JPS6312295B2 - - Google Patents

Description

[Detailed description of the invention]

The present invention relates to a voice input Japanese typewriter using a voice recognition device. In conventional Japanese typewriters, when creating a general document, an operator selects a character from a table in which characters such as kanji, kana characters, alphanumeric characters, symbols, etc. are all arranged in parallel, and then presses the key. With this method, there are thousands of types of characters on the table, so Japanese typewriters are not devices that anyone can use immediately like English typewriters, and require the training of specialized operators. In response to this, a Japanese typewriter has recently been proposed that uses a keyboard with only kana, alphanumeric, and symbols to input kana characters according to the pronunciation of the characters in the input sentence, and has the function of converting kana to kanji. There is. Various Kana-Kanji conversion methods have been proposed, ranging from a simple dictionary lookup machine that displays the kanji that correspond to the kana in the range specified by the operator, to a machine that simply inputs the text as it is and converts it into a text. Some also perform automatic grammatical analysis, semantic analysis, etc., and automatically convert kana to kanji. This automatic kana-kanji conversion device has a word dictionary and a grammar dictionary that correspond to the Japanese language dictionary. Japanese typewriters equipped with this automatic kana-kanji conversion function require much less learning to input than conventional Japanese typewriters, and anyone can input. However, in this case as well, in order to input general documents, 140 to 150 input keys are required when inputting kana characters, kana lowercase characters, alphanumeric characters, symbols, etc. After all, a certain degree of proficiency is required. Therefore, a voice input Japanese typewriter using a voice recognition device in the input section may be considered as a device that does not compensate for these input disadvantages and allows anyone to obtain a certain level of input speed without requiring any level of proficiency. That is, a voice recognition device is used in place of the kana keyboard of the Japanese/Japanese typewriter having the automatic kana/kanji conversion function. Various speech recognition devices have been proposed, but speech recognition is classified into word recognition and monosyllable recognition depending on the target. Among these, when considering the application of monosyllable recognition to voice typewriters, since the target is monosyllables, in principle, an infinite number of languages can be recognized. Regarding the monosyllable recognition device, see the Acoustical Society of Japan Speech Study Group Material No. 877-46 (December 1977 issue)
``Monosyllabic voice typewriter (Hokkaido University Applied Electrical Research Institute)'', as shown schematically in Figure 1. In Figure 1, 1 is a microphone, 2
is a preamplifier, 3 is a 16-channel bandpass filter (hereinafter referred to as 16CH BPF), and 4 is a 16-channel analog-to-digital converter (hereinafter referred to as 16CH BPF).
5 is a read-only memory that logarithmically converts the input value (hereinafter referred to as logarithmic conversion).
6 is a pitch extractor, 7 is a counter timer, 8 is a microprocessor (hereinafter referred to as micro CPU), 9 is a bus line (hereinafter referred to as BUS).
10 is a read-only memory (hereinafter referred to as ROM), 11 is a random access memory (hereinafter referred to as RAM), 12 is a flop disk drive device (hereinafter referred to as FDD), 13 is a character Display device (hereinafter referred to as CRT),
14 is a keyboard device (hereinafter referred to as KB), 15
is a hardware arithmetic unit. The voice typewriter having the above configuration operates as follows. First, since the audio input from the microphone 1 is attenuated at 6 dB/octave, the preamplifier 2 emphasizes the high range at a rate of 6 dB/octave to flatten this power spectrum. The output is input to 16CH BPF3, the frequency of 200 to 4400Hz is divided into 16 channels, and the output is sent to A/D converter 4 for each CH.
Enter. After being converted into a 12-bit digital quantity by the A/D converter 4, it is input to the logarithmic conversion ROM 5, where it is converted into an 8-bit logarithm. This process is done in order to approximate the logarithmic nature of human auditory characteristics, and also takes into consideration the advantage of being able to perform calculations using only addition and subtraction when normalizing the power spectrum. The output of the A/D converter 4 is drawn into the micro CPU 8 through the BUS LINE 9 of the micro CPU 8. The micro CPU 8 has a ROM 10 for control and a RAM for a data area.
