JPH0628830A - Speech recognizing device - Google Patents

Abstract

PURPOSE:To speedily retrieve the recorded content of a tape. CONSTITUTION:When a key word is inputted from a keyboard 7 through a key word voice input part 11 or a key word preparing part 8, the key word is supplied through a selective processing part 9 to a key word processing part 10. When the recording position of the key word is preliminarily recorded in the header of a tape 4, the position preliminarily recorded in the header of the tape 4 is referred by the key word processing part 10, and the same key word as the key word supplied from the selective processing part 9 is retrieved. When the same key word as the key word supplied from the selective processing part 9 is present among the key words preliminarily recorded in the header of the tape 4, the recording position of the key word preliminarily recorded in the tape 4 is referred by the key word processing part 10, and the reproduction of the tape 4 is started from the position.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a voice recognition device suitable for use in searching the contents of voice recorded on a tape or a disk.

[0002]

2. Description of the Related Art For example, in a tape recorder,
The state of the meeting can be recorded on tape. At this time, the index signal can be recorded at an arbitrary position as needed. As a result, it is possible to retrieve this index signal from an arbitrary position on the tape and reproduce it from that position.

[0003]

However, in the conventional apparatus, for example, when it is desired to confirm a part relating to a predetermined item among the items discussed at the meeting, all the index signals are sequentially detected from the beginning of the tape and the partial signal is partially detected. However, there is a problem that not only the user feels troublesome but also it takes time to search the position. In addition, when the index signal is not recorded in advance at that position, reproduction must be performed from the start point to the end point.

The present invention has been made in view of such a situation, and makes it possible to quickly search a predetermined recording content.

[0005]

According to a first aspect of the present invention, there is provided a voice recognition device, which comprises a voice input section 1 as voice input means for inputting voice and a tape 4 as a recording medium for recording the input voice. A keyboard 7 or a keyword voice input section 11 as a keyword input means for inputting a keyword, a standard pattern memory 24 as a storage means for storing the input keyword, and a voice recorded on the tape 4 (voice input section 1 From the voice recorded on the tape 4 or the voice reproduced from the tape 4), and a voice recognition unit 3 as a keyword extraction unit for extracting a keyword stored in the standard pattern memory 24. The position where the keyword is recorded on the tape 4 is detected, and the keyword is recorded at a predetermined position on the tape 4 such as a header. Characterized in that it comprises a keyword processing section 10 as a keyword processing means for recording.

In the voice recognition device according to a second aspect of the present invention, the keyboard 7 or the keyword voice input unit 11 inputs a keyword by characters or voice.

In the voice recognition device according to the third aspect, the standard pattern memory 2 as a storage means for storing the keyword is provided.
4, a voice including a keyword is recorded, and the recording position of the keyword included in the voice is recorded in a predetermined position such as a header as a tape 4 and a standard pattern memory 24. And a keyword processing unit 10 as a search means for searching the tape 4 for a voice including the keyword.

[0008]

In the voice recognition device according to the first aspect of the present invention, the keyword contained in the voice recorded on the tape 4 or the voice reproduced from the tape 4 is extracted, and the extracted keyword indicates the position recorded on the tape 4. It is detected and recorded in the header of the tape 4 together with the keyword. Therefore, referring to the header of the tape 4, at the position where the keyword is recorded,
It can be accessed quickly.

In the voice recognition device according to the second aspect, the keyboard 7 or the keyword voice input unit 11 is
Since the keyword can be input by letters or voice, the keyword can be confirmed by letters or voice.

In the voice recognition apparatus according to the third aspect, the keyword is stored in the standard pattern memory 24. Then, the voice including the keyword stored in the standard pattern memory 24 is searched from the tape 4 in which the voice including the keyword is recorded and the recording position of the keyword included in the voice is recorded. Therefore, the predetermined contents recorded on the tape 4 can be confirmed quickly.

[0011]

DESCRIPTION OF THE PREFERRED EMBODIMENTS FIG. 1 is a block diagram showing the configuration of an embodiment of a voice recognition device of the present invention. The voice input unit 1 outputs a signal obtained by A / D converting the input voice, that is, a digitized voice signal to the voice recording unit 2 and the voice recognition unit 3. The audio recording unit 2 records the audio signal output from the audio input unit 1 on the tape 4. The audio reproducing unit 5 is a tape 4
The audio signal recorded in is reproduced and supplied to the audio output unit 6 and the audio recognition unit 3. The audio output unit 6 is the audio reproduction unit 5
Alternatively, the voice signal supplied from the keyword processing unit 10 is converted into voice and output.

The keyword voice input unit 11 supplies the input voice as a keyword to the selection processing unit 9. The keyboard 7 is for typing a keyword as a character (character string), and the typed character is output to the keyword creating unit 8. The keyword creating unit 8 creates a keyword as a voice signal from the keyword as a character output from the keyboard 7 and supplies it to the selection processing unit 9. The selection processing unit 9 selects either the keyword supplied from the keyword voice input unit 11 or the keyword supplied from the keyword creation unit 8, and the voice recognition unit 3
And output to the keyword processing unit 10.

As shown in FIG. 2, the voice recognition section 3 is composed of an acoustic analysis section 21, an input pattern memory 22, a DP matching section 23, a standard pattern memory 24, and a word spotting determination section 25. Acoustic analysis unit 21
When a voice signal as a keyword is supplied from the selection processing unit 9, the keyword (voice signal) is subjected to frequency analysis processing such as linear prediction analysis processing to calculate a voice characteristic parameter (feature parameter series). , To the standard pattern memory 24. Standard pattern memory 24
Stores the characteristic parameter calculated from the keyword supplied from the acoustic analysis unit 21 as a standard pattern.

In addition, when the audio signal is supplied from the audio input unit 1 or the audio signal reproduced from the tape 4 by the audio reproducing unit 5 is supplied to the acoustic analysis unit 21,
A linear prediction analysis process is performed on the voice signal to calculate a voice feature parameter (feature parameter series), and the voice feature parameter is supplied to the input pattern memory 22. The input pattern memory 22 temporarily stores the characteristic parameter of the voice supplied from the acoustic analysis unit 21 as an input pattern.

The DP matching unit 23 calculates the pattern distance between the input pattern temporarily stored in the input pattern memory 22 and the standard pattern stored in the standard pattern memory 24 by DP matching. That is, the DP matching unit 23 sets one of the time axis of the standard pattern or the time axis of the input pattern, for example, the time axis of the standard pattern, so that the time axis of the other pattern is the most similar (the standard pattern). And the input pattern) so that the distance between the standard pattern and the input pattern (time-normalized distance) is artificially distorted. Then, the DP matching unit 23, together with the distance between the standard pattern and the input pattern, the position at which the time axis of the standard pattern corresponds to the time axis of the input pattern when the distance is obtained, that is, the standard in the input pattern. The position of the pattern is supplied to the word spotting determination unit 25.

The DP matching unit 23 performs the above processing on all the standard patterns stored in the standard pattern memory 24, and determines the minimum distance in all the standard patterns and the position where the minimum distance is obtained. Obtained and supplied to the word spotting determination unit 25.

The word spotting determination unit 25 uses the DP
From the minimum distances in all the standard patterns supplied from the matching unit 23 and the positions where they are obtained,
All the standard patterns included in the input pattern are extracted (recognized), and the keywords for the standard patterns are output to the keyword processing unit 10 (FIG. 1).

The keyword processing unit 10 includes a voice recognition unit 3
The position (recording position) where the keyword output from the (word spotting determination unit 25) is recorded on the tape 4 is detected, and is recorded together with the keyword on, for example, the header of the tape 4. Further, the keyword processing unit 10 uses the tape 4
The keyword recorded in the header of is read and supplied to the voice reproducing unit 5 or the display unit 12. The display unit 12 displays the keyword supplied from the keyword processing unit 10 as an image.

Next, the operation for registering a keyword will be described. The keyword input from the keyboard 7 via the keyword voice input unit 11 or the keyword creation unit 8 is supplied to the acoustic analysis unit 21 and the keyword processing unit 10 of the voice recognition unit 3 via the selection processing unit 9.
In the keyword processing unit 10, the keyword supplied from the selection processing unit 9 is supplied to the voice output unit 6 via the display unit 12 or the voice reproduction unit 5 for input confirmation. In the display unit 12, the keyword input from the keyboard 7 via the keyword creating unit 8 is displayed as an image (character), and in the voice output unit 6, the keyword input from the keyword voice input unit 11 is output as voice. .

It should be noted that one keyword is used for the keyboard 7
If both are input with the keyword voice input unit 11, the keyword input by voice (the keyword input by the keyword voice input unit 11) can be confirmed as characters, and the keyword input by characters ( A keyword input from the keyboard 7 can be confirmed by voice.

On the other hand, in the acoustic analysis section 21 of the speech recognition section 3, the speech signal serving as the keyword supplied from the selection processing section 9 is subjected to linear prediction analysis to calculate the characteristic parameter (characteristic parameter series) of the speech, and the standard is calculated. It is supplied to the pattern memory 24. In the standard pattern memory 24, the characteristic parameter calculated by the keyword supplied from the acoustic analysis unit 21 is stored as a standard pattern.

Further, the operation of recording the keyword on the tape 4 will be described. In the voice input unit 1,
The audio signal is A / D converted and supplied to the audio recording unit 2 and the audio recognition unit 3. In the audio recording unit 2, the audio signal supplied from the audio input unit 1 is recorded on the tape 4.

On the other hand, the speech signal supplied to the speech recognition section 3 is subjected to linear prediction analysis processing in the acoustic analysis section 21 to calculate a characteristic parameter (characteristic parameter series) of speech. Then, in the input pattern memory 22, the characteristic parameter of the voice supplied from the acoustic analysis unit 21 is temporarily stored as an input pattern.

In the DP matching section 23, the pattern distance between the input pattern temporarily stored in the input pattern memory 22 and the standard pattern stored in the standard pattern memory 24 is calculated by DP matching. That is, in the DP matching unit 23, one of the time axis of the standard pattern and the time axis of the input pattern, for example, the time axis of the standard pattern, is artificially set so as to be most similar to the input pattern which is the other pattern. After distortion, the distance between the standard pattern and the input pattern (time-normalized distance) is obtained. Then, in the DP matching unit 23, together with the distance between the standard pattern and the input pattern,
The position where the time axis of the standard pattern corresponds to the time axis of the input pattern when the distance is obtained, that is, the position of the standard pattern in the input pattern is obtained and supplied to the word spotting determination unit 25.

In the DP matching unit 23, all the standard patterns stored in the standard pattern memory 24 are subjected to the above processing, and the minimum distance in all the standard patterns and the position where the minimum distance is obtained. Is calculated and supplied to the word spotting determination unit 25.

In the word spotting determination unit 25, all the standard patterns included in the input pattern are extracted (recognized) from the minimum distance in all the standard patterns supplied from the DP matching unit 23 and the position where the minimum distance is obtained. Then, the keyword for the standard pattern is output to the keyword processing unit 10 (FIG. 1).

The keyword processing unit 10 detects the position where the keyword output from the voice recognition unit 3 (word spotting determination unit 25) is recorded on the tape 4,
It is recorded in the header of the tape 4 together with the keyword.

In addition to the above, the keywords can be recorded on the tape 4 as follows. That is, in the voice input unit 1, the voice signal is A / D converted,
It is supplied to the voice recording unit 2. In the audio recording unit 2, the audio signal supplied from the audio input unit 1 is recorded on the tape 4. After the recording of the audio signal is completed, the audio signal recorded on the tape 4 is reproduced by the audio reproducing unit 5 and supplied to the voice recognizing unit 3. Acoustic analysis unit 2 of voice recognition unit 3
In 1, the audio signal recorded on the tape 4 is subjected to the linear predictive analysis processing, and the audio characteristic parameter (characteristic parameter series) is calculated. Then, the input pattern memory 22
At, the characteristic parameter of the voice supplied from the acoustic analysis unit 21 is temporarily stored as an input pattern.

After that, the voice recognition unit 3 and the keyword processing unit 10 perform the above-described processing for recording the keyword on the tape 4, and the keyword and the recording position of the keyword are recorded in the header of the tape 4.

Next, when reproducing the audio signal recorded on the tape 4, when a keyword is input from the keyboard 7 through the keyword audio input unit 11 or the keyword creating unit 8, the keyword is input through the selection processing unit 9. Are supplied to the keyword processing unit 10. Keyword processing unit 10
In, the keyword supplied from the selection processing unit 9 is
It is supplied to the audio output unit 6 via the display unit 12 or the audio reproduction unit 5 for input confirmation. In the display unit 12, the keyword input from the keyboard 7 via the keyword creating unit 8 is displayed as an image, and in the voice output unit 6, the keyword input from the keyword voice input unit 11 is output as voice.

When the keyword and the recording position of the keyword are recorded in the header of the tape 4 in advance, the keyword recorded in the header of the tape 4 is referred to in the keyword processing unit 10 and supplied from the selection processing unit 9. The same keyword as the searched keyword is searched. If the same keyword as the keyword supplied from the selection processing unit 9 is included in the keywords recorded in the header of the tape 4 in advance, the keyword processing unit 10 records the keyword of the keyword recorded in the header of the tape 4 in advance. The recording position is read, the position is further searched, and the reproduction of the tape 4 is started from that position. Further, if the position of the start point of the voice including the keyword is recorded in the header, the voice including the keyword can be reproduced from the start point.

When the keyword and the recording position of the keyword are not recorded in the header of the tape 4 in advance, the audio reproduction unit 5 starts reproduction of the audio signal recorded on the tape 4 and supplies it to the audio recognition unit 3. To be done.
A linear prediction analysis process is performed in the acoustic analysis unit 21 of the voice recognition unit 3, and a voice characteristic parameter (feature parameter series)
Is calculated. Then, in the input pattern memory 22, the characteristic parameter of the voice supplied from the acoustic analysis unit 21 is temporarily stored as an input pattern.

In the DP matching unit 23, one of the time axis of the standard pattern stored in the standard pattern memory 24 or the time axis of the input pattern stored in the input pattern memory 22, for example, the time axis of the standard pattern, The distance between the standard pattern and the input pattern (distance normalized by time) is obtained by artificially distorting it so that it is most similar to the other pattern, the input pattern. Then, in the DP matching unit 23, along with the distance between the standard pattern and the input pattern, when the distance is obtained, the time axis of the standard pattern corresponds to the time axis of the input pattern, that is, the standard in the input pattern. The position of the pattern is obtained and supplied to the word spotting determination unit 25.

In the DP matching unit 23, all the standard patterns stored in the standard pattern memory 24 are subjected to the above processing, and the minimum distance in all the standard patterns and the position where the minimum distance is obtained. Is calculated and supplied to the word spotting determination unit 25.

In the word spotting determination section 25, all the standard patterns included in the input pattern are extracted (recognized) from the minimum distance in all the standard patterns supplied from the DP matching section 23 and the position where the minimum distance is obtained. Then, the keyword for the standard pattern is output to the keyword processing unit 10 (FIG. 1).

In the keyword processing unit 10, the keyword supplied from the voice recognition unit 3 (word spotting determination unit 25) is supplied to the voice output unit 6 via the display unit 12 or the voice reproduction unit 5. The keywords included in the audio signal recorded on the tape 4 are displayed as images on the display unit 12, and the keywords included in the audio signal recorded on the tape 4 are output as audio on the audio output unit 6.

