JPH0352058A - Document processor for voice input - Google Patents

Abstract

PURPOSE:To decide and determine the attribute of a displayed character by using the features of a voice and to improve the operability by providing a voice feature extracting means and a decision means which determines the attribute of the character. CONSTITUTION:The voice feature extracting means 2-2 extracts the features of the voice such as the voice power, voice pitch, and voice section length in a voice section. The decision means 24 decides the respective features of the voice obtained by the voice feature extracting means 22 to determined and send the thickness, density, font, etc., of the display character to an editing means 20. Consequently, a user need not sends an indication with a switch, etc., unlike before; and, for example, a thick character is displayed with a loud voice and a double-sized character is displayed by slow voicing, thereby improving the operability.

Description

[Detailed Description of the Invention] [Summary] This invention relates to a voice input document processing device that creates documents, etc. by voice input, and in particular, when characters are displayed, the density, size, font, etc. of the characters can be adjusted according to the size of the voice. An object of the present invention is to provide a document processing device with excellent operability, which determines and determines the attributes of displayed characters based on the characteristics of the voice, regarding a document processing device that inputs instructions based on the pitch and length of the voice. A signal conversion means for converting input speech into an electrical signal, a speech analysis means for analyzing the converted electrical signal, and a speech section cutting means for cutting out a speech section according to the analysis result.
Recognize the audio of the extracted audio section! ! means of knowledge,
! ! A character code generation means for converting the tA result into a character code, an editing means for editing a document from the converted character code, an audio feature extraction means for extracting the characteristics of the input voice from the converted electric signal, and an extracted character code. A determination means for determining the attributes of a character indicated by a character code by determining the characteristics of the voice. [Industrial Field of Application] The present invention relates to a voice input document processing device that creates documents, etc. by voice input. [Prior Art] In a conventional voice input document processing device, as shown in FIG. 6, uttered voice is converted into an audio electrical signal by an electrical signal converter 30 (for example, a microphone). The electrical signal is analyzed by the audio analysis section 32 for its audio spectrum and audio power (volume) every 10 to several tens of milliseconds, and a time series of vector quantity spectrum and scalar quantity power is output. In the speech segment extraction unit 34, speech segments are extracted using the analyzed power time series. In other words, the beginning and end of the voice can be found on the voice power time series. These start and end points coincide with the start and end points of the voice on the spectral time series. The recognition unit 36 performs recognition using the spectrum between the start and end determined by the speech segment extraction unit 34. Various recognition methods have been developed, including methods using dynamic programming (DP method), methods using hidden Markov models (HMM), methods using discriminant analysis, and methods using neural networks. There is a method. Recognition units include monosyllables, characters, words, phrases, and sentences.
No matter what unit of recognition it is, it is a word! ! (Japanese, English, etc.) have homonyms, so it is necessary to select candidates for kana-kanji conversion. Therefore, in many cases, the recognition unit 36 only outputs a symbol string of phonemes (a string of consonants and vowels, which in the case of Japanese is almost always written in Roman letters). Candidate selection for kana-kanji conversion is performed by the character code generation section 38 and first instruction section 40. That is, first, the phoneme symbol string sent from the recognition section 36 is converted into a kanji string, and the first candidate is sent to the editing section 4.
2 on the display section 44. Next, if the displayed kanji candidate is not desirable, the user presses the first instruction section 40 (specifically, the next candidate key).
In the character code generation section 38, when the first instruction section 40 is pressed, the next kanji candidate is displayed. Here, if the correct kanji is displayed, the user makes the following utterance. At that point, the previous kanji candidate is confirmed. In editing g42, the text (displayed to be input) is edited according to instructions from the second instruction section 46 (consisting of switches, etc.). The specific contents of editing include changing the thickness and width of the characters, changing the font (Mincho, etc., Gothic, etc.), changing the darkness of the characters,
This includes deleting incorrect entries. The display unit 44 displays character strings (sentences, etc.) on, for example, a CRT display, and displays them as stripes. Then, the printing unit 48 prints the &W collected character strings (sentences, etc.) on paper. [Problem to be solved by the invention] However, with such a conventional device, an instruction from the second instruction unit 46 is required to change the size of characters, for example, and therefore voice input is not required. It was extremely inconvenient to use, as it was necessary to interrupt the process to operate the switch. The present invention has been devised to address the above-mentioned conventional problems, and provides a voice input document processing device with excellent operability that can determine the attributes of displayed characters by determining them based on the characteristics of voice. With the goal. [Means for Solving the Problems] In order to achieve the above object, the device according to the present invention has the following features:
T! ! It is structured like I. In the figure, audio is converted into an electrical signal by signal conversion means 10. The converted electrical signal is analyzed by the audio analysis means l2. According to the analysis result, a speech section is cut out by the speech section cutting means 14. The voice of the cut out voice section is recognized by the recognition means 16 as L!
The recognition result is converted into a character code by the character code generating means 18. Furthermore, the document is edited by the editing means 20 using the converted character code. Here, the voice feature extracting means 22 extracts the characteristics of the input voice from the computer signal, and the determining means 24 determines the extracted voice characteristics and determines the attributes of the character indicated by the character code. [Operation] In the present invention, the speech feature extraction means 2 extracts speech characteristics such as speech power, speech pitch, and speech section length within a speech section. The determining means 24 determines each feature of the voice obtained by the voice feature extracting means 22, determines the size, sharpness, font, etc. of displayed characters, and transmits them to the editing means 20. That is, determining means 2
4 is the second instruction unit 4 in the conventional device based on the audio information.
Substitute 6. Therefore, the user does not need to give instructions using switches or the like as in the past; for example, if the user speaks loudly, the text will be displayed in plain text, or if he speaks slowly, the text will be displayed in double-width text. [Embodiments] Hereinafter, preferred embodiments of the apparatus according to the present invention will be described based on the drawings. 2nd E! I shows the overall configuration of the first embodiment,
In the figure, blocks with the same symbols as in the conventional example have the same functions. Here, the pitch calculation section 50 calculates the average pitch F of the speech within the speech section (from the beginning to the end) obtained by the speech section cutting section 34. The power calculation section 52 calculates the average power P of the speech within the speech section (from the start to the end) obtained by the speech section extraction section 34. Furthermore, the section length calculation section 54 calculates the average power P of the speech within the speech section (starting end to end) obtained by the speech section extraction section 34. The length of the speech interval (time length) L is calculated from the end. Then, the phoneme number calculation unit 56 calculates the phoneme number C of the first candidate (the most plausible candidate) among the recognition candidates obtained by the recognition unit 36. The determination unit 58 includes a pitch calculation unit 50, a power calculation unit 52,
fli obtained by the section length calculation unit 54 and the phoneme number calculation unit 56
F. P. L. Make a comprehensive judgment based on C and I! Notify the collection department 42. The editing section 42 displays the character attributes determined based on these results on the display section 4.
4 and printed in print 84B. Here, as a criterion for judgment, for example, the criterion shown in FIG. 4 is used. That is, as understood from FIG.
0, the value F. obtained by the power calculation unit 52. P makes a judgment [5
When the value is larger than the predetermined illll[Fth, pth in step 8, it is determined that the text is displayed in dark or italic text. Next, the section length calculation unit 54 calculates the number of phonemes [the value L obtained in step 5B]. When the can L/C in C is larger than a predetermined threshold [T}{, the determination unit 58 determines to display it in double-width characters. Next, FIG. 3 shows a second embodiment of the present invention. Threshold value Pth of the judgment criteria used in the first implementation (see 41st!I)
Fth. The TH was determined based on the assumption that the number of people would be unspecified, and it did not take into account the subtle differences in the voices of the operators who actually input voice. Therefore, in this embodiment, the voice characteristics of the operator who actually inputs voice are stored, and the voice is used as a threshold value for the judgment criterion. In this embodiment, for example, the determination criteria shown in FIG. Sometimes it is determined to be displayed in dark letters. Further, when the value F obtained by the pitch calculation section 50 is 50% or more larger than [Fb of the previous utterance stored in the previous feature storage section 60, it is determined that the value is displayed in italics. Furthermore, the ML. At C, the value of L/C is 50% from the value of fllL b /C b stored in the previous feature storage unit 60 before the first utterance.
If the value is larger than that, it will be determined to display double-width characters. As described above, according to this embodiment, it is possible to grasp even subtle differences in the voices of operators who actually perform voice input, thereby further improving the operability of the present device. [Effects of the Invention] As explained above, according to the present invention, the size, density, font, etc. of displayed characters can be converted by changing the volume, pitch, or speed of voice input when inputting voice. This can be done by simply changing the

