JP2000076241A - Voice recognizor and voice input method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To discriminate an editing command from an input word which is inputted in voices. SOLUTION: When a voice break is detected by a voice break detector 3 of this voice input device, a main recognition engine 4 recognizes a voice with the detected voice break used as a trigger and the recognized voice is stored in a text area by a document preparing device 6. If a new recognition trigger is detected while the engine 4 is executing its recognizing operation, the syllable of the detected trigger is recognized by a sub-recognition engine 5. This recognized syllable is incompletely recognized by the engine 4 and accordingly the incomplete word is deleted from the text area. Meanwhile, the word recognized by the engine 5 is interrupted as an editing command and a sentence editing device 7 edits a text in response to the editing command.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

TECHNICAL FIELD The speech input word processor

2. Description of the Related Art Conventionally, there is a speech input word processor for creating sentences by speech input. This is to convert the voice input from a microphone etc. into a character string using a voice recognition device,
This is a system for creating a text by inserting and adding it to the editing memory of the character processing device. However, in actual text creation, not only insertion of characters but also editing operations such as deletion of already input characters are required.

[0002]

In the case of a system in which this editing command for deletion or the like is realized by pronouncing “sakuju” by voice input, when the word “deletion” is intentionally input in a document However, there is a disadvantage in that the "delete" input by voice is treated as an edit command, and the word "delete" cannot be input. As described above, in a system for inputting an edit command by voice, it is difficult to input a word indicating the edit command as a word in a document.

In the case of a system for inputting an editing command from a device other than voice recognition such as a keyboard, there is a disadvantage that the voice input and the keyboard input must be used properly, and the system becomes inconvenient for the user. .

SUMMARY OF THE INVENTION The present invention has been made in view of the above-mentioned prior art, and has as its object to provide a speech recognition apparatus and a speech input method which distinguish between a sentence input word to be inputted by speech and an edit command, and have good operability. Aim.

It is another object of the present invention to provide a speech recognition device and a speech input method which can be used by a user without difficulty by using a word inserted by a speaker in a sentence as an editing command. I do.

[0006]

In order to achieve the above object, the present invention has the following arrangement. That is,
First recognition means for recognizing a word from an input voice;
Second recognition means for recognizing an instruction from an input voice;
A speech recognition apparatus comprising: an editing unit that accumulates a word recognized by the first recognition unit to create a sentence and edits the sentence according to an instruction recognized by the second recognition unit.

[0007] Alternatively, a sentence creating step of adding a word speech-recognized by the first recognizing means to a sentence and a sentence editing step of editing the sentence based on an instruction speech-recognized by the second recognizing means. Voice input method to be equipped.

[0008] Alternatively, a sentence creating step of adding a word speech-recognized by the first recognizing means to a sentence, and a sentence editing step of editing the sentence based on an instruction speech-recognized by the second recognizing means. A computer-readable storage medium storing a program for executing the steps.

[0009]

FIG. 1 is a block diagram of a speech input word processor according to a first embodiment of the present invention. The audio signal input from the microphone 1 is subjected to noise removal and volume level stabilization by the microphone amplifier 2 so as to be easily processed for voice recognition, and then sent to a main recognition engine 4 and a sub-recognition engine 5 to be described later. Can be Further, the microphone amplifier 2 has a function of converting the presence or absence of a voice input into digital information and sending the digital information to the voice segment detection device 3.

The voice segment detecting device 3 includes a microphone amplifier 2
It has a function of detecting the break of the input word from the presence / absence information of the voice input from the PC and controlling the recognition operation of the main recognition engine 4. Further, it has a function of controlling the recognition operation of the sub-recognition engine 5 together with the presence / absence information of voice and the information of whether or not the main recognition engine 4 is currently performing the recognition operation. This control procedure will be described later.

The main recognition engine 4 includes a voice recognition chip that performs A / D conversion on the input analog voice information and processes the digital information. Further, since the speech recognition chip cannot receive the next speech during the recognition operation, it also has a function of outputting to the speech segmentation detection device 3 that the recognition operation is being performed. The recognition dictionary of the main recognition engine 4 is composed of words used in general sentences.

The sub-recognition engine 5 has the same structure as the main recognition engine 4, but includes a different dictionary.
The dictionary of the sub-recognition engine 5 is mainly composed of words that are referred to as “solo”, for example, “that” or “not”
It stores words that are not included in the text or command to be input by voice, such as "Um" or "No", but are muttered as expected. In the figure, an output indicating that the sub-recognition engine 5 is performing a recognition operation is not described, but this is not described because there is no need for this output. Like the main recognition engine 4, there is no problem even if there is a signal line indicating that the operation is in progress.

The document creation device 6 includes a main recognition engine 4
It has a function to insert a character string of a word output from a word processor into a document creation memory of a word processor.