11, FDD 12 for storing data such as registered audio,
KB14 for manually inputting to CRT13 and micro CPU8 for outputting and displaying voice recognition results
are connected to form a small electronic computer system. Further, the output of the 16CH BPF 3 is input to the pitch extractor 6, and the configuration is configured to trigger the 16CH A/D converter every pitch period. One method for the pitch extractor 6 is proposed in, for example, Japanese Patent Application Laid-Open No. 54-162405 entitled "Pitch Frequency Extraction Apparatus", and it may be considered that pitch frequencies are extracted by such a method. A pitch frequency exists in the case of a vowel or a voiced consonant, but in the case of a voiceless consonant, the sound source is noise and the pitch frequency does not exist. In order to sample these unvoiced consonants, a counter timer 7 generates pulses every 1 msec at a constant frequency, for example, 1 KHz, and inputs them to the 16CH A/D converter 4.
The 16CH A/D converter 4 samples the output of the 16CH BPF 3 every pitch cycle in the case of a vowel or voiced consonant, and every fixed cycle in the case of a voiceless consonant, and converts it from analog to digital to convert the power of the input speech waveform. Spectral envelope characteristics can be obtained. The digitized input voice waveform obtained in the above manner is subjected to power spectrum normalization, and then the square distance is calculated between it and each data of the registered voice waveform. Calculation of this squared distance requires several thousand times, and since real-time processing cannot be performed using the micro CPU 8, a hardware calculator 15 is used to calculate the squared distance.
The calculation is performed using hardware. The squared distance is calculated between the input speech waveform and the registered speech waveform, and the monosyllable that shows the lowest value of the squared distance is recognized as the input speech. When the recognized monosyllables are sequentially output and grouped into words or phrases, they are input to the automatic kana-kanji conversion device and converted into sentences containing kanji and kana. For this kana-kanji conversion, "Research and Practical Report Vol. 26, No. 1 (January 1977) published by Nippon Telegraph and Telephone Public Corporation Telecommunications Research Institute, P.353-P.369 Prototype of online kana-kanji conversion system (Kimura et al.) ” may be used. FIG. 2 shows an example of the automatic kana-kanji conversion processing procedure. Briefly explaining the processing procedure shown in FIG. 2, when a kana sentence separated into clauses is input, the input clauses are processed by the longest match method. The longest match method is a method in which an independent word with the longest word length is detected in a bunsetsu unit by comparison with a dictionary headword, and the input bunsetsu is separated into an independent part and an attached part. If the adjunct separated by this process can be further separated, it is separated, and then grammar matching processing is performed to test whether the connection between the separated independent word and the adjunct or between adjuncts is possible. If the separation is appropriate, the process moves on to the next process, and if it is inappropriate, the process returns to the longest match process, detects the independent word with the next longer word length, and performs the same process. If the grammar matching process determines that it is appropriate, the separation of the input phrase into independent words and dependent words is completed. Next, if the separated independent words have homophones, the semantic information of the homophones that are candidates for separation selection is retrieved from the dictionary, and the semantic information of the homophones that are the most connected between the preceding and following clauses is extracted. Perform semantic processing to select the larger one. In addition, if the candidate homophones cannot be narrowed down to one by this process alone, the frequency information of the homophones that are also candidates is extracted and the most frequent one is used as the final independent word for the bunsetsu. It performs output processing. If all of the above processes are non-conforming after performing longest match processing and grammar matching processing, there is a possibility that a prefix exists at the beginning of the clause, so prefix processing will remove the part that seems to be the prefix of the clause. All you have to do is remove it and perform the longest match process again. Here, taking "Hatsusei Niyori" and "Okonatsuta" as examples as specific inputs, processing procedures and processing results such as longest match processing, grammar matching processing, and prefix processing are shown in Tables 1 and 2, respectively.

