JPH04242300A - Voice recognition device - Google Patents

Abstract

PURPOSE:To obtain the voice recognition device which automatically stops a tape recorder when there is a message for voice input interruption even in the absence of an operator who confirms the message from a display part and automatically restarts the rotation of the tape recorder when the message for voice input interruption is released. CONSTITUTION:This device is equipped with an up-counter 15a which counts the feature quantity from a feature detection part 7 and outputs a data-full signal when the counted value reaches a specific value, a down-counter 15b which counts down each time the matching distance is read in from an input buffer memory and outputs a data-empty signal when reaching a specific counted value, and a tape voice input judging circuit part 15c which judges that the input buffer memory is full when an end detection signal is inputted after the data-full signal is inputted and stops the tape recorder, and judges that the input buffer memory is empty when the data empty signal is inputted from the down counter and then restarts the rotation of the tape recorder.

Description

[Detailed description of the invention]

0001

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a voice recognition apparatus, and more particularly to control for preventing erroneous recognition of voice signals from a tape recorder.

0002

[Prior Art] Conventional speech recognition devices generally use microphones for input.
In some cases, line input from a precoder or the like is also possible, and in this explanation, a voice recognition device that also allows line input will be used.

FIG. 2 is a schematic diagram of a conventional speech recognition device. In the figure, 1 is a microphone, 2 is a tape recorder, 3 is a microphone input terminal, and 4 is a line input terminal.

[0004] 5 is a changeover switch, and microphone 1
Alternatively, when an audio signal from either tape recorder 2 is input, the switch can be made manually.

Reference numeral 6 denotes an audio input section, which receives a digital audio signal (which is amplified, band-limited, A/D converted, etc.) input from the microphone 1 or tape recorder 2.
This is to output audio data (hereinafter referred to as audio data).

[0006] Reference numeral 7 denotes a feature detection unit, which detects a voice section, extracts feature quantities such as spectral and cepstrum by known analysis methods such as FET and LPC analysis, and outputs an end detection signal indicating the end of the section. It is something.

[0007] Reference numeral 8 denotes a matching unit, which performs a similarity calculation between the features of the speech data obtained by the feature extraction unit 7 and standard patterns (features of speech to be recognized such as syllables and words). and outputs a distance value (known distance value such as city value distance, WLR, etc.).

9 is a minimum value search unit, and a matching unit 8
The distance value output from (hereinafter referred to as matching distance) is n
The minimum distance is selected, and the standard pattern corresponding to the minimum distance value is output to the display unit 10 as recognized voice data (referred to as recognized voice data). .

Reference numeral 10 denotes a display unit that displays the output recognized speech data and indicates whether the next speech input is possible or whether it is necessary to wait for the next input. This is to prevent the input buffer memory from becoming full if speech is continuously input in units of words, phrases, etc., and the next distance value is input before it is recognized. Reference numeral 11 denotes a control section that outputs timing signals to each section to obtain predetermined data.

A conventional speech recognition device configured as described above will be explained. For example, when the selector switch 5 is switched to input a reproduced audio signal from the tape recorder 2, the audio input section 6 outputs reproduced audio data from which noise and the like have been removed to the feature extraction section 7.

The feature extraction section 7 extracts feature amounts from the input reproduced audio data using a known analysis method and outputs them to the matching section 8. Next, the matching unit 8 performs the feature extraction unit 7
A similarity calculation is performed between the feature amount of the reproduced audio data obtained in step 1 and the standard pattern, and a matching distance is output to the minimum value search section 9 based on the end detection signal. Then, the minimum value search unit 9 temporarily stores the matching distance output from the matching unit 8 in the input buffer memory, selects the minimum distance,
The standard pattern corresponding to the minimum distance value is output to the display unit 10 as recognized reproduced audio data. Then, the display unit 10 displays the recognized reproduced audio data, and displays whether the next audio input is possible and whether it is necessary to wait for the next input.

Furthermore, when the selector switch 5 is switched to input the audio signal from the microphone, each part operates in the same manner as described above and displays the above-mentioned necessary message on the display part 10, so that the operator can The next voice was input via the microphone 1 in response to the message.

[0013]

[Problem to be Solved by the Invention] In the conventional speech recognition device as described above, when the display section is notified from the minimum search section whether the speech is able to be recognized or not, and the display section displays this in a message. , operated by microphone.
If the computer is inputting audio, you can input or stop audio depending on the message, but the
When inputting audio continuously with the precoder, the minimum search unit updates the input buffer memory to the matching distance when the matching distance is input, regardless of the state of the input buffer memory, so the input buffer memory is updated to the matching distance on the display unit. Even if a message is displayed indicating that the tape recorder cannot be used, the problem is that the tape recorder continuously outputs the reproduced audio signal and updates the input buffer memory as a result, making it impossible to recognize the audio. .