[Brief explanation of drawings]

Figure 1 is a diagram explaining the principle of the present invention, and Figure 21! I is a block diagram of the first embodiment of the present invention, FIG. 3 is a block diagram of the second embodiment of the present invention,
Fig. 4 is a specific example of the judgment criteria used in the 3111 actual JV example, Fig. 5 is a specific example of the judgment criteria used in the second example, and 6th v;! i is an explanatory diagram of the configuration of a conventional example. 10...Signal conversion means, 12...Speech analysis means, 14...Speech section extraction means, I6...Recognition means, 18...Character code generation means, 20...Editing means, 22. ...Voice feature extraction means, 24...Judgment means, Principle explanatory diagram of the present invention No. 1rEJ Fig. 2 p>p th F>Fth L/C>TH If then then dark italic characters double-width characters Here, pth, Fth, TH are predetermined a values If P>1.5XPb If dark text (bold text) F>1.5XFb If italic text L/C> 1.5XLb/Cb
Double-width characters Here, Pb, Fb, Lb, and Cb are each value before one utterance. Figure 6

Claims (1)

[Scope of Claims] Signal converting means (10) for converting input speech into an electrical signal, speech analysis means (12) for analyzing the converted electrical signal, and speech section cutting means for cutting out a speech section according to the analysis result. (14), and a recognition means (16) for recognizing the voice of the cut out voice section.
) and a character code generation means (
18) and editing means (2) for editing the document using the converted character code.
0), a voice feature extracting means (22) for extracting the characteristics of the input voice from the converted electrical signal, and a determining means (22) for determining the attributes of the character indicated by the character code by determining the characteristics of the extracted voice. 24) A voice input document processing device comprising: