JP2019090917A - Voice-to-text conversion device, method and computer program - Google Patents

Abstract

To provide accurate text data without imposing a burden on an engine for converting voice data into text data in processing of converting voice data into a text.SOLUTION: A voice data-to-text conversion device 1 divides voice data by prescribed unit time with a division processing section 12 and generates compressed data obtained by deleting a silent part from divided data by a compression processing section 13. When pre-connection text data obtained by converting the compressed data into a text is received from a voice-to-text conversion engine after the compressed data is transmitted to the voice-to-text conversion engine, a connection processing section 15 generates post-connection text data obtained by connecting the pre-connection text data in a pre-division order.SELECTED DRAWING: Figure 1

Description

  The present invention relates to a technology for converting speech data into text.

In recent years, the demand for services for converting speech data into text has been increasing. Although voice data is created in the creation of the minutes, text data is more convenient for later confirmation than voice data.
With regard to this point, for example, Patent Document 1 proposes a method of simultaneously digitizing call speech and distributing data to a plurality of callers who are in voice communication.
Further, in Patent Documents 2 and 3, an apparatus for converting voice data into text data by an operator is proposed.

Unexamined-Japanese-Patent No. 2010-43101 JP 2008-9693A JP, 2013-182353, A

  However, in the technique described in the above-mentioned patent document, since the voice data including the silent part is converted into text, an unnecessary processing load is imposed on the device. That is, although the silent part is not textified, applying textification processing will execute useless processing.

  On the other hand, when long-time voice data or the like in which the data size tends to be large is divided and converted into text as needed, incorrect conversion may be performed at divided positions, and accurate conversion into text may not be possible.

  Therefore, the present invention has an object of obtaining accurate text data without putting a load on an engine for converting speech data into text data in text conversion processing of speech data.

  In order to achieve the above object, an audio data text converting apparatus according to the present invention is an apparatus for converting audio data into text, and can communicate with an audio text converting engine which converts the audio data into text through a network A division processing unit configured to divide the audio data by a predetermined unit time to generate divided data; a compression processing unit configured to generate compressed data in which a silent portion is deleted from the divided data; A compressed data transmission unit for transmitting the compressed data, a pre-combination text data receiving unit for receiving the pre-combination text data obtained by converting the compressed data into text data from the speech-to-text engine, the pre-combination text data And combining processing means for generating text data after combining in the order before division. And butterflies.

  Further, the dividing means divides the audio data by the silent portion to generate divided data when the sectioning position for each predetermined unit time from the start position of the audio data is a silent portion, and the audio data is generated. When the demarcation position for each predetermined unit time from the start position of is a talkative part, the speech data is divided by a silent part within a certain time before the talkative part to generate divided data It may be

  The image processing apparatus further includes volume adjustment means for adjusting the volume of the audio data to a predetermined volume, and the division means divides the audio data adjusted to a predetermined volume by a predetermined unit time to generate divided data. It is also possible to

  Further, the information processing apparatus further comprises identification information issuing means for issuing identification information to the compressed data, and the combination processing means combines the pre-combination text data in the order before division based on the identification information. Text data may be generated.

  Further, it is configured to be communicable with the user terminal used by the user via the network, and the voice data receiving means for receiving the voice data from the user terminal, and the post-join text data to the user terminal. And a post-combination text data transmission means for transmitting.

  Further, according to another aspect of the present invention, there is provided an audio data text conversion method for converting audio data into text data, the method comprising: communication with an audio text conversion engine for converting the audio data into text data via a network A division process of dividing the audio data by a predetermined unit time to generate divided data, a compression process of generating compressed data in which silent parts are deleted from the divided data, and the speech-to-text engine; A compressed data transmission process for transmitting the compressed data, a pre-combination text data reception process for receiving the pre-combination text data obtained by converting the compressed data into a text form from the speech-to-text engine, the pre-combination text data Execute combining processing to generate text data after combining combining in the order before division And features.