The present invention has been made to solve the above problems, and even if there is no operator to check the message from the display, if there is a message to interrupt voice input, the screen will be displayed. - It is an object of the present invention to provide a voice recognition device that can automatically stop a recorder and automatically reproduce the rotation of a tape recorder when a message to interrupt voice input is released.

[0015]

[Means for Solving the Problems] A voice recognition device according to the present invention is a tape recorder that outputs a voice signal recorded on a tape when a playback instruction is given, and stops playback when a stop command is given. Detect the audio section from the audio signal from the da,
It has a feature detection section that outputs an end detection signal that indicates the end of the speech section after outputting the feature amount of the voice, and calculates the matching distance of the feature amount of the voice from the feature detection section every time the end detection signal is output. In a speech recognition device that has a minimum search unit that temporarily stores a predetermined number of matching distances in an input buffer memory that can store a predetermined number of matching distances, recognizes speech in the order in which it is stored, and outputs it to a display unit, a feature value is detected from a feature detection unit. an up counter that counts up each time it is output to the matching section, and when the count value reaches a predetermined number before the input buffer memory is filled, outputs a data full signal to clear the count value; The number of stored items in the buffer memory is set in advance, and the minimum value search unit counts down each time the matching distance is read from the input buffer memory, and the count value becomes a count value that allows it to be determined that the input buffer memory is empty. When it arrives, the date
After outputting the data empty signal, inputting the data full signal from the down counter which sets the storage number to the set value, and the data full signal from the up counter, it is determined that the input buffer memory is full. A tape audio device that outputs a stop signal to the tape recorder, and when a data empty signal is input from the down counter, it determines that the input buffer memory is empty and outputs a playback signal to the tape recorder. It is equipped with an input determination circuit section.

[0016]

[Operation] In the speech recognition device of the present invention, when the feature detection section calculates the feature amount of the speech from the tape recorder and outputs it to the matching section, the up counter causes the matching section to detect the feature amount. It counts up each time it is output, and when the count value reaches a predetermined number before the input buffer memory of the minimum value search means is filled, it outputs a data full signal to the tape audio input determination circuit, and
Clear the count value.

Next, the tape voice input judgment circuit section judges that the input buffer memory is full when an end detection signal indicating the end of the voice section is input from the characteristic section after the data full signal is input. and outputs a stop signal to the tape recorder to stop the tape recorder from playing the tape.

Next, when the matching distance of the input buffer memory is read by the minimum value search means, the down counter counts down, and when the count value reaches a count value at which it can be determined that the input buffer memory is empty, the down counter counts down. - After outputting the tape empty signal to the tape audio input determining circuit section, the number of stored tapes is set to a preset number.

Then, the tape audio input judgment circuit section
When a tape empty signal is input, it is determined that the input buffer memory is empty, and a playback signal is output to the tape recorder to start playback again.

[0020]

DESCRIPTION OF THE PREFERRED EMBODIMENTS FIG. 1 is a schematic diagram of a speech recognition apparatus according to the present invention. In the figure, 1 to 11 are the same as those in FIG. 2 above. Reference numeral 15 denotes an input buffer monitoring section, which has a circuit described below.

15a is an up counter, which increases the matching distance of the audio data output from the matching section 8 by 1.
The counter is incremented one by one, and when the value reaches n-2 from the number n stored in the input buffer memory, a data full signal indicating that the input buffer is full is output to the tape audio input judgment circuit described later. It is something.

Reference numeral 15b is a down counter, which counts down when each n-2 matching distance stored in the input buffer memory of the minimum value search unit 9 is output, and the value indicates that the input buffer memory is empty. If the value reaches a value that can be determined, a data empty signal is output to a tape audio input determining circuit, which will be described later.

15c is a tape audio input judgment circuit;
When a data full signal is input from the up counter 15a, preparations are made to stop the playback of the tape recorder 2, and when an end detection signal is input from the feature detection section 7, a stop signal is immediately sent to the tape recorder 2. When a data empty signal is input from the down counter 15b, a reproduction signal is immediately output to the tape recorder 2 to output reproduced audio data.

Note that in this case, the input buffer memory is n
The termination detection signal is always outputted while reaching the end point.

The operation of the speech recognition device configured as described above will be explained below. Switch the selector switch 5 to operate the tape recorder 2 and play the reproduced audio data.
output the data to the audio input section 6.

