JPH06102894A - Speech recognizing device - Google Patents

Abstract

PURPOSE:To provide the speech recognizing device which can securely record only a speech pattern and securely confirm the speech pattern stored as a feature pattern by reproducing the recorded speech pattern and is improved in recognition rate. CONSTITUTION:A speech section detection part 13 detects the speech section of a speech pattern supplied from a speech analytic part 12 on the basis of a low-level threshold value L and a high-level threshold value H, and generates a sound-recording request signal and a sound-recording interruption signal. A feature pattern storage part 14 stores a speech pattern corresponding to the detected speech section and a sound recording control part 16 writes the speech pattern in the speech pattern storage part 17 in response to the sound recording request signal outputted from the speech section detection part 13 and quits the writing in response to the sound-recording interruption signal.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention comprises, for example, a voice recording / reproducing apparatus for recording an input voice signal. By reproducing the voice recorded by this voice recording / reproducing apparatus, a voice pattern registered as a characteristic pattern can be reproduced. A voice recognition device that can be confirmed.

[0002]

2. Description of the Related Art As is well known, a voice recognition device converts an input voice signal into a digital voice pattern, detects a voice section from this voice pattern, and uses a voice pattern corresponding to this voice section as a characteristic pattern. It is stored in the characteristic pattern storage unit. In this voice recognition device, the detection accuracy of the voice section greatly affects the recognition rate. Therefore, if the voice pattern stored in the characteristic pattern storage unit can be confirmed, the failure of voice pattern registration can be reduced, and the recognition rate can be improved.

Therefore, a voice recording / reproducing device is combined with a voice recognition device to store a voice pattern corresponding to a voice section in a characteristic pattern storage unit of the voice recognition device, and a voice corresponding to the voice pattern stored in this storage unit. An apparatus has been developed in which a pattern is recorded in a voice recording / reproducing apparatus and the recorded voice pattern is reproduced to confirm the voice pattern stored in the characteristic pattern storage unit. Figure 4
Shows a conventional voice recognition device in which a voice recording / reproducing device is combined.

In FIG. 4, a voice input unit 31 formed of, for example, a microphone converts voice into an electric signal and outputs it. The output end of the voice input unit 31 is connected to the voice analysis unit 32. The voice analysis unit 32 has an A / D converter (not shown) and converts the voice signal supplied from the voice input unit 31 into a digital signal to create a voice pattern. The output end of the voice analysis unit 32 is connected to the voice section detection unit 33. This voice section detection unit 33
Uses two thresholds, for example, to detect a voice section in which noise is removed from the voice pattern supplied from the voice analysis unit 32. A characteristic pattern storage unit 34 and a recognition calculation unit 35 are connected to the voice section detection unit 33. The feature pattern storage unit 34 is
The voice pattern corresponding to the detected voice section is stored. Further, the recognition calculation unit 35 calculates the degree of similarity between the voice pattern output from the voice section detection unit 33 and the voice pattern stored in the characteristic pattern storage unit 34.

On the other hand, in the voice recording / reproducing apparatus 36, the input end of the voice recording control unit 37 is connected to the output end of the voice input unit 31, and the output end is connected to the voice pattern recording unit 38 and the voice reproducing unit 39. ing. The voice recording control unit 37 has an A / D converter (not shown) and a so-called voice trigger mechanism, and converts the voice signal supplied from the voice input unit 31 into a digital signal by the A / D converter. At the same time as generating the pattern, the voice trigger mechanism detects the beginning (starting end) of the audio signal. As a result, when the beginning of the audio signal is detected, the subsequent audio patterns are output to the audio pattern storage unit 38. The voice pattern storage unit 38 is, for example, a RAM (Random Acc
ESS Memory) and records the voice pattern supplied from the voice recording controller 37. The voice reproduction unit 39 has a D / A converter (not shown), converts the voice pattern read from the voice pattern recording unit 38 into an analog signal under the control of the voice recording control unit 37, and outputs the voice signal. Output.

[0006]

The voice memory control unit 3
7 detects the start of the voice pattern by the voice trigger mechanism as described above, and automatically starts recording the voice pattern. This voice trigger mechanism has a low-level threshold (L) and a high-level threshold (H), and uses these thresholds to determine the voice section. It could not be detected accurately.