  A computer program according to another aspect of the present invention is a computer program for converting voice data into text, and is configured to be communicable with a voice-to-text engine that converts the voice data into text via a network. A division process of dividing the voice data by a predetermined unit time to generate divided data, a compression process of generating compressed data obtained by deleting silent parts from the divided data, and the voice-to-text engine A compressed data transmission process for transmitting the compressed data, a pre-combination text data reception process for receiving the pre-combination text data obtained by converting the compressed data into a text from the speech-to-text engine, the pre-combination text data A combining process for generating post-join text data combined in the order before division To the execution.

  According to the speech-to-text device according to the present invention, accurate text data can be obtained without putting a load on an engine for converting speech data into text data in the textification process of speech data.

It is the functional block diagram which showed the function of the speech-to-text device concerning the embodiment of the present invention. FIG. 6 is a diagram showing an example of data stored in an identification information storage unit in the voice-to-text device according to the present embodiment. It is a conceptual diagram explaining the process performed by the audio text conversion apparatus which concerns on this embodiment, (a) When a break position is provided in a silent part, (b) When a break position is provided in a sound part Show. It is a conceptual diagram explaining the process performed by the audio text processing apparatus which concerns on this embodiment, and shows the state by which audio | speech data was divided | segmented in the silent part. It is a conceptual diagram explaining the process performed by the speech-to-text apparatus which concerns on this embodiment, and shows the state in which compression data were produced | generated. It is a sequence diagram which shows the flow of a series of processes performed by the speech-to-text apparatus based on this embodiment. It is the functional block diagram which showed the function with which the user terminal provided with the speech-to-text processing part concerning another embodiment of the present invention is provided. It is a sequence diagram which shows the flow of a series of processes performed by the user terminal provided with the audio text conversion process part which concerns on this embodiment. It is a conceptual diagram explaining the process performed by the user terminal provided with the audio text conversion process part which concerns on this embodiment, (a) State where a break position is set, (b) voice data is divided in a silent part State (c) when compressed data is generated. In another embodiment of the present invention, it is a sequence diagram showing a flow of processing when making voice data into text at any time.

Hereinafter, a speech-to-text device according to an embodiment of the present invention will be described with reference to the drawings.
The voice-to-text device is a device for text-to-speech voice data, and is a device that performs predetermined processing on the voice data and sends processed voice data to the voice-to-text engine to make it into text.

  The voice-to-text device 1 shown in FIG. 1 is configured to be able to communicate with the voice-to-text engine 2 and the user terminal 3 via a network NW such as the Internet.

  The voice-to-text device 1 includes an identification information storage unit 1A, a volume adjustment unit 11, and division processing using an arithmetic device such as a central processing unit (CPU) or a storage device such as a random access memory (RAM) or a read only memory (ROM). A functional block composed of the unit 12, the compression processing unit 13, the identification information issuing unit 14, the combination processing unit 15, and the communication processing unit 16 is configured.

The identification information storage unit 1A is a storage unit that stores identification information issued and added to compressed data by the identification information issuing unit 14, and information related to the compressed data.
For example, as shown in FIG. 2, the identification information storage unit 1A stores the file name, the file size, the creation date, etc. of compressed data in addition to the identification information issued and added for each compressed data described later, as shown in FIG. .
Here, the identification information is information for identifying individual compressed data and making it possible to grasp the coupling order, and is referred to in the coupling processing performed by the coupling processing unit 15.

The volume adjustment unit 11 executes a process of adjusting the volume of audio data to a predetermined volume.
This processing is, for example, to suppress audio data to -10 [db], whereby erroneous conversion is made in the textification processing even if there is a portion with different volume in the audio data before processing You can prevent that.

The division processing unit 12 divides voice data by a predetermined unit time to generate divided data.
In this process, voice data is divided from the start position every predetermined unit time. At this time, when the break position is a silent portion, the voice data is divided at the silent portion, and when the break position is a talkative portion, the sound is silent at a silent portion within a predetermined time before the break position. Divide data to generate divided data.
The division processing unit 12 in this example divides the audio data adjusted to a predetermined volume by the volume adjustment unit 11 to generate divided data.