The audio input section 6 amplifies the reproduced audio signal,
The reproduced audio data that has been subjected to band limitation, A/D conversion, etc. is output to the feature detection section 7.

The feature detection unit 7 detects a voice section from the input reproduced voice data using a known analysis method, sequentially extracts and outputs the feature quantities of the reproduced voice data, and detects the end of the voice section. Outputs a termination detection signal to notify. Next, the matching section 8 calculates the degree of similarity between the feature amount of the reproduced audio data obtained by the feature detecting section 7 and the standard pattern, and outputs the matching distance to the minimum value searching section 9.

Then, the minimum value search section 9 performs the matching section 8.
Temporarily store and read the matching distances output from n into a buffer memory, select the minimum distance,
The standard pattern corresponding to the minimum distance value is output to the display section 10 as recognized and reproduced audio data. And display section 10
displays the recognized voice data and indicates whether the next voice input is possible or whether it is necessary to wait for the next voice input.

Further, the up counter 15a counts up the matching distance output from the matching section 8, and the value increases from the storage capacity n of the input buffer memory to n-2.
When the value reached is reached, the tape audio input judgment circuit 15c
When the input buffer memory is full, a data full signal is output.

When the tape audio input judgment circuit 15c receives the data full signal, it prepares to stop the tape recorder 2, and when the end detection signal is input from the feature detection section 7, it prepares to stop the tape recorder 2.
A stop signal is output to the tape recorder 2 to stop it.

At this time, as explained above, the minimum value search unit 9 reads the matching distances temporarily stored in the buffer memory, selects the minimum distance, and recognizes the standard pattern corresponding to the minimum distance value to reproduce the reproduced voice. Since it is output as data,
The down counter 15b counts down each time the matching distance is output and read from the input buffer memory, and when the value reaches 0, for example, it outputs a data empty signal indicating that the input buffer memory is empty. The input signal is output to the voice input judgment circuit 15c.

Then, the tape audio input determination circuit 15c outputs a reproduction signal that causes the tape recorder 2 to reproduce and output the audio again, and outputs the reproduced audio data, so each part is the same as above. make certain movements.

[0033]

As described above, according to the present invention, each time the feature amount of the sound from the tape recorder is determined by the feature detection section and outputted to the matching section, the number of outputs is counted. However, when the value reaches a predetermined number before the input buffer memory of the minimum value search section is filled, the tape recorder is stopped, and each time the matching distance stored in the input buffer memory is read out, By counting the number of readings, determining from that value that the input buffer memory is empty, and outputting a playback signal to the tape recorder to start playback again, the display section Even if there is no operator to check the message, the tape recorder can be stopped automatically or the tape recorder can be replayed according to the message on the display to avoid mistakes. The effect of preventing recognition has been achieved.

[Brief explanation of the drawing]

FIG. 1 A schematic configuration diagram of a speech recognition device of the present invention.

[Figure 2] Schematic configuration diagram of a conventional speech recognition device

[Explanation of symbols]

1. Microphone 2 Tape recorder 5 Selector switch 6 Audio input section 7 Feature detection section 8 Matching section 9 Minimum value search unit 10 Display section 11 Control section 15a Up counter 15b Down counter 15c Tape audio input judgment circuit

Claims (1)

[Claims] Claim 1: Detecting an audio section from an audio signal from a tape recorder that outputs an audio signal recorded on a tape when there is at least a playback instruction, and stops playback when a stop command is given, a feature detection unit that outputs an end detection signal indicating the end of the voice section after outputting the voice feature, and the end detection signal outputs a matching distance of the voice feature from the feature detection unit; In the speech recognition device, the speech recognition device has a minimum search unit that temporarily stores the matching distance in an input buffer memory capable of storing a predetermined number of matching distances, recognizes the speech from the stored order, and outputs it to the display unit. Each time a feature quantity is output from the matching section to the matching section, the count is counted up, and when the count value reaches a predetermined number before the input buffer memory is filled, a data full signal is output and the count value is cleared. An up counter to store the matching distance in the input buffer memory is set in advance, and each time a matching distance is read from the input buffer memory by the minimum value search unit, the count value is counted down and the count value is stored in the input buffer memory. When the counter reaches a count value that can be determined to be empty, a data empty signal is output, and then a down count is set to the set storage number, and after a data full signal from the up counter is input, the data empty signal is output. When an end detection signal is input, it is determined that the input buffer memory is full, and a stop signal is output to the tape recorder, and when a data empty signal is input from the down counter, the input buffer memory is determined to be full. 1. A speech recognition device comprising: a tape audio input determining circuit section that determines that the tape is empty and outputs a reproduction signal to the tape recorder.