JPS6184771A - Voice input device - Google Patents

Abstract

PURPOSE:To confirm input results in the state convenient for use which agrees with the request of a user at each time, by setting various execution modes and making it possible that the user selects freely these modes. CONSTITUTION:A voice synthesizing part 6 and a character display part 19 request data for voice synthesis and character display of one word or clause to an input voice recording part and a KANA (Japanese syllabary) - KANJI (Chinese character) conversion output recording part 13 by a command from a control part 6 and synthesize a voice and perform D/A conversion and output a reading voice and display a sentence where KANJI, KANA, etc. are mixed. In this case, a coefficient which modifies parameters is outputted from a modify ing signal generating part 16, which modifies synthesis parameters, for the purpose of changing the tone color of the synthesized voice in accordance with a command from the user. The user inputs confirming/correcting information 21 from a key input part 22. When confirming information is inputted, the control part 6 outputs the next word or clause.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Application of the Invention] The present invention relates to a voice input device, and more particularly to a method for recognizing input results in a voice typewriter.

[Background of the invention]

A voice input device, especially converting input voice into a string of characters. In so-called voice typewriters, it is important to check whether the input voice is correctly converted into a character string. The conventional device is, for example,
As described in issue G, syllables can be checked for each input, as described in JP-A-54-136134, input speech and characters can be reproduced and displayed in a time-related manner and read together, or converted. Excellent ideas have been used, such as synthesizing speech from characters by regular synthesis, playing it alternately with the input speech, matching it, and outputting the input speech with different sounds such as echoes depending on the input information specifying the character type.

In systems that perform kana-kanji conversion or conversion of homophones into character strings, there is conversion of homophones and allographs, and rather than outputting audio from the converted character string by rule synthesis, it is possible to convert the input audio and the converted characters. It is more reasonable to match columns. However, a significant number of people do not like hearing their own voice played back, for example on a tape recorder. Therefore, it is desirable to reproduce the personal characteristics of the timbre in a modified voice.

Further, the optimal format for reading confirmation differs depending on the degree of completion of the target text, but an input system with a flexible configuration that can handle these situations has not been realized.

When inputting information, there is also a problem in that if you check while inputting, your thinking is interrupted.

[Purpose of the invention]

SUMMARY OF THE INVENTION An object of the present invention is to solve the above-mentioned problems and provide an input result confirmation means suitable for a flexible and easy-to-use voice input system that meets the requirements for special treatment.

[Summary of the invention]

In order to achieve the above object, the present invention first provides means for collectively recording input audio in a reproducible audio format. This recording means records signals that are input externally at the same time as the input, or signals that indicate the boundaries of words or phrases that are automatically determined and output during recognition or post-recognition Japanese processing for hypothetical kanji conversion. Provide a means to associate and record. Furthermore, the input audio is combined into a! according to an externally designated mode, while inputting, or according to a user's designation after inputting. After the recognized results are converted into kana/kanji characters, etc., the converted results are provided with a means for recording the converted results with symbols indicating the boundaries of words or phrases.

When reproducing input audio during reading confirmation, there is a means for reproducing the audio by changing the timbre according to commands given by the user from the outside, a means for changing the playback speed according to the instructions, and a command. control means for reproducing word or phrase by word or for continuously reproducing the entire input up to a point in time specified separately; Provide an easy-to-understand display aid (cursor, blinking, color change, etc.).

By setting a mode in which the functions of these methods are arbitrarily combined and allowing the user to freely select these modes, input results can be created in an easy-to-use state that meets the user's needs at the time. It becomes possible to perform confirmation.

[Embodiments of the invention]

An embodiment of the present invention will be described below with reference to FIG.

An input voice 1 is converted into a digital signal by an A/D converter 2, and is recorded in an analysis multi-input voice recording unit 4 by an analysis unit 3. The analysis method may be any method as long as it is consistent with the recognition method of the recognition unit. Furthermore, since it is also used for reading and compositing, a method suitable for single compositing is preferable, but it is of course possible to perform processing separately for both. Although it is preferable for efficiency that the two be the same, in many cases, the analysis of the sound source of the synthesized sound is not used for recognition, so it is necessary to perform it only for one synthesis. The recognition unit 5 requests input voice 8 according to the recognition command signal 7 from the control unit 6.
The analyzed input voice is recognized and the recognition result is written into the recognition result recording section 9. Many publicly known techniques are known for the configuration of the recognition unit 5, and no matter which system is adopted, the details thereof are not essentially related to the present invention, so a description thereof will be omitted. [Kana-kanji conversion processing unit] 0 outputs a recognition result request signal 12 in response to a kana-kanji conversion processing command signal 11 from the control unit 6,
The recognition result is taken in, the input is converted into a sentence containing kanji while finding the word or phrase boundary through processing such as morphological analysis, and the output is written to the kana-kanji conversion output recording section I3, and the found word or phrase, etc. The symbol 14 at the boundary position is input to the audio recording section 4 and the kana-kanji conversion output recording section 1.
3. Record the information in the corresponding predetermined location. Known technology can be used to convert kana character strings into kanji, so the explanation will be omitted here6.Note that the boundaries between words and phrases can be synchronized with input by the speaker using a separate switch etc. during voice input. Of course, it doesn't matter if you enter it.