Here, a specific example of audio data division processing by the division processing unit 12 will be described with reference to FIG.
The audio data 100 is configured by sound parts 101a, 101b, 101c, 101d and 101e, and silent parts 102a, 102b, 102c, 102d and 102e. In the following, for the convenience of description, the sound producing units 101a, 101b, 101c, 101d, and 101e may be collectively referred to as the sound producing unit 101 when not focusing on each component. Similarly, silent sections 102a, 102b, 102c, 102d and 102e may also be referred to as silent sections 102.
Note that the sound producing unit 101 is a portion having a certain volume or more. The silent portion 102 is a silent portion or a portion having a volume less than a predetermined level and regarded as silent. The reference of the volume divided into the sound part 101 and the silent part 102 can be set arbitrarily.

With respect to such voice data, the division processing unit 12 sequentially sets a break position with 50 [sec] as a unit time from the start position of the voice data.
The example of FIG. 3A shows the case where the break position 100a is set to the silent portion 102b. In this case, the audio data 100 is divided with the division position 100a or the silent portion 102b as a division position.
On the other hand, the example of FIG. 3B shows the case where the break position 100b is set to the sound part 101c. In this case, the audio data 100 is divided with the silent portion 102b within 10 [sec] before the separation position 100b as the division position.
As described above, when the division processing unit 12 performs division processing on the audio data 100, as shown in FIG. 4, the audio data 100 is at a predetermined division position (in the example of FIG. 3, the silent portion 102b). A plurality of divided data 110a and 110b are obtained. The divided data 110 b includes audio data for the time from the break position 100 b to the split position 102 b.
By dividing the speech data 100 in this manner, it is possible to prevent unnatural division of a word having a single meaning. As a result, it is possible to prevent an erroneous conversion from being made in the textification process by the speech-to-text engine 2.

The compression processing unit 13 generates compressed data in which the silent portion is deleted from the divided data generated by the division processing unit 12.
The specific example of this process is shown in FIG. 4 and FIG.
FIG. 4 shows a state in which the audio data 100 is divided between the sound producing unit 101 b and the sound producing unit 101 c by the processing that follows the example of FIG. 3. The silent portion 102 b is divided into silent portions 102 b ₁ and 102 b ₂ .
Compression processing section 13 from this state, deleting silence _{102a, 102b 1, 102b 2,} 102c, and 102d as shown in FIG. As a result, the audio data 100 becomes a plurality of compressed data 120a and 120b in which the sound production unit 100 is continuous.

The identification information issuing unit 14 issues and gives identification information to each compressed data generated by the compression processing unit 13.
The identification information is information for identifying each compressed data and making it possible to grasp the order, and is stored in the identification information storage unit 1A and referred to in the combining process performed by the combining process unit 15 Be done.

The combining processing unit 15 executes processing for generating post-combination text data by combining pre-combination text data in the order before division.
Here, when generating the compressed data, the voice-to-text device 1 transmits the compressed data to the voice-to-text engine 2 together with a request for processing the compressed data into a text. When pre-combination text data in which the compressed data is converted into text data is generated in the voice-to-text engine 2, the voice-to-text apparatus 1 receives the pre-combination text data from the voice-to-text engine 2. At this time, the pre-combination text data holds identification information attached to the compressed data before being converted into text, and the combination processing unit 15 refers to the identification information storage unit 1A to perform pre-combination text data In order of voice data and combine them.
As a result, post-combination text data in which speech data is converted into text is generated.

The communication processing unit 16 is a processing unit that executes transmission and reception of various data with the voice-to-text engine 2 and the user terminal 3 via the network NW such as the Internet.
The communication processing unit 16 transmits, for example, compressed data and receives uncombined text data with the voice-to-text engine 2. Further, with the user terminal 3, voice data is received, and text data after combination is transmitted.

The voice-to-text engine 2 is a data processing engine that converts voice data into text.
The speech-to-text engine 2 has functional units such as identification means for identifying words in speech data, speech data of the words to be identified and text data associated with each other, and dictionary means serving as a reference destination of conversion processing. There is.