[0037] Therefore, the tape 4, is for example, "Hello, Tokyo Sony Taro. Simply send your answer on the other day of inquiries. Our Walkman We now will 8 type released. Models that meet your requirements in this Is WM-EX70, 19,500 yen, please purchase it at a nearby electronics store that sells Sony products.
Please request if you want a detailed catalog. We will send it to you. We hurried and responded. Is recorded in the standard pattern memory 24, for example, “Tokyo, Sony, reply, walkman, type, circle, request, model, product, purchase, electronics store, use, catalog, hope, Billing 1,
When keywords such as “2, 3, 4, 5, 6, 7, 8, 9, 0” are registered, “Tokyo, Sony, Answer, Walkman, 8, Type, Request, Model, 70, 19, 50”
The keywords “0, Yen, Sony, Product, Electronic store, Usage, Catalog, Hope, Claim, Reply” are output from the voice output unit 6 or the display unit 12. From this keyword, the outline of the content of the voice (message) can be inferred.

The voice recognition apparatus of the present invention has been described above. However, if the present invention is applied to, for example, a voice mail system, the contents of a voice mail sent can be confirmed quickly.

In this embodiment, the tape is used as the recording medium for recording the audio signal. However, if the audio signal is recorded on a disc which can be randomly accessed, such as an optical disc or a magneto-optical disc, The keyword can be searched more quickly.

[0040]

According to the voice recognition device of the first aspect, the keyword contained in the voice recorded in the recording medium or the voice reproduced from the recording medium is extracted, and the extracted keyword is recorded in the recording medium. The detected position is detected and recorded together with the keyword at a predetermined position on the recording medium. Therefore, it is possible to quickly access the position where the keyword is recorded by referring to the predetermined position on the recording medium.

According to the voice recognition device of the second aspect,
The keyword input means inputs a keyword by characters or voice. Therefore, the device can be used according to the preference of the user.

According to the voice recognition device of the third aspect,
Keywords are stored in the storage means. Then, the voice including the keyword is recorded, and the voice including the keyword stored in the storage unit is searched from the recording medium in which the recording position of the keyword included in the voice is recorded. Therefore, it is possible to quickly confirm the predetermined content recorded on the recording medium.

[Brief description of drawings]

FIG. 1 is a block diagram showing the configuration of an embodiment of a voice recognition device of the present invention.

FIG. 2 is a more detailed block diagram of a voice recognition unit 3 of the embodiment of FIG.

[Explanation of symbols]

 1 voice input unit 2 voice recording unit 3 voice recognition unit 4 tape 5 voice reproduction unit 6 voice output unit 7 keyboard 8 keyword creation unit 9 selection processing unit 10 keyword processing unit 11 keyword voice input unit 12 display unit 21 acoustic analysis unit 22 input Pattern memory 23 DP matching unit 24 Standard pattern memory 25 Word spotting determination unit

 ─────────────────────────────────────────────────── ─── Continuation of the front page (72) Inventor Yuko Tanaka 6-735 Kita-Shinagawa, Shinagawa-ku, Tokyo Inside Sony Corporation (72) Inventor Yasuhiko Kato 6-7-35 Kita-Shinagawa, Shinagawa-ku, Tokyo Sony Corporation (72) Inventor Kazuo Ishii 6-735 Kitashinagawa, Shinagawa-ku, Tokyo Sony Corporation

Claims (3)

[Claims] 1. A voice input unit for inputting voice, a recording medium for recording the voice input by the voice input unit, a keyword input unit for inputting a keyword, and a keyword input by the keyword input unit. Storing means for storing the keyword, keyword extracting means for extracting the keyword stored in the storing means from the voice recorded in the recording medium, and the keyword extracted by the keyword extracting means is recorded in the recording medium. A voice recognition device, comprising: a keyword processing means for detecting a position and recording the keyword at a predetermined position on the recording medium together with the keyword. 2. The keyword input means inputs the keyword by characters or voice.
The voice recognition device described in. 3. A storage means for storing a keyword, a voice including the keyword is recorded, and
A recording medium in which a recording position of the keyword included in the voice is recorded at a predetermined position; and a searching unit that searches the recording medium for a voice containing the keyword stored in the storage unit. And a voice recognition device.