That is, FIG. 5 (a) shows a case in which the recording request signal is not generated corresponding to the low level threshold value (L) at the beginning of the voice section, so that the recording request signal is not generated. There is. Therefore, in this case, the beginning of the voice is not recorded. Further, in FIG. 5B, there is noise before the voice section, and this noise exceeds the low level threshold value (L) and further exceeds the high level threshold value (H). A signal is generated and only noise is recorded.

As described above, the voice pattern recorded by using the conventional voice recording control unit 37 may lack the start end or may include noise, so that the voice section cannot be accurately recorded. It was a thing.

In addition, conventionally, the voice section detection unit 33 and the voice recording control unit 37 separately detect the voice patterns and store them separately in the characteristic pattern storage unit 34 and the voice pattern storage unit 38. Therefore, the voice pattern stored in the characteristic pattern storage unit 34 and the voice pattern stored in the voice pattern storage unit 38 do not match in the voice section, so that the voice pattern stored in the feature pattern storage unit is accurately confirmed. It was impossible.

The present invention has been made to solve the above problems, and an object of the present invention is to reliably record only a voice pattern without lacking the beginning of the voice pattern or recording only noise. Yes, you can match the voice section of this recorded voice pattern with the feature pattern,
An object of the present invention is to provide a voice recognition device capable of surely confirming a voice pattern stored as a characteristic pattern by reproducing a recorded voice pattern and improving a recognition rate.

[0011]

In order to solve the above problems, the present invention provides a voice analysis unit for generating a voice pattern composed of a digital signal corresponding to an input voice, a low level threshold and a high level threshold. If the set voice pattern generated by the voice analysis unit exceeds the high level threshold within the first time period after the voice pattern exceeds the low level threshold, it corresponds to the time when the voice pattern exceeds the low level threshold. And outputs a recording request signal, and outputs a recording interruption signal when the voice pattern falls below the low level threshold within a second time longer than the first time after exceeding the high level threshold. A voice section detecting unit that detects a section as a voice section after the voice pattern exceeds a high level threshold and after the second time elapses until the voice pattern falls below the low level threshold, and the detected voice section. Storage means for storing a voice pattern corresponding to the voice pattern, a calculation section for calculating the similarity between the voice pattern stored in the storage means and the voice pattern output from the voice section detection section, and the voice section detection section. Recording means for recording the voice pattern in response to the output recording request signal and stopping recording of the voice pattern in response to the recording interruption signal; and reproducing means for reproducing the voice pattern recorded by the recording means. It is equipped with.

Further, the recording means writes the voice pattern in the storage portion for storing the voice pattern and the recording request signal output from the voice section detection portion, and in response to the recording interruption signal. A writing circuit for stopping the writing of the voice pattern, a reading circuit for reading the voice pattern stored in the storage section, and a reproducing means for converting the read voice pattern into voice and outputting the voice. There is.

[0013]

That is, in the present invention, when the voice pattern supplied from the voice analysis unit exceeds the high level threshold within the first time after the voice pattern supplied from the voice analysis unit exceeds the low level threshold, the low level is detected. When a recording request signal is output in response to the time when the sound level exceeds the threshold value of, and the sound pattern falls below the low level threshold value within the second time longer than the first time after exceeding the high level threshold value. , Output the recording interruption signal.
Furthermore, when the voice pattern exceeds the high level threshold within the first time after the voice pattern exceeds the low level threshold, and further falls below the low level threshold after the second time, Is detected as a voice section, and the storage means stores the voice pattern corresponding to the detected voice section as a characteristic pattern. Further, the recording control means records a voice pattern in accordance with the recording request signal output from the voice section detection unit, and if the recording request signal is output from the voice section detection unit during the recording, it records until that time. Stop the recording by treating the voice pattern as noise.
Therefore, it is possible to reliably record only the voice pattern without missing the beginning of the voice pattern or recording only the noise. Moreover, since the voice section of the voice pattern stored in the storage means and the voice section of the voice pattern recorded by the recording control means match each other, the voice pattern stored in the storage means is reproduced by reproducing the recorded voice pattern. You can check the pattern.

[0014]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below with reference to the drawings.

The voice input unit 11 composed of, for example, a microphone converts voice into an electric signal and outputs it. The output end of the voice input unit 11 is connected to the voice analysis unit 12. The voice analysis unit 12 is used in the A / D converter 1
2a and converts a voice signal supplied from the voice input unit 11 into a digital signal to create a voice pattern. The output end of the voice analysis unit 12 is connected to the voice section detection unit 13. The voice section detection unit 13 is provided with a microprocessor 13a and a memory 13b. The memory 13b has two programs, which are a program necessary for the operation of the microprocessor 13a, a low level threshold L set based on ambient noise picked up by the voice input unit 11, and a high level threshold H. The threshold value and the like are stored in advance. Also, the microprocessor 1
3a detects a voice section from the voice pattern supplied from the voice analysis unit 12 by an operation described later, and generates a recording request signal or a recording interruption signal.