  The user terminal 3 is a terminal used by a user of an audio data text converting system constituted by an audio data text converting device and an audio data text converting engine, and the user converts predetermined audio data into text by the user terminal 3 Get text data.

  The user terminal 3 is constituted by, for example, a terminal such as a so-called smart phone, a tablet terminal, a personal computer, etc., executes transmission and reception of data via the voice text processing apparatus 1 and the network NW such as the Internet, and inputs and outputs various data. Can be performed.

Subsequently, a flow of a series of processes executed by the speech to text device 1 according to the present embodiment will be described with reference to FIG.
First, the user transmits a request for converting the voice data into a text as well as desired voice data to the voice-to-text device 1 from the user terminal 3 (S101).

In the voice-to-text device 1 having received the voice data, the volume adjuster 11 adjusts the volume of the voice data to a predetermined volume (S102).
When the volume of the audio data is adjusted, the division processing unit 12 divides the audio data after the volume adjustment by a predetermined unit time to generate divided data (S103).
In division processing for generating divided data, audio data is divided at a division position for each predetermined unit time from the start position. At this time, when the break position is a silent portion, the voice data is divided by the silent portion, and when the break position is a sound portion, the silent portion is within a predetermined time before the sound portion. The voice data is divided to generate divided data.

Subsequently, the compression processing unit 13 generates compressed data in which the silent portion is deleted from the divided data generated by the division processing unit 12 (S104).
When the compressed data is generated, the identification information issuing unit 14 issues and adds identification information to each compressed data (S105).
The identification information is information for identifying each compressed data and making it possible to grasp the order, and is registered in the identification information storage unit 1A according to the issue (S106).

The compressed data is transmitted from the speech processing apparatus 1 to the speech processing engine 2 together with the text conversion request (S107).
In response to the reception of the compressed data, the voice-to-text engine 2 executes a process of converting the compressed data into text (S108).
When the compressed data is converted into text by the text conversion process and the pre-combination text data is generated, the pre-combination text data is transmitted to the speech to text apparatus 1 (S109).

  The speech processing apparatus 1 receives the pre-combination text data from the speech-to-text engine 2 and the combination processing unit 15 generates post-combination text data by combining the pre-combination text data in the order before division (S110). In this process, the combining processing unit 15 arranges and combines the pre-combination text data in the order of voice data with reference to the identification information storage unit 1A based on the identification information held by the pre-combination text data.

The generated combined text data is transmitted by the communication processing unit 16 to the user terminal 3 (S111).
Thereby, the user can obtain data in which desired voice data is converted into text.
According to the voice-to-text device 1 according to the present embodiment, since the silent part is deleted in advance in the voice-to-text device, the voice-to-text engine does not make silent parts into text, and the processing load of the voice-to-text engine Can be reduced.
In addition, since the voice data is divided by the silent portion at the break position of the predetermined unit time or the silent portion existing before the break position, the word is not divided midway, and erroneous conversion is performed. You can prevent.

Next, a modification of the speech-to-text device according to the above-described embodiment will be described with reference to FIG.
In the present embodiment, the above-described voice-to-text device 1 is incorporated as a voice-to-text processing unit 41 into a user terminal 4 configured by a so-called smart phone, personal computer or the like, and functions as application software.
In the following description, the same reference numerals as in the above description are assigned to functional units that perform the same functions as the functional units included in the above-described voice-to-text device 1.

In the present embodiment, the user terminal 4 has a functional unit including a voice-to-text processing unit 41, an input / output processing unit 42, and a communication processing unit 43, and via the voice-to-text engine 2 and the network NW such as the Internet. It is configured to be communicable.
Further, the voice-to-text processing unit 41 includes an identification information storage unit 1A, a volume adjustment unit 11, a division processing unit 12, a compression processing unit 13, an identification information issuing unit 14, and a combination processing unit 15.