The document editing device 7 performs an operation corresponding to the meaning of the word output from the sub-recognition engine 5, and has a function of, for example, deleting a word in the text being created by the document creation device 6. The storage device 8 is for storing the text created by the document creation device 6 and the document editing device 7.

The display 9 displays a sentence or the like being created.

It is also possible to print out a sentence created by an external I / F (not shown) or transfer the sentence via a communication line.

In this embodiment, the A / D converter is built in each of the main recognition engine 4 and the sub recognition engine 5, but one A / D converter is provided independently of the microphone amplifier 2. A device may be provided to pass this digital output to the recognition engine.

FIG. 2 is a flowchart showing an operation algorithm of the voice segment detection device 3.

In step S11, the voice presence / absence information input from the microphone amplifier 2 is inverted, and a trigger for the main recognition engine 4 used in the subsequent steps is created. In step S12, the trigger for the main recognition engine 4 is acquired, and in step S13, it is checked whether or not the edge of the signal change falls. If it does not fall, the process returns to step S11 to continue this process.

If the edge of the main recognition engine 4 is falling, the flow goes to step S14 to acquire the current operation state of the main recognition engine 4. Next, step S15
It is checked whether or not the main recognition engine 4 is recognizing. If it is, the process branches to step S16. If it is not, the process branches to step S17.

In step S16, the sub-recognition engine 5 is placed in a standby state for voice input in place of the main recognition engine 4 being recognized, and a mode for acquiring voice input from the microphone amplifier 2 is set. On the other hand, in step S17,
The sub-recognition engine 5 is stopped so that the voice input from the microphone amplifier 2 is not processed.

After steps S16 and S17 have been completed, the entire process ends in step S18.

With the above-described algorithm, the speech segmentation detection device 3 causes the sub-recognition engine 5 to recognize the speech input during the recognition operation of the main recognition engine 4.

In the present embodiment, the trigger for step S12 is created by inverting the voice presence / absence signal in step S11. However, in the case of a configuration in which the rising edge triggers the recognition engine from the beginning, step S11 Becomes unnecessary.

Further, the voice segment detection device 3 can be realized by software or by hardware combining a flip-flop circuit or the like. In the case of realizing by software, C (not shown)
The PU executes the program of the procedure of FIG. 2.

FIG. 3 is a timing chart showing an example of how speech input is actually processed by the apparatus shown in FIG. Specifically, when the user wants to input the text by saying “My name is Tanaka” to microphone 1,
Imagine a situation where you mistakenly say "Tamura" and then immediately rush "No".

The state of this sound is described in a graph G01 sentence in the figure. The point is that the voice of "difference" is generated closely to "tamura". In such a case, in the conventional speech recognition system, in such a case, since the speech recognition engine is recognizing the speech "tamura", the speech "different" cannot be captured.

The graph G02 sound level is an image diagram showing a state in which the sound input from the microphone 1 is output from the microphone amplifier 2.

The graph G03 voice presence / absence information shows digital information obtained by converting the analog signal into voice presence / absence information. In this figure, when a voice is being input, it is indicated by a high level, and when it is not input, it is indicated by a low level.

The graph G04 main recognition trigger is obtained by inverting the above-mentioned graph G03 voice presence / absence information.
Enters a voice input state, ends the input by a rising edge, and starts a recognition process. Timings “A” and “C” in the figure are described as an example in which the recognition processing of the main recognition engine 4 is started by a rising edge.

The graph G05 main recognition state indicates the state of the main recognition engine 4 in which the recognition process has been started by the graph G04 main recognition trigger. The state is high during recognition, and low when recognition is completed. Shown by level. "B" and "D" in the above are "A"
This represents the timing of the end of the recognition process according to the start of the recognition of “B”. When the recognition is started at the timing of "A", the recognition ends at the timing of "B" (fall). In the case of the present embodiment, the word "I" is recognized and output normally.

On the other hand, the word "difference" should be input from the timing "E", but is input from the timing "F" when the recognition is completed because the main recognition engine is busy. . Therefore, "C"
When the recognition is started at the timing of, since the location where the voice for recognition is started is “F” in the figure, the first part where the voice that “should be different” should normally be input is missing. For example, only the voice of “ga” in which the first sound “chi” is missing is input to the main recognition engine 4. Therefore, the recognition result output at the timing “D” is incorrect.

The graph G06 sub-recognition trigger is the graph G
When the main recognition state of the graph G05 is at a high level at the timing "E" of the fall of the 04 main recognition trigger, the voice segment detection device 3 generates the falling edge as a falling edge.
With this trigger, the sub-recognition engine 5 enters a voice input waiting state. Therefore, the voice “No” is taken into the sub-recognition engine 5 from the beginning and the timing “C ′”
Indicates the start of recognition.

