JPS6037500B2 - voice typewriter - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to an improvement in a voice typewriter for inputting text information by voice.

The work of inputting text information into a computer is an important work in information processing work, and improving its efficiency is considered essential for improving the efficiency of the entire information processing work.

Particularly in our country, where typewriters are not commonly used, a large amount of training is required to develop a new character data entry skill. Therefore, it is strongly desired to develop an easy-to-use character data input means that replaces the conventional keyboard. Attempts were made to develop a voice typewriter as a promising candidate for this purpose. A voice typewriter is a device that converts human voice into character data codes through a pattern recognition process that analyzes and identifies the voice uttered by a human using an electronic circuit. However, recognizing natural speech is an extremely difficult problem, and it is said that there is almost no possibility that an economical speech typewriter will be realized in the near future. The biggest reason why speech pattern recognition is difficult is that it is difficult to identify the contributing factors. For example, even a detailed analysis does not yield significant and safe discriminative features among the pa, forced, and ta sounds, making them the most difficult to distinguish. This is because it is difficult to distinguish the consonants lpl, lkl, and ltl from each other. The present invention aims to solve the problem of consonant recognition by inputting difficult-to-recognize consonant information using a keyboard or the like, thereby realizing and providing a voice typewriter that is an accurate and inexpensive means of inputting character information. .

The voice typewriter according to the present invention has a voice input section that has a function of recognizing and encoding vowel parts of speech, a keyboard section for inputting consonant information, and a voice typewriter that integrates consonant information and vowel information provided by both of the above. It is composed of a character information synthesis section that synthesizes and outputs character information.

According to such a configuration, the recognition of vowel sounds is simpler than the one-stage phoneme recognition, so the overall size is small and economical, and the number of keyboards can be reduced as the keyboard only needs to be able to discriminate consonant parts. A character information input means that is small in number and therefore easy to operate is realized.

The principles of the present invention will be explained in detail below with reference to drawings of embodiments.

FIG. 1 is a block diagram showing an example of the overall configuration of a voice typewriter according to the present invention.

Keyboard section 10' is provided with a keyboard for specifying the consonant section of a phoneme. A consonant information signal C is generated by selectively pressing a key on this keyboard. An example of the key arrangement on the keyboard section 10 is as shown in FIG. 2. Keys marked with the letter "a" designate vowels and similar phonemes. For example, the phoneme lal, l
il, lul, lel, l o l, l of l, lyul
, lyol, lwal, and INl are selected by this key. We will call these phonemes and vowel groups. lkal, lkil for keys marked with the letter "k"
, lkul, lkel, and lkol are selected. Below, the same plane has g (ga sound), s (sa sound), z (za sound), t (ta sound), d (da sound), n (na sound), h (ha sound). (P sound), b (B sound), m (M sound), r (
It is equipped with keys such as ``La Gyon''. In addition, keys for punctuation marks "",".","linebreak","blank", etc. may be provided as in a normal typewriter. The consonant signal C, which is the output of the keyboard section described above, may be determined as follows. a: 0, k: 1, g: 2, s: 3, z: 4
, t: 5, d: 6, n: 7, h: 8, p: 9, b: 10
, m:11, r:12, .:13, o:14, line feed: 1
Blank: 16 For example, when the b key is pressed, the code 10 is output as the consonant signal C, and when the m key is pressed, the code 11 is output.

The voice input section is composed of a microphone 20 and a voice recognition section 30. The operator of this voice typewriter, for example,
To input the character gi', simultaneously press the "k" key and say "ki".

This audio signal is converted into an electrical signal i by the two microphones shown in FIG. 1, and sent to the audio recognition section 301.
The speech recognition unit 30 analyzes and identifies the speech signal i input from the microphone 20, and identifies the vowel parts of the vowel groups lal, l il,! ul, l e l, l o l,
It is determined which of lyal, IWl, lyol, lyal, and INI it belongs to and outputs it as a vowel signal v.
As an example, the following code is associated with this vowel signal v. la! :0, lil:1, lul:2, l e l:3, l o l:4, lyal:5lWI
:6, lyol:7, lwal:8lNI:9 The speech recognition unit 30 can be configured as follows, for example.

That is, an analysis unit for converting an input audio signal i into a matrix pattern having a time axis and a frequency axis by frequency analysis and time sampling, and a starting point of the input matrix pattern. a vowel extraction section for erasing the consonant section of the vowel section to create a vowel section pattern, a standard pattern storage section for storing a standard pattern that is a standard matrix pattern for each of the vowel group, and the vowel section pattern. and means for comparing the vowel and the standard pattern, and determines that the vowel is the same as the standard pattern that is determined to be the closest as a result of the comparison. In addition,
The vowel pattern is extracted when the power level of the voice is high.
This is performed based on criteria such as that the sound is voiced (that is, that a pitch component exists) and that the frequency component is stationary.
Character information synthesis section 4 River, consonant signal c given from the keyboard section 10 and vowel signal v given from the speech recognition section 30
Based on this, character information is synthesized and output as character code r.