The input / output processing unit 42 is a functional unit that inputs and outputs various data, and is configured of a display, a touch panel, and the like.
The communication processing unit 43 is a functional unit for executing transmission and reception of data via the voice-to-text engine 2 and the network NW, and is realized by a browser program or the like.

Next, in the user terminal according to this example, a flow of processing for converting voice data into text is shown in FIG. The same processes as those in the above-described example described with reference to FIG.
First, when the user requests the user terminal 4 to execute textification processing for desired voice data stored in the terminal, the volume of the voice data is adjusted to a predetermined volume by the voice adjustment unit 11 (S102).
When the sound volume of the audio data is adjusted, the division processing unit 12 divides the sound data after the volume adjustment by a predetermined unit time to generate divided data (S103), and the compression processing unit 13 divides the sound data. The compressed data from which the silent part is deleted is generated from the divided data generated by the (S104).
Identification information is issued and added to each compressed data by the identification information issuing unit 14 (S105), and the identification information is identified as information for identifying individual compressed data and making it possible to grasp the order thereof. It is registered in the information storage unit 1A (S106).

The compressed data is transmitted from the user terminal 4 to the voice-to-text engine 2 together with the text-to-text request (S107).
When the text-to-text processing of the compressed data is executed in the voice-to-text engine 2 (S108), the pre-combination text data is transmitted to the user terminal 4 (S109).

The user terminal 4 that has received the pre-combination text data from the speech-to-text engine 2 generates the post-combination text data by combining the pre-combination text data in the order before division by the combining processing unit 15 (S110).
As described above, if the textification processing unit 41 is installed as application software in the user terminal 4 and can be executed, voice data can be transmitted from the user terminal 4 directly to the speech processing engine 2 to obtain text data. You can do so.

In the above embodiment, not only voice data completed as one file but also voice data being created can be converted to text as needed.
The process in this case will be described with reference to FIG.
FIG. 9A shows a state in which voice data is being generated as voice data being recorded. In the voice data 100, the break positions are sequentially set with 50 [sec] from the start position as the unit time, and in FIG. 9A, the break positions 100b are provided in the talks portion 101c being created. ing.

In this example, first, after the sound volume is adjusted for the section of 50 [sec] divided by the division position 100b, the processing after the above-described division is executed.
That is, for the example of FIG. 9A, as described above, the audio data 100 is divided with the silent portion 102b within 10 [sec] before the separation position 100b as the division position, and is shown in FIG. The divided data 110a is generated as shown in FIG.

Compression processing section 13 from this state, delete silence 102a, the 102b ₁ as shown in FIG. 9 (c), to generate compressed data 120a. The compressed data 120a is textified by the speech-to-text engine and transmitted to the user terminal.
On the other hand, the divided data 110b following the divided data 110a is converted into text after waiting for voice data to be generated for a predetermined unit time (50 [sec] in this example) from the dividing position 100b.

The flow of processing according to this example will be described with reference to FIG. 10, using the above-described case of the user terminal 4 as an example.
First, the user requests the user terminal 4 to execute text conversion processing, for example, for voice data being recorded.
In response to this, the text conversion processing unit 41 determines whether the audio data has reached a predetermined time (S202). When the audio data has reached a predetermined time, the audio data is divided at the predetermined time, and the audio adjustment unit 11 adjusts the volume of the audio data to a predetermined volume (S102).
When the sound volume of the audio data is adjusted, the division processing unit 12 divides the sound data after the volume adjustment by a predetermined unit time to generate divided data (S103), and the compression processing unit 13 divides the sound data. The compressed data from which the silent part is deleted is generated from the divided data generated by the (S104).
Identification information is issued and added to each compressed data by the identification information issuing unit 14 (S105), and the identification information is identified as information for identifying individual compressed data and making it possible to grasp the order thereof. It is registered in the information storage unit 1A (S106).

