KR100314659B1 - Automatic transformation device and method for telephone speech database - Google Patents

Abstract

The present invention relates to an automatic telephone database construction apparatus and method for converting an existing voice database recorded in a quiet environment into a telephone voice through a telephone line.

The telephone voice database automatic building apparatus of the present invention comprises: a telephone network matching unit for making and receiving a telephone call and for outputting a special tone for converting a signal between a digital voice signal and an analog telephone voice and an end of a voice file; Voice database and voice to read the voice file of the voice database to the telephone network matching unit, convert it as necessary, output it, record digital telephone voice passing through the telephone line output from the telephone network matching unit, and make a telephone voice file to build a telephone voice database. And a host including a file reproducing unit, a call processing control unit, a telephone voice file recording and storage unit, a telephone voice database, and a system control unit.

The present invention automatically establishes a telephone voice database without an operator and a user who provides a voice, thereby reducing operation time and cost for the operator and the user, and comparing the recognition performance between the existing voice data and the telephone voice data. This study analyzes the influence of telephone channel on recognition performance and enables the development of voice recognizer optimized for telephone channel.

Description

Automatic transformation device and method for telephone speech database

The present invention relates to an apparatus for constructing a telephone speech database essential for telephone line speech recognition, and more particularly, to an apparatus and method for automatically constructing a telephone speech database for automatically converting a large-scale speech database recorded in a quiet environment into telephone speech data.

In order to implement a voice recognition service in a telephone environment, there is a problem in that a voice recognition engine needs to be trained with voice data in a telephone environment. To this end, a device for building a telephone voice database is required.

A general telephone voice database construction device collects the user's voice directly through a telephone line. The technology of collecting multiple voices simultaneously in multiple channels for efficient collection process (see Korean Patent Application No. 1997-052666). Technology that automatically collects the user's voice without an operator using announcements (see Korean Patent Application No. 1997-047152), and can easily label voice to reduce the time required to edit the collected voice data. Techniques for providing an environment (see Korean Patent Application No. 1997-047337), and a technique for automatically extracting a voice section and storing it as a voice file are known.

There is also a technique using a simulator without using a telephone line environment (see US Pat. No. 5,475,792), but since the telephone channel has various factors affecting the performance of voice recognition, modeling optimized for the actual telephone channel environment is not easy.

For example, the conventional telephone voice database for telephone line voice recognition is constructed by an automatic telephone voice collection apparatus as shown in FIG. 1.

As shown in FIG. 1, the configuration of a conventional telephone voice collecting device includes: a telephone line matching unit 10 for detecting whether a telephone call is received from a user and converting an analog signal from a telephone line and a digital signal in the apparatus; A voice signal processor 12 which separates only the voice of the user from the signal mixed with the voice of the user input through the telephone line matching unit 10 and the signal output through the guide broadcast, and extracts only the voice section of the user; A voice file storage unit 16 for storing the voice section data detected by the voice signal processing unit 12 as a voice file; And a call control and voice recording unit 14 for controlling call control and voice file storage in the telephone line matching unit 10.

However, it is not easy to make a device for building a telephone voice database as shown in FIG. In addition, there is a problem of mobilizing a large number of people, it is necessary to check whether data is lost due to carelessness of users, and data labeling problem is complicated. In addition, it is difficult to quantitatively determine the improvement of recognition performance due to channel distortion and the improvement due to algorithm improvement.

The use of various voice databases recorded in the existing quiet environment can save a lot of time and money, and the voice recognition engine in the telephone environment can be analyzed through the comparison of voice recognition performances. It is advantageous for development.

In order to create a telephone voice database using the existing voice database utilizing these advantages, there is a method of using a channel simulator instead of passing the existing voice database over a real telephone line. It is difficult to know whether it reflects enough, and it is not easy to control various parameters.

Another way to create a phone voice database using an existing voice database is to create a voice file as an analog voice and play it through a telephone receiver, and then collect it using a recording device connected to another phone. Cannot be ignored, and it is difficult to analyze the improvement of recognition performance considering only channel distortion, and it is necessary to deal with the problem of distinguishing the beginning and the end of the speech section and the labeling problem.

An object of the present invention to solve the above problems is to collect the voice data collected and labeled in the existing quiet environment through the telephone network, instead of collecting the user's voice directly through the telephone to build a telephone voice database, The present invention provides an apparatus and method for automatically constructing a telephone voice database that can easily build a telephone voice database.

1 is a block diagram showing a general phone voice data collection device

2 is a block diagram of the automatic construction of telephone voice database according to the present invention

3 is a flowchart illustrating an automatic conversion process of a voice file into a phone voice file according to the present invention.

4 is a flowchart illustrating a process of automatically building a telephone voice database according to the present invention.