A feature pattern storage unit 14, a recognition calculation unit 15, and a recording control unit 16 are connected to the voice section detection unit 13. The characteristic pattern storage unit 14 stores a voice pattern corresponding to the detected voice section in the voice section detection unit 13, and is configured by, for example, a RAM. Further, the recognition calculation unit 15 calculates the degree of similarity between the voice pattern output from the voice section detection unit 13 and the voice pattern stored in the characteristic pattern storage unit 14. The input end of the recording control unit 16 is connected to the output end of the voice input unit 11, and the output end is connected to the voice pattern storage unit 17 and the voice reproducing unit 18.

The recording controller 16 includes an A / D converter 16
a, a writing circuit 16b, and a reading circuit 16c are provided. The A / D converter 16a includes the voice input unit 11
The voice signal supplied from the device is converted into a digital signal to generate a voice pattern. The writing circuit 16b writes the voice pattern output from the A / D converter 16a in the voice pattern storage unit 17 in response to the recording request signal supplied from the voice section detection unit 13, or the recording interruption signal. The writing of the voice pattern is controlled to be stopped according to. Further, the read circuit 16c sequentially reads the audio patterns stored in the audio pattern storage unit 17 according to a read signal supplied from the outside and supplies the audio patterns to the audio reproduction unit 18.
The voice pattern storage unit 17 is composed of, for example, a RAM, and stores the voice pattern supplied from the storage control unit 16. The voice reproduction unit 19 has a D / A converter 18a and the like, and converts the voice pattern supplied from the voice pattern storage unit 17 into an analog signal under the control of the storage control unit 16 and outputs the voice signal. To do. next,
The operation of the microprocessor 13a constituting the voice section detection unit 13 will be further described with reference to FIGS.

First, the microprocessor 13a sequentially takes in the voice patterns output from the voice analysis unit 12 into the characteristic pattern storage unit 14, and determines whether or not the voice pattern exceeds a low level threshold L (ST1, 2).
As a result, when the threshold L of the low level is exceeded, it is determined whether the threshold H of the high level is exceeded within a predetermined short time, for example, 20 ms (ST3). As a result, when the threshold value H of the high level is exceeded within a predetermined short time, the time point of exceeding the threshold value L of the low level is regarded as the beginning (starting end) of the voice pattern, and the time point of exceeding the threshold value L of the low level is used as a reference. A recording request signal is output as (ST4).

However, noise has a characteristic that it exceeds the high level threshold value H only for a short time. Therefore, in order to remove noise, it is determined whether or not the threshold value L has become equal to or lower than the low level threshold L within a predetermined time longer than the predetermined short time, for example, 1 s after the start of the voice is detected (ST5,
6) If the threshold value L becomes a low level or less within a predetermined time, the voice pattern during this period is regarded as noise and a recording interruption signal is output (ST7).

On the other hand, when it is determined in step 5 that the predetermined time has elapsed after outputting the recording request signal, it is determined whether or not the voice pattern is below the low level threshold L (ST8). ), If it is below the low level threshold L, it is considered that the voice section has ended, and the recording request signal is stopped (ST9). That is, the recording request signal is output corresponding to the voice section, and the characteristic pattern storage unit 1
In 4, the voice pattern of the range regarded as the voice section by the voice section detection unit 13 is stored.

The writing circuit 16b, which constitutes the recording control unit 16, receives the recording request signal from the voice section detection unit 13 and outputs the voice pattern output from the A / D converter 16a to the voice pattern storage unit 17. Write in sequence, that is, record a voice pattern. This writing circuit 16b
When the recording stop signal is output from the voice section detector 13 during the writing, the writing is stopped and the writing address is cleared. After that, when the recording interruption signal is stopped, the writing circuit 16a sequentially writes the audio patterns output from the A / D converter 16a in the audio pattern storage unit 17 again in response to the recording request signal. After this,
When the recording request signal output from the voice section detection unit 13 stops, the writing circuit 16a stops writing the voice pattern.