For example, the consonant signal is c=1(k) and the vowel signal is v=4(l
ol), the character code "ko" (for example, JIS code) is output. Also, the consonant signal c is 13, 14, 1
5, 16 In other words, if it is one of “ ”, “.”, “line feed”, or “blank”, the special character code specified by the consonant signal is output regardless of the vowel signal v. It shall be. FIG. 3 shows an example in which the character information synthesizing section 40 as described above is constructed by a memory exclusively used for displaying.

That is, it is configured so that it can be read out by two-dimensional address designation (c, v) using a consonant signal c and a vowel signal v.
The consonant signal specifies a code of 10 to 16 with 5 bits, and the vowel signal specifies a code of 0 to 9 with 4 bits. It is assumed that a character code determined by the combination of the consonant signal c and the vowel signal v is programmed into the address (c, v) of the read-only memory 41. For example, following the above example, the code for the character "K" is recorded at address (1, 4). Furthermore, the code "." is programmed at the address (13, v) regardless of v.
In reality, the problem is when to output the character code r.

For example, in the case of a normal character, the consonant key is pressed, the voice is input, the vowel part recognition is completed, and then ``'', ``'' again.

For special characters such as ", which do not require voice input, it is appropriate to output the character code r when the key is pressed. Controlling the output time in this way is simple, so we will omit it. Fig. 4 1 is a diagram showing an example of the configuration of a speech recognition section 30, which is one main component of the speech typewriter according to the present invention.

The analysis unit 31 has a spectrum analysis function, detects the spectrum envelope of the input audio waveform i, expresses it as a vector aj of appropriate dimensions, and inputs it to the input pattern buffer 32 at a constant frame period. Assume that the following audio pattern has been input to the input pattern buffer 32 at the moment when the input of the audio pattern is finished. A=a,,a2,
...ai...a, {11 This is called an input pattern.

The end position detection section 33 determines the end position e, which is the address on the input pattern buffer where the final vector al of the input pattern A is stored, and sends it to the matching section 35. on the other hand,
The standard pattern storage unit 34 stores standard patterns of each phoneme included in the vowel group, and the vector bi is stored in a time-series format in the same way as the input pattern. Now, representing that one piece, B=bQ,...bi...bJ
Indicated by ■.

The pattern matching section 35 compares the input pattern A and the standard pattern B and calculates the distance d between them. In this case, the ends of both patterns are aligned and compared based on the information of the end layer e. FIG. 5 is a conceptual diagram showing how input pattern A and standard pattern B are compared.

Mainly the vowel portion of the input pattern A is compared with the standard pattern 8. Therefore, vowel parts can be directly compared without being influenced much by the consonant part. Therefore, it is assumed that the matching unit 35 calculates the distance between the input pattern A and the standard pattern B.

Here, the distance between the ll and ll' vectors is shown. Further, the denominator J is used to compensate for the tendency that the larger the length J of the standard pattern, the larger the sum of the numerators. Such distance calculations are performed with respect to all the standard patterns stored in the standard pattern storage section 34, and the results are sequentially sent to the determination section 36 as a signal d. Judgment section 36
Then, the distances d that are sequentially input are compared and the minimum value is determined. The vowel corresponding to this minimum value is taken as the determination result, and its code is output as the vowel signal v. In the previous example, a method was described in which the consonant part and the vowel part are separated in advance and the vowel part is recognized later, but in reality, separating the consonant part and vowel part requires a complicated method, and it cannot be done completely correctly. It may not be possible to do so.

When matching is performed by aligning the ends of the input pattern and the standard pattern as in the configuration shown in FIG. 4, there is no need to separate the consonant and vowel parts, and accurate recognition is possible. Although the configuration of the present invention has been described above based on examples, these descriptions do not limit the scope of the present invention.

In particular, the layout of the keyboard, the display of characters, etc. are not limited to the example shown in FIG. Furthermore, manual input means such as a light ben may be used. Further, other methods can be considered for classifying vowels and consonants, and the present invention is not limited to the examples described in the main text. Furthermore, in the above description, for example, “
When inputting the character ``K'', the user presses the ``k'' key and pronounces ``Ki'', but it is also possible to pronounce only the vowel part as ``i''.This may give the user an unnatural feeling. However, the speech recognition section can be simplified.

[Brief explanation of drawings]

FIG. 1 is an embodiment for explaining the basic configuration of the present invention. 10 is a keyboard section, a second microphone, a third speech recognition section, and 40 is a character information synthesis section. FIG. 2 is a diagram showing an example of a keyboard section. FIG. 3 shows an example of the configuration of the character information synthesis section. Block 41 in the figure is a read-only memory. FIG. 4 is a diagram showing an example of the configuration of the voice recognition section. 31 is an analysis section, 32 is an input pattern buffer, 33 is an end position detection section, 34 is a standard pattern storage section, 35 is a matching section, and 36 is a determination section. FIG. 5 is a conceptual diagram showing how input pattern A and standard pattern B are compared. Many drawings, 2nd drawing, 3rd drawing, 4th drawing, younger brother S drawing

Claims (1)

[Claims] 1. A means for manually inputting at least consonant information, a speech recognition unit for recognizing vowels input by voice, and generating character information by combining the consonant information and vowel information obtained by both of the above. A voice typewriter comprising: a means for outputting a voice.