The compressed data is transmitted from the user terminal 4 to the voice-to-text engine 2 together with the text-to-text request (S107).
When the text-to-text processing of the compressed data is executed in the voice-to-text engine 2 (S108), the pre-combination text data is transmitted to the user terminal 4 (S109).
The user terminal 4 that has received the pre-combination text data from the speech-to-text engine 2 generates the post-combination text data by combining the pre-combination text data in the order before division by the combining processing unit 15 (S110).
It is determined whether or not the audio data still continues (S202), and if the audio data that has not been converted into text continues, the process returns to S201 to continue the conversion of the audio data into text. On the other hand, when the voice data to be converted to text is finished, the series of processes is finished.
As a result, the voice data being created is also textified as needed.

  In the above embodiment of the present invention, the division of the audio data is performed in the silent part within a predetermined time before the division position, when the division position becomes the talk part. However, the present invention is not limited to this, and it may be a silent part within a fixed time after the break position.

  Further, the volume adjustment of the audio data may not be performed before the generation of the divided data, and may be performed at the time of generation of the divided data or at the time of generation of the compressed data.

DESCRIPTION OF SYMBOLS 1 voice text-to-speech device 11 volume adjustment unit 12 division processing unit 13 compression processing unit 14 identification issuing unit 15 combination processing unit 16 communication processing unit 1 A identification information storage unit 2 voice text conversion engine 3 user terminal 4 user terminal 41 voice text conversion processing Unit 42 Input / output processing unit 43 Communication processing unit NW network

Claims (7)

A device for converting voice data into text,
A speech-to-text engine for converting the speech data into text, and communicable via a network,
Division processing means for dividing the audio data by a predetermined unit time to generate divided data;
Compression processing means for generating compressed data obtained by deleting silent parts from the divided data;
Compressed data transmitting means for transmitting the compressed data to the voice to text engine;
Pre-combination text data receiving means for receiving pre-combination text data obtained by converting the compressed data into text form from the speech-to-text engine;
Combining processing means for generating post-join text data by joining the pre-join text data in a pre-division order;
An audio-to-text device characterized in that. The dividing means divides the voice data by the silent portion to generate divided data when the break position at a predetermined unit time from the start position of the voice data is a silent portion, and starts the voice data. When the demarcation position for each predetermined unit time from the position is a talkative part, the voice data is divided by a silent part within a predetermined time before the talkable part to generate divided data.
A speech to text apparatus according to claim 1. The apparatus further comprises volume adjustment means for adjusting the volume of the audio data to a predetermined volume,
The dividing unit divides the audio data adjusted to a predetermined volume at a predetermined unit time to generate divided data.
An apparatus according to claim 1 or 2. Identification information issuing means for issuing identification information to the compressed data;
The combination processing means generates combined text data by combining the pre-combination text data in the order before division based on the identification information.
The speech-to-text device according to any one of claims 1 to 3. It is configured to be communicable with the user terminal used by the user via the network.
Voice data receiving means for receiving the voice data from the user terminal;
The combined text data transmitting means for transmitting the combined text data to the user terminal;
An apparatus for converting text to speech according to any one of claims 1 to 4. A method for converting voice data into text,
A speech-to-text engine for converting the speech data into text and a computer configured to be communicable via a network,
A division process of dividing the audio data by a predetermined unit time to generate divided data;
Compression processing for generating compressed data in which silent parts are removed from the divided data;
Compressed data transmission processing for transmitting the compressed data to the voice-to-text engine;
A pre-combination text data reception process for receiving pre-combination text data obtained by converting the compressed data into text data from the speech-to-text engine;
Performing a combining process of combining the pre-combination text data in a pre-division order to generate post-combination text data;
An audio text conversion method characterized by A computer program for converting voice data into text,
A voice-to-text engine for converting the voice data into text, and a computer configured to be able to communicate via a network,
A division process of dividing the audio data by a predetermined unit time to generate divided data;
Compression processing for generating compressed data in which silent parts are removed from the divided data;
Compressed data transmission processing for transmitting the compressed data to the voice-to-text engine;
A pre-combination text data reception process for receiving pre-combination text data obtained by converting the compressed data into text data from the speech-to-text engine;
Performing a combining process of combining the text data before combining in the order before splitting to generate text data after combining;
Computer program.

Images