Haiyomi is sent to the Boyomi recognition processing section 1000.

The distance calculation unit 1010 performs distance calculations for each numerical word candidate that has been converted into a block, one digit at a time, with respect to the standard button stored in the standard button memory 1020, and sends the result to the determination unit 1030.

The standard button memory 1020 differs from the standard button memory 920 in that only a limited number of button readings corresponding to the number readings are stored. The determination unit 1030 also outputs the total distance and the category name to the output terminal 810 as a recognition result. Further, the detailed operations of the distance i calculation section 1010 and the determination section 1030 of the stick reading recognition processing section 1000 are described, for example, in the r-voice slamming method proposed in Japanese Patent Application No. 58440/1983 filed by the applicant of the present invention. ', and by limiting the slope of the voice stationary part bus, the transient part bus is limited to the range where the rate of speech is expected.

In addition, in response to this, a path setting process tentatively called optimization is repeated a plurality of times.

As can be understood from the above explanation, according to the method of the present invention, when a continuous number voice is input, after determining whether it is a number with a scale or a number with a scale,
It is sufficient to have a configuration that performs the corresponding voice recognition processing, and the specific configuration is not limited to the above-described embodiment.

In addition, the main components 400 and 5 shown in FIG.
00, Boo,? 10, 720, 740,
The processing of 780.910, 930, 1010, and 1030 can be performed by a CPU (central processing unit).

(Effects of the Invention) As is clear from the above explanation, according to the present invention, when arbitrary numbers are uttered continuously, it is detected whether the input voice has a scale or the number is read in a rounded manner, and after the detection, it is possible to Since the corresponding recognition processing is performed, it is possible to perform the recognition processing of the uttered continuous numeric voice with high accuracy, thus improving the recognition rate.
The effect of reducing the number of calculations to N can be expected. Furthermore, since it has the functions of continuous number recognition methods for both stick reading and placekeeping, it can be applied to continuous number recognition devices that meet the needs of users in the future, and has a wide range of applications.

[Brief explanation of drawings]

Figure 1 is a block diagram for explaining an embodiment of the continuous digit speech recognition method of the present invention, Figure 2 is a block diagram for explaining the conventional continuous digit speech recognition method, and Figure 3 is a block diagram for explaining the conventional continuous digit speech recognition method. A voice power diagram when uttering a 4-digit number, Figure 4 is a diagram showing the contents of the title stamp memory, and Figure 5 is a diagram showing the relationship (rules) between numbers and scale. 100... Input terminal, 200... Frequency analysis section 300... Logarithmic conversion section, 400... Speech interval foot section 500... Spectrum conversion section 600... Re-sampling section 700... Input Form recognition determination unit 710... Broncization unit, 720... Distance calculation unit 730... "place" standard slam memory 740... Form determination unit, 780... Control unit 80
0... Recognition processing unit, 810... Output terminal 800
... Scale recognition processing section 810 ... Distance lrT calculation section 920 ... Standard button memory 930 ... Judgment section, 940 ... Rule table 1000 ... +5 reading recognition processing section 1010 ... Alternate layer calculation Section 1020... Standard button memory 1030... Judgment section. Patent applicant Oki Electric Industry Co., Ltd. 1., I'll
′′ , ',. Agent: Patent attorney Takashi Ogaki, '-. zo1 鑓ride0stopkukei''S3□?Recognition method 9emeiguchi Figure 2

Claims (1)

[Claims] (1) When recognizing the input voice of continuous numbers of any number of digits that are continuously uttered, it is determined whether the input format of the continuous numbers in the input voice is continuous numbers with place value or continuous numbers with bar reading. A continuous number speech recognition method, characterized in that a scale recognition process or a reading recognition process is performed on the input voice according to a result of an input form recognition determination.