Explanation of symbols on the main parts of the drawings

10: telephone line matching unit 12: voice signal processing unit

14: call control and voice recording unit 16: voice file storage unit

50: local telephone station (PSTN) of public network 51: outgoing telephone line

52: incoming call line 60: telephone network matching unit

70: host computer 71: call processing control unit

72: system control unit

73: Clean Speech database

74: voice file playback unit 75: telephone voice file recording and storage unit

76: telephone voice database

In order to achieve the above object, the present invention makes and receives a telephone call to a local telephone station of a public network, converts a digital voice signal input from a voice file reproducing unit into an analog telephone voice, transmits it through a transmission telephone line, and is received through a receiving telephone line. A telephone network matching unit for converting an analog telephone voice into a digital voice signal and outputting the telephone voice file recording and storage unit and outputting a special tone for notifying the end of the voice file; A call processing control unit controlling digital voice data input / output to the telephone network matching unit, an operation of making and disconnecting a call from one trunk line interface to the other trunk line interface of the telephone network matching unit, and outputting a special tone; A voice file reproducing unit for reading and downsampling the voice file from the voice database, performing conversion necessary for passing the telephone line, and then transmitting the converted digital voice data to the telephone network matching unit; Record the digital voice signal outputted from the telephone network matching unit until the telephone network matching unit receives a tone indicating the end of playing the voice file, and when the recording of one voice file is completed, the telephone voice file is stored in the telephone of the mass storage device. Telephone voice file recording and storage unit for storing in the voice database; And a system controller for controlling the call voice database automatic construction device by controlling the operations of the call processing controller, the voice file reproducing unit, and the phone voice file recording and storage unit.

Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings.

Figure 2 is a block diagram showing the configuration of the automatic phone voice database construction device implemented by the present invention, the local telephone station 50 of the public network to be able to build a phone voice database using an existing voice database; A host computer including a call processing control unit 71, a voice file reproducing unit 74, a phone voice file recording and storage unit 75, a voice database 73, a phone voice database 76 and a system control unit 72. 70 and; A telephone network matching section 60 is connected to a local telephone station 50 of the public network via a transmission telephone line 51 and a reception telephone line 52, and is also connected to the host computer 70.

Specifically, the local telephone station 50 of the public network is dialed, and the A-Law PCM format digital voice signal input from the voice file reproducing unit 74 is converted into an analog voice signal and transmitted through the transmission telephone line 51. Detects the incoming call of the local telephone company 50 of the public network to answer the call, and converts the analog voice signal received through the receiving phone line 52 into a digital voice signal of the A-Law PCM format to record and store the phone voice file. A telephone network matching unit 60 for outputting to 75 and processing necessary DTMF tones generation; Digital voice data input / output of the telephone network matching unit 60, generation / release of a call from a transmission telephone line 51 to which voice data is to be exported to a reception telephone line 52 to receive voice data passing through a local telephone station 50 of a public network; A call processing control unit 71 for controlling transmission and reception of tones which can distinguish when one voice file is reproduced; Read the voice file from the recorded voice database (73) in the existing quiet environment, remove the file header information, pass it through a low pass filter, downsample it to 8 kHz, quantize it into 8-bit A-Law PCM format, and digitally convert it. A voice file reproducing unit 74 for transmitting the converted digital voice data to the telephone network matching unit 60; When the telephone network matching unit 60 converts the analog voice signal received from the telephone line into a digital voice signal and transmits it, the telephone network matching unit 60 matches the telephone network until the DTMF tone indicating the end of the voice file reproduction is received. The digital voice signal output from the unit 60 is recorded, and when the recording of one voice file is completed, the digital voice signal is converted into a 16-bit linear PCM format, and the converted phone voice file is converted into a phone voice database of the mass storage device (76). Telephone voice file recording and storage unit 75 to store; And a system controller 72 for totally controlling the telephone voice database automatic construction device of the present invention by controlling operations of the call processing control unit 71, the voice file reproducing unit 74, and the telephone voice file recording and storing unit 75. It is configured to include.

A process of automatically converting a voice file into a phone voice file in the apparatus for automatically building a phone voice database configured as described above is illustrated in FIG. 3.

First, the voice file reproducing unit 74 reads one voice file from the voice database 73 while the telephone is connected from the transmission telephone line 51 of the telephone network matching unit 60 to the receiving telephone line 52 (S1). ).

Then, the audio file reproducing unit 74 removes the file header information (S2), and passes through a low pass digital filter of 4 kHz or less in order to eliminate the interference effect (aliasing) that may occur during the downsampling process (S3). ), The audio data having a sampling rate of 16 kHz or more is downsampled to 8 kHz (S4).

In addition, in order to recognize the voice data passing through the telephone network by the local telephone company 50 in the public network, the digital voice data is assumed to be connected to the E1 trunk and the digital voice data is in 8-bit A-Law PCM format. For this purpose, the digital voice of 16-bit linear PCM format downsampled to 8 kHz is quantized into 8-bit A-Law PCM format digital voice (S5).