Note that each process of inputting audio, recognizing it, and converting it into kana-kanji, etc. can be executed in parallel in a pipeline, or in batches for each part, and these modes can be selected. It is clear that this can be easily changed by controlling each command from the control section 6 and the output timing of the control section 11 according to a command from the user.

This will be explained next when reading together. In response to a command from the control unit 6, the speech synthesis unit 6 and the character display unit 19 send data for speech synthesis and character display for one word or a segment to each input voice recording unit 4 and the kana-kanji conversion output recording unit 13. is requested, the voice is synthesized, D/A converted, and output as voice for reading aloud, and displayed as a mixed sentence such as kanji and kana. At this time, the control unit 6 changes the timbre of the synthesized voice according to the command from the user.
By the modification signal generation unit 16 that modifies the synthesis parameters,
Outputs the coefficients that transform the parameters. For example, if the speech synthesis section is an LSP synthesizer well known to those skilled in the art, each LSP parameter may be multiplied by a certain value to move the formant position of the synthesized speech, or the pitch frequency may be multiplied by a certain value to It is possible to change the speaking speed of the voice or synthesized sound by changing the height of the synthesizer or by controlling the synthesis parameter to be supplied to the synthesizer at an interval different from the analysis interval. User confirms/corrects information 21 in Key Human Resources Department 2
Input from 2. When the confirmation information is input, the control unit 6 outputs the next word or phrase. For correction information,
The information in the corresponding portion of the kana-kanji output recording section 13 is corrected, and the process proceeds to the next step.

In the case of modifying words or phrase boundaries, the input audio recording unit 4
The position of the corresponding boundary symbol is also corrected to avoid any discrepancy in the correspondence between reading and character display.

In addition, according to an instruction from the user, until the input 21 for error correction is input, the audio and characters are played continuously regardless of the word or phrase boundary, and the correction manual 21 is performed.
If there is, it is possible to easily set a mode in which output is stopped at the end of the word or phrase at the time of input, and restarted after correction processing. In addition, it is also possible to display characters in units of sentences or paragraphs first, and to indicate the position of the word or phrase that corresponds to the output of the reading voice using a cursor, brillex, color display, etc. It is.

Kanji has many homonyms, and it is easier to detect errors when they are displayed within the entire sentence.

Although the above embodiment has been described for Japanese, it goes without saying that the same structure can be applied to languages that have the same pronunciation but different tsuzuri.

〔Effect of the invention〕

As described above, according to the present invention, when a sentence including homophones and allographs is input by voice, the input voice and the sentence can be checked and corrected in correspondence with each other with great ease.

[Brief explanation of the drawing]

FIG. 1 is a block diagram for explaining one embodiment of the present invention.

Claims (1)

[Claims] 1. A means for recording input voice, a means for recognizing input voice, a means for converting the recognized result into characters, etc., and a means for converting a character string converted into characters, etc. a character display means for displaying single words or phrases or longer words, a means for inputting confirmation and correction information of the displayed content by the character display means, and a means for inputting information for checking and correcting the displayed content by the character display means, and inputting audio in the input voice recording means for each word or phrase. An audio input device characterized by comprising means for reproducing and outputting the audio in sections. 2. The voice input device according to claim 1, wherein the voice reproduction means includes timbre modification means so that the reproduced sound has a different timbre from the input voice. 3. The voice reproduction means is controlled to sequentially output in units of words or phrases each time information is input from the confirmation and correction information input means, and the character display means corresponds to the voice output from the voice reproduction means. 2. The voice input device according to claim 1, further comprising display assisting means for clearly identifying the position of the character. 4. The voice input device according to claim 1, wherein the voice reproduction means has a speed changing means so that the reproduced sound has a speaking speed different from that of the input voice. 5. The input voice recording means has a sufficient capacity and has a storage means having a sufficient capacity to record the output of the character etc. conversion means, and the input voice recording means has a storage means having a sufficient capacity to record the output of the character etc. conversion means, and the time of inputting voice, the time of performing recognition operation, and 2. The voice input device according to claim 1, wherein the voice input device is characterized in that the time points at which confirmation and correction are performed can be separated and processed separately. 6. The above-mentioned patent characterized in that the confirmation and correction means is normally in a state where confirmation information is input, and also has a mode in which correction information can be input only when correction is necessary. A voice input device according to claim 1.