On the other hand, when a read signal is externally supplied to the sound recording control unit 16, the read circuit 16c sequentially reads the sound patterns stored in the sound pattern storage unit 17 and supplies the sound patterns to the sound reproducing unit 18. This voice reproduction unit 18
Then, the supplied voice pattern is output as voice converted into an analog signal by the D / A converter 18a.

According to the above embodiment, the voice section detecting unit 13
Outputs a recording request signal or a recording stop signal according to the level of the voice pattern, and the recording control unit 16 causes the voice section detection unit 1
A voice pattern is stored in the voice pattern storage unit 17 based on a recording request signal or a recording interruption signal output from the recording unit 3, or the storage is stopped. Therefore, only the noise is not stored in the voice pattern storage unit 17, and the start end of the voice pattern is not stored.

In addition, the voice pattern stored in the voice pattern storage unit 17 and the voice pattern stored in the characteristic pattern storage unit 14 are in the same voice section. Therefore, by reproducing the voice pattern stored in the voice pattern storage unit 17, the voice pattern stored in the characteristic pattern storage unit 14 can be confirmed.

In the above embodiment, the storage controller 16
Has an A / D converter 16a, converts the audio signal supplied from the audio input unit 11 into a digital signal, and writes the digital signal in the audio pattern storage unit 17 by the writing circuit 16. However, the present invention is not limited to this. , Without A / D converter,
The voice pattern supplied from the voice analysis unit 12 may be written in the voice pattern storage unit 17 by the writing circuit 16. Besides, the present invention is not limited to the above-described embodiments, and it goes without saying that various modifications can be made without departing from the spirit of the invention.

[0026]

As described above in detail, according to the present invention, the voice section detection unit outputs the recording request signal and the recording stop signal according to the voice pattern and the low level threshold and the high level threshold, and The voice section is detected in response to the recording request signal or the recording stop signal, and the recording control unit records a voice pattern or stops recording based on the recording request signal or the recording interruption signal. Therefore, it is possible to reliably record only the voice pattern without missing the beginning of the voice pattern and to match the recorded voice pattern with the feature pattern of the voice stored in the feature storage unit. By playing
A voice recognition device capable of confirming the voice pattern stored in the characteristic pattern storage unit and improving the recognition rate can be provided.

[Brief description of drawings]

FIG. 1 is a configuration diagram showing an embodiment of an apparatus of the present invention.

2 is a flowchart shown to explain the operation of FIG. 1. FIG.

3 is a waveform diagram shown to explain the operation of FIG. 1. FIG.

FIG. 4 is a configuration diagram showing a conventional voice recognition device.

FIG. 5 is a waveform diagram shown to explain the operation of FIG.

[Explanation of symbols]

11 ... Voice input section, 12 ... Voice analysis section, 13 ... Voice section detection section, 14 ... Feature pattern storage section, 15 ... Recognition calculation section, 16 ... Storage control section, 17 ... Voice pattern storage section, 1
8 ... Voice reproduction unit.

Claims (2)

[Claims] 1. A voice analysis unit for generating a voice pattern composed of a digital signal corresponding to an input voice, a low level threshold and a high level threshold are set, and the voice pattern generated by the voice analysis unit is set to When the high level threshold is exceeded within a first time after the low level threshold is exceeded, a recording request signal is output corresponding to the time when the low level threshold is exceeded, and the voice pattern is high level. After the threshold value is exceeded, if the low level threshold value is reached within the second time period that is longer than the first time period, a recording interruption signal is output, and after the voice pattern exceeds the high level threshold value, A voice section detection unit that detects a section until the threshold value becomes a low level threshold or less after a second time elapses, and a voice pattern corresponding to the detected voice section is stored as a characteristic pattern. A storage unit, a calculation unit that calculates a similarity between the voice pattern stored in the storage unit and the voice pattern output from the voice section detection unit, and a recording request signal output from the voice section detection unit. A recording means for recording the voice pattern and stopping the recording of the voice pattern in response to the recording interruption signal; and a reproducing means for reproducing the voice pattern recorded by the recording means. Voice recognition device. 2. The recording means writes the voice pattern in the storage portion in response to a recording request signal output from the voice section detection portion and a storage portion for storing the voice pattern, and in response to the recording interruption signal. A writing circuit for stopping the writing of the voice pattern, a reading circuit for reading the voice pattern stored in the storage section, and a reproducing means for converting the read voice pattern into voice and outputting the voice. The voice recognition device according to claim 1, wherein