When the voice file reproducing unit 74 transmits the digital voice signal converted into the channel designated by the control of the call processing control unit 71 to the telephone network matching unit 60, the telephone network matching unit 60 converts it into an analog voice signal. To the outgoing telephone line 51. Then, the analog voice signal to the transmission telephone line 51 is received by the reception telephone line 51 through the local telephone station 50 of the public network (S6).

The received analog voice signal is converted into a digital voice signal by the telephone network matching unit 60 and then recorded by the telephone voice file recording and storage unit 75 (S7), and when the recording is completed, it is converted back to the 16-bit linear PCM format. (S8), it is stored in the telephone voice database 76 as a telephone voice file (S9).

4 is a flowchart illustrating a process of automatically converting a large number of voice files into a phone voice file in the apparatus for automatically building a phone voice database according to the present invention.

This will be described in detail.

First, in order to build a telephone voice database using the voice database, the system control unit 72 makes a telephone network matching unit 60 call a local telephone station 50 of the public network through the call processing control unit 71 to transmit a telephone line 51. ) To form a local loop through the local telephone station 50 of the public network (S10).

When the local loop is formed, the voice file reproducing unit 74 reads one voice file from the voice database 73 (S11), and converts the read voice file into an 8 kHz 8-bit A-Law PCM (S12). At this time, the voice file read out from the voice database 73 by the voice file reproducing unit 74 is designated by the system control unit 72 with reference to the voice file list. As described above, the conversion of the voice file is performed by removing the header information of the read voice file, passing through a low pass digital filter, downsampling at 8 kHz, and then again using 8-bit A-Law PCM format digital voice. This is done through a series of conversions.

The telephone network matching unit 71 converts the digital voice file converted into the 8 kHz 8-bit A-Law PCM input from the voice file reproducing unit 64 into an analog voice signal and exports it to the transmission telephone line 51 to reproduce the voice file. Start, and receives the analog voice signal to the receiving telephone line 52 through the local telephone station 50 of the public network to convert the digital voice signal and output to the telephone voice file recording and storage unit 75 to start the telephone voice recording (S13).

At this time, the telephone voice file recording and storage unit 75 records the telephone voice under the control of the system control unit 72.

When the end of the voice file is detected (S15) while the voice file is being reproduced by the transmission telephone line 51 (S15), the telephone network matching unit 60 controls the call processing control unit 71 to transmit DTMF tones to the transmission telephone line 51. Export '#' (S16).

At this time, the end of the voice file is recognized by the end of file (OFF) of the file or by the DTMF tone added to the end of the file. When the voice file playing unit 74 recognizes the end of the voice file, it notifies the system controller 72 of the end of the file. . Then, the system control unit 72 controls the telephone network matching unit 60 to transmit the DTMF tone '#' indicating the end of the voice file through the call processing control unit 71.

Then, it is checked whether the DTMF tone '#' indicating the end of the voice file is detected at the reception telephone line 52 (S17). If the DTMF tone '#' is not detected from the reception telephone line 52, the voice recording is continued (S18). Then, the flow advances to step S14 to continuously play back the voice file, and the end of the voice file is detected to confirm whether the DTMF tone '#' has been detected on the receiving phone line 52 through the transmission telephone line 51.

However, when the DTMF tone '#' is detected in the reception telephone line 52 as a result of the checking in step S17, the voice recording in the telephone voice file recording and storage unit 75 ends (S19). At this time, the end of the voice recording, the telephone network matching unit 60 notifies the system control unit 72 through the call processing control unit 71 that the DTMF tone '#' has been detected on the receiving telephone line 52, the system control unit 72 This is achieved by controlling the telephone voice file recording and storage unit 75.

When the voice playback and recording of one voice file is completed, the phone voice file recording and storage unit 75 converts the voice recording completed in the A-Law PCM format into a 16-bit linear PCM format and then stores the phone voice database. Stored in (76) (S20).

At this time, the telephone voice file stored in the telephone voice database 76 is labeled and stored with the name specified by the system controller 72.

Then, it is checked whether all files in the voice database 63 have been converted (S21). As a result of the verification, when all the files in the voice database 63 have been converted, the telephone is disconnected, the local loop is formed with the local telephone station 50 of the public network, and the telephone voice database construction work is terminated (S22). However, if there is a voice file to be converted in the voice database 63, the process returns to step S11 to repeat the above steps S12 to S21 to read the voice data file and convert it into a telephone voice.

The end of the telephone voice database construction work is also confirmed by the system control unit 72 by referring to the voice file list that all voice files are converted into telephone voice files, and the call processing control unit 71 and the telephone voice file recording and storage unit 75. By controlling

As described above, by automating the telephone line passage of the already labeled voice data, it is possible to automate the construction of the telephone voice conversion and the telephone voice database for a large voice database.