The sub-recognition engine 5 whose recognition has begun is normally recognized as the word "difference" at the timing of "H" in the graph G07 sub-recognition state.

As described above, in the system of the present embodiment, it is possible to normally capture a muttered voice like a solitary message without sufficient recognition time.

FIG. 4 is a flowchart showing an algorithm for processing words recognized by the main recognition engine 4 and the sub recognition engine 5 as a document processing command. This procedure is executed by the document creation device 6 and the document editing device 7.

After obtaining the recognition result (= word character string) from the main recognition engine 4 in step S20, the process proceeds to step S20.
At 21, the character string is added to the text area S22. By the way, the process of acquiring the character string and adding it to the text is the same as the process of storing the character string input from the keyboard in the memory in a general word processor, so that the details are omitted.

Next, in step S23, the sub recognition engine 5
It is determined whether or not has been recognized. If not, the process returns to step S21.

When the sub-recognition engine 5 has completed the recognition, the same word as the word recognized by the sub-recognition engine has been incompletely recognized by the main recognition engine 4 immediately before. In the above-described example, “gaou” is recognized by the main recognition engine. Therefore, after acquiring the recognition result of the sub-recognition engine 5 in step S24, one word stored immediately before in the text area is deleted in step S25. As a result, "gau" in the above-described example is deleted.

Next, in step S26, a word output from the sub-recognition engine 5 is searched from the dictionary in which the command is registered.
And deletes one word stored immediately before the word deleted in step S25 from the text area. By this processing, the “tamura” in the above-described example is deleted.

If the word is not a negative word, step S2
Go to 8 and end.

The above is the algorithm for processing the monologue recognized by the sub-recognition engine 5 as a word processor.

As described above, the word recognized by the sub-recognition engine is analyzed, and if the word is registered as a command in the dictionary, the processing corresponding to the command is performed.

In this example, only the negative word is processed as a command. However, "up", "do", "magi", "hidari", etc., which mean the movement of the cursor, are placed in the conversion table and If you add processing to judge and analyze,
Further functions are improved and usability is improved. Also in this case, a word serving as a command is input while recognition is being performed by the main recognition engine, and is recognized by the sub recognition engine. In the processing method of the cursor movement of the document editing apparatus 7, since the cursor movement processing of a normal word processor can be used as it is, the details are omitted in this embodiment.

FIG. 5 shows a part of the conversion table used in step S26 of FIG. In FIG. 5, the right column shows the recognized word, the left column its meaning,
That is, it indicates a command. Note that, in this, a no operation is a word that does not need to be processed because it has no meaning.

As described above, the operation can be simplified by identifying words for text input and words for editing by the device. In addition, as a command for editing, a so-called self-indulgently generated word from the user is used.
There is no burden on the user.

[Second Embodiment] Many functions of the speech recognition system described in the first embodiment can be realized by a computer system. for that reason,
Without distinguishing the main recognition engine and the sub recognition engine,
A voice input within a predetermined time after a break in voice input as a sentence is detected is recognized as a command. This predetermined time is programmable. In this way, the trigger for command recognition can be set as a sound pronounced within a predetermined time from the start of recognition of a word input as a sentence. By setting this predetermined time to a value corresponding to the time required for the recognition processing by the main recognition engine in the first embodiment, a speech recognition system similar to that of the first embodiment can be realized.

FIG. 6 is a block diagram of a computer system for realizing voice recognition and text input / editing by software.

In FIG. 6, the audio signal input from the microphone 1 is amplified by the microphone amplifier 2 and
, And stored in the memory 602.

In the memory 602, in addition to the audio data, C
The program executed by the PU 604 is also stored.
This program executes the procedures of FIGS.
Is a program to be executed. However, in the present embodiment, not only editing of text obtained by recognition, but also voice recognition itself is performed by executing a program by a computer system, so that the voice recognition by the main recognition engine and the sub-recognition engine is one. Is realized by the program module.

FIG. 7 is a flowchart for realizing, by a computer, the procedure of FIG. 2 executed by the voice segment detecting device 3 in the first embodiment. First, in step S701, a recognition trigger is obtained from the audio data stored in the memory 602. This method may be the same as in the first embodiment. In step S702, speech recognition from a recognition trigger to a speech break is performed. Step S70
At 3, it is determined whether the time interval from the immediately preceding recognition trigger is greater than a predetermined threshold Tth. If it is larger, in step S704,
For example, a flag or the like indicates that the speech recognition result in step S702 is a result of recognition by the main recognition engine.
Indicates that the recognition result is obtained by the sub-recognition engine.