【table】

【table】

【table】

[Table] As shown above, by connecting a monosyllabic speech recognition device and an automatic kana-kanji conversion device, a voice input Japanese typewriter converts words or phrases input by voice into sentences containing kanji and kana and outputs them. Although it is possible in principle to create one, it has the following serious drawbacks. That is, the kana characters in Japanese are phonetic characters, each of which has one independent utterance. Therefore, in general, the sentence or word you want to input can be input by converting it into kana characters and pronouncing the kana literally, but there are some sounds that are produced by stopping the vocal cords for one syllable, such as "hatsusei" and "okonatsuta". There is. In this case, it is the consonant 'tsu'. This consonant ``tsu'' has no independent vocalization and is actually pronounced silently. In the case of conventional automatic kana-kanji conversion devices, this was handled by inputting the consonant "tsu" symbol from the keyboard, but when considering countermeasures for the voice input method, the first thing to do is to input the consonant "tsu" symbol and the consonant. It may be possible to improve the monosyllable recognition device so that it can distinguish between sounds without "tsu", but this is not a good idea since the amount of reference data increases, the memory capacity to store it increases, the cost goes up, and the recognition processing time increases. Alternatively, it is also possible to assign a special pronunciation sound to the consonant "tsu". For example, if you create a single continuous sound called ``komoji'' and assign it, it can be treated the same as input from the keyboard. However, in such cases, the above-mentioned "hatsusei" or "okonatsuta"
etc., it is necessary to pronounce it as "ha, komoji, se, i" or "o, ko, na, komoji, ta", etc., which creates a sense of incongruity between the character string and the pronunciation. This is not a good idea because it makes it difficult to input and loses the advantage of voice input. Another method is to omit the pronunciation of the consonant "tsu" and input it. Said “Hatsusei”
and ``Okonatsuta'' become ``Hasei'' and ``Okonata.'' In this case, since part of the input information is omitted, there is naturally a risk that errors will increase due to the decrease in input information. In other words, ``Hatsusei'' becomes ``Hasei,''``Itsushiki'' becomes ``Ishiki,'' and ``Datsukai'' becomes ``Dakai,'' which is not a good idea because it becomes difficult to distinguish them from other words. Another possible method is to input the part of the consonant "tsu" from the keyboard. Taking "Hatusei" as an example, "Ha", "Se", and "I" are entered by voice, and the consonant "tsu" is entered by keyboard, which temporarily interrupts the flow of operation, which is also a problem with the consonant "tsu". This cannot be said to be an improvement measure. The present invention compensates for the shortcomings of these improvement measures, and makes it possible to input a given character string with a pronunciation that is as close to natural as possible without increasing the number of sounds recognized by the speech recognition unit. This provides a kana-kanji conversion method for voice input. In other words, in the utterance during input, if there is a consonant ``tsu'' in the character string, it will be uttered by replacing it with the character ``tsu'', which appears in the headword of the independent word dictionary used for automatic kana-kanji conversion processing. All the consonant sounds "tsu" have been converted to the direct sound "tsu", but other matters such as spelling, part of speech, etc. have been left unchanged. Automatic kana-kanji conversion processing is performed using an independent word dictionary. It is something to do. Therefore, the conversion is the same in both utterances and headwords. For example, "Hatuseiniyori"
becomes "Hatuseiniyori" and "Okonatsuta" becomes "Okonatsuta". This converted utterance is different from consonant utterance, but when it comes to monosyllabic utterances, there is almost no discomfort and it is close to natural. Also, the order in which the headwords are arranged by converting the headwords in the dictionary varies depending on the dictionary used, but for example, in the case of Gakken Japanese Language Dictionary by Gakushu Kenkyusha,
As for the headword arrangement for the consonant "tsu", the consonant "tsu" is placed after the direct consonant "tsu". For example, the order is ``hatsuka'' and ``hatsuka''. In a dictionary with such an arrangement, the arrangement will remain unchanged during character conversion by sentence invention. An embodiment of the present invention will be described below. FIG. 3 shows an example of a general independent word dictionary, and FIG. 4 shows an example of an independent word dictionary used in the present invention. Dictionaries used for automatic kana-kanji conversion have some kind of coded data stored in various storage devices, but JIS-6220 codes are used as dictionary entry characters due to capacity and simplicity. As a notation code, it can represent kanji.
It is common to use the JIS-6226 code.
Note that the codes and formats of other information such as part of speech and frequency information differ depending on each system. In the independent word dictionary created in this way shown in Figure 3, the consonant "tsu" on the kana heading is
The independent word dictionary in Figure 4 is the one converted to the direct sound ``tsu''. In the examples in this list, the relevant headings are marked with *, but the word arrangement has not changed at all. Furthermore, when the present invention is implemented, all the consonants "tsu" on the headings in the affix dictionary, adjunct dictionary, and inflection table are also converted. In this case as well, there is a notation section separate from the kana headings, similar to the independent word dictionary, and the notation section is JIS-
6226 code. Also, the consonant "tsu" on the notation shall be preserved as is. Table 3 shows the processing procedure and processing results when the character strings ``by vocalization'' and ``gyonatsuta'' are input using the dictionary in which the present invention is implemented in this way.
It is shown in Table 4.

【table】

【table】

【table】

[Table] First, the vocal input for "by vocalization" is "hatsusei niyori", which can be uttered without feeling too strange to the original character string. Next, a verification process is performed using this as input. However, even though the dictionary heading is converted according to the utterance, the notation part is not converted at all, so if there are no errors in the verification and decomposition, the final output will be the desired ``according to the utterance''. As is clear from the above explanation, the same result can be obtained with ``Gyonatsuta''. According to the present invention, the consonant "tsu", which does not have its own pronunciation in writing, is replaced with the direct sound "tsu", which is done by inputting vocalizations that do not seem too strange to the original character string. The automatic kana-kanji conversion processing unit also allows conventional key input, etc.
It becomes possible to perform processing using exactly the same processing procedure as kana-kanji conversion processing. When the present invention is implemented here, some of the headings that were conventionally distinct from each other become the same. For example, in Table 5, there are the notations ``Chief'' and ``Ataxia'' for the heading ``Chitsuchiyo''. The appearance of many homonyms in this way is an undesirable phenomenon in order to obtain correct notation, but for example, if we take Shogakukan's Newly Selected Japanese Dictionary as an example, the present invention has a total number of 70,000 words in the header characters. The increase in homophones due to heading conversion is approximately 15
There are only categorical positions, and this increase in homophones poses almost no problem in practical terms. An example is shown in Table 5.

【table】

[Table] In addition, when inputting something other than voice input into the automatic kana-kanji conversion processing unit having a dictionary group implementing the present invention, it is also possible to input the consonant "tsu", but in that case, the input consonant "tsu" If it is uniquely converted to the direct sound "tsu" character, it will be processed without any problem. In addition, if the conversion of the consonant "tsu" is complicated, leave the heading section containing the conventional consonant "tsu",
Another possible method is to add a new term in which the consonant "tsu" is converted to the direct sound "tsu" character. However, this method requires an increase in the storage capacity of the memory because items are added. In any case, the automatic kana-kanji conversion device embodying the present invention performs the conversion process for the input of the consonant "tsu" as described above, or saves the heading of the consonant "tsu" to input it from a conventional keyboard. It is also possible to process both raw character string input and voice input mixed together. As described above, according to the present invention, a voice input typewriter can be realized without disturbing the naturalness of voice.

[Brief explanation of the drawing]

Figure 1 is a diagram showing the monosyllabic speech recognition unit in a voice input Japanese typewriter, Figure 2 is a diagram showing the automatic kana-kanji conversion processing procedure for converting a kana input sentence into a sentence containing kanji and kana, and Figure 3 is a general diagram. FIG. 4 is a diagram showing an example of an independent word dictionary used in the present invention. 1...Microphone, 2...Preamplifier, 3
...16CH BPF, 4...16CH A/D converter, 5...Logarithmic conversion ROM, 6...Pitch extractor, 7...Counter timer, 8...Micro
CPU, 9...BUS LINE, 10...ROM, 1
1...RAM, 12...FDP, 13...CRT, 1
4...KB, 15...Hardware computing unit.

Claims (1)

[Scope of Claims] 1. In a kana-kanji conversion method that converts a kana character string obtained by monosyllable recognition of input speech into a sentence containing kanji and kana and outputs the result, the consonant "tsu" is changed to the direct sound "tsu" at the time of input. A dictionary for input character verification is provided that converts the consonant "tsu" that occurs in kana headings into the direct sound "tsu", and the kana character strings input in bunsetsu units are converted into kana headings of the dictionary. Test the kana character string by comparing it to independent words and attached words,
A conversion method in a voice input Japanese typewriter, characterized by outputting the notation of a corresponding word or phrase.