The recognition result thus obtained is shown in FIG.
According to the above procedure, the recognition result indicated as obtained by the main recognition engine is added to the text area, and the recognition result indicated as obtained by the sub-recognition engine is regarded as an edit command. If the command is a delete command, the immediately preceding word is deleted from the text as in step S27 in FIG. In FIG. 4, incomplete recognition results are deleted in step S25 in order to perform recognition in real time, but in this embodiment, step S25 is not necessary in order to sequentially recognize syllables.

As described above, the present invention can also be realized by the configuration shown in FIG.

[0054]

[Other Embodiments] Even if the present invention is applied to a system including a plurality of devices (for example, a host computer, an interface device, a reader, a printer, etc.), an apparatus (for example, a copying machine) Machine, facsimile machine, etc.).

Another object of the present invention is to provide a system or an apparatus with a storage medium in which the program code of the procedure shown in FIGS. 4 and 7 for realizing the functions of the above-described embodiment is supplied to a system or an apparatus. (Or CPU or MPU) by reading and executing the program code stored in the storage medium.

In this case, the program code itself read from the storage medium implements the functions of the above-described embodiment, and the storage medium storing the program code constitutes the present invention.

As a storage medium for supplying the program code, for example, a floppy disk, hard disk, optical disk, magneto-optical disk, CD-ROM, CD
-R, a magnetic tape, a nonvolatile memory card, a ROM, or the like can be used.

When the computer executes the readout program code, not only the functions of the above-described embodiment are realized, but also the OS (Operating System) running on the computer based on the instruction of the program code. ) Performs part or all of the actual processing, and the processing realizes the functions of the above-described embodiments.

Further, after the program code read from the storage medium is written into a memory provided on a function expansion board inserted into the computer or a function expansion unit connected to the computer, based on the instructions of the program code, The case where the CPU of the function expansion board or the function expansion unit performs part or all of the actual processing, and the function of the above-described embodiment is realized by the processing.

[0060]

As described above, according to the present invention, it is possible to provide a speech recognition apparatus and a speech input method which can distinguish a sentence input word and an edit command to be inputted by speech and have good operability.

Also, by using words related to editing of a sentence inserted by a speaker in a sentence as an editing command, it is possible to provide a voice recognition device and a voice input method that can be used by a user without difficulty.

[0062]

[Brief description of the drawings]

FIG. 1 is a block diagram of a speech recognition system.

FIG. 2 is a flowchart showing an algorithm of the voice segment detection device 3.

FIG. 3 is a timing chart showing a state of processing audio.

FIG. 4 is a flowchart showing an algorithm for processing a result of the sub-recognition engine 5.

FIG. 5 is a diagram showing an example of a conversion table describing the meaning of a word.

FIG. 6 is a block diagram of a computer system for realizing a speech recognition system.

FIG. 7 is a flowchart of a procedure for recognizing a sentence and a command in the system of FIG. 6;

[Explanation of symbols]

 Reference Signs List 1 microphone 2 microphone amplifier 3 voice segment detection device 4 main recognition engine 5 sub-recognition engine 6 document creation device 7 document editing device 8 storage device 9 display

Claims (9)

[Claims] 1. A first method for recognizing words from input speech
A second recognizing means for recognizing an instruction from an input voice, and a sentence created by accumulating words recognized by the first recognizing means, and recognizing by the second recognizing means. Editing means for editing the sentence in accordance with a given instruction. 2. The second recognizing means recognizes an instruction from the pronounced voice when the voice is pronounced within a predetermined time after the recognition by the first recognizing means is started. The speech recognition device according to claim 1, wherein: 3. The speech recognition apparatus according to claim 2, wherein the predetermined time is a time from when the first recognition unit starts recognition to when it ends. 4. An instruction recognized by the second recognition unit includes an instruction to delete a word recognized by the first recognition unit immediately before the instruction is recognized. The speech recognition device according to claim 1. 5. A sentence creating step of adding a word speech-recognized by the first recognizing means to a sentence, and a sentence editing step of editing the sentence based on an instruction speech-recognized by the second recognizing means. A voice input method comprising: 6. The second recognizing means recognizes an instruction from the pronounced voice when the voice is pronounced within a predetermined time from the start of the recognition by the first recognizing means. The voice input method according to claim 5, wherein a sentence is edited in the sentence editing step based on the instruction. 7. The voice input method according to claim 6, wherein the predetermined time is a time from when the first recognition unit starts recognition to when it ends. 8. The instruction recognized by the second recognizing means includes a word deletion instruction, and the word recognized by the first recognizing means immediately before the deletion instruction is recognized is sent to the sentence editing step. 6. The method according to claim 5, wherein
Voice input method described in. 9. A sentence creating step of adding a word speech-recognized by the first recognizing means to a sentence, and a sentence editing step of editing the sentence based on an instruction speech-recognized by the second recognizing means. A computer-readable storage medium storing a program for executing the steps.