As described above, the present invention can automatically build a telephone voice database using an existing voice database without an operator and a user who provides a voice, thereby reducing work time and cost for the operator and the user.

In addition, it is possible to compare the recognition performance between the existing voice data and the telephone voice data in the voice recognition system, thereby analyzing the influence of the telephone channel on the recognition performance, and thus, the development of the voice recognition system optimized for the telephone channel is possible. do.

The technical spirit of the present invention has been described above with reference to the accompanying drawings, but this is by way of example only and not intended to limit the present invention. In addition, it is obvious that any person skilled in the art can make various modifications and imitations without departing from the scope of the technical idea of the present invention.

Claims (7)

A local telephone station 50 of the public network so that a telephone voice database can be constructed using an existing voice database; A host computer including a call processing control unit 71, a voice file reproducing unit 74, a phone voice file recording and storage unit 75, a voice database 73, a phone voice database 76 and a system control unit 72. 70 and; A telephone network matching unit 60 connected to the local telephone station 50 of the public network via a transmission telephone line 51 and a reception telephone line 52, and also connected to the host computer 70; The telephone network matching unit 60 makes a call to the local telephone station 50 of the public network, converts the digital voice signal input from the voice file reproducing unit 74 into an analog voice signal, and transmits it through the transmission telephone line 51. It detects the incoming call of the local telephone company 50 of the public network to answer the call and converts the analog voice signal received through the receiving phone line 52 into a digital voice signal and outputs it to the telephone voice file recording and storage unit 75, It also exports DTMF tones that indicate the end of a voice file, The call processing control unit 71 receives a digital voice data input / output of the telephone network matching unit 60 and a receiving telephone line 52 to receive voice data passing through a local telephone station 50 of a public network from a transmission telephone line 51 to which voice data is to be exported. Controls the generation / release of a call to and the transmission and reception of a tone which can be distinguished when one voice file is played. The voice file reproducing unit 74 reads the voice file from the voice database 73 recorded in the existing quiet environment, performs conversion necessary for passing the telephone line, and then transmits the converted digital voice data to the telephone network matching unit. The telephone voice file recording and storage unit 75 records the digital voice signal output from the telephone network matching unit 60 until the telephone network matching unit 60 receives a DTMF tone indicating the end of the voice file reproduction, When recording of one voice file is completed, the format of the voice file is converted, and the converted phone voice file is stored in the phone voice database 76 of the mass storage device. The system control unit 72 controls overall operation of the telephone voice database by controlling operations of the call processing control unit 71, the voice file reproducing unit 74, and the phone voice file recording and storage unit 75. Phone speech database automatic build device. Using the existing voice database to automatically build a phone voice database, A first step of making a local loop through a local telephone station 50 of a public network by dialing a transmission telephone line 51 from a transmission telephone line 51 of the telephone network matching unit 60; A second step of the voice file reproducing unit 74 reading and converting one voice file from the voice database 73; The telephone network matching unit 71 starts the reproduction of the voice file through the transmission telephone line 51, receives it from the receiving telephone line 52, and records the telephone voice file and starts recording the telephone voice file in the storage unit 75. Step 3; A fourth step of, if the end of the voice file is detected during the reproduction of the voice file, outputting a DTMF tone indicating the end of the voice file from the telephone network matching unit 60 to the transmission telephone line 51; If the DTMF tone indicating the end of the voice file is not detected on the incoming telephone line 52, repeating step 4 while continuing to record the telephone voice, and recording and storing the telephone voice file when the DTMF tone indicating the end of the voice file is detected. A fifth step of ending the voice recording in the unit 75; A sixth step of converting the format of the recorded voice into a telephone voice database 76 when all of the voice playback and recording of one voice file are completed; And Check that all files in the voice database 63 have been converted into telephone voice files. If all files in the voice database 63 have been converted into telephone voice files, hang up the telephone to complete the telephone voice database construction operation. And a seventh step of returning to the second step in order to convert the phone to a voice file. The method of claim 2, The voice file conversion in the second step, Remove the header information of the read voice file, pass through a lowpass digital filter to eliminate the interference effects that may occur during downsampling, then downsample 8 kHz and convert to 8-bit A-Law PCM. How to build telephone voice database automatically. The method according to claim 2 or 3, DTMF tones indicating the end of the voice file in the fourth step, The voice file reproducing unit 74 recognizes the EOF of the voice file or the DTMF tone added to the end of the voice file and informs the system control unit 72, so that the system control unit 72 matches the end of the voice file in the telephone network matching unit 60. A method for automatically constructing a telephone voice database, characterized in that it is sent through a transmission telephone line (51) by controlling the call processing control unit (71) to send a DTMF tone. The method according to claim 2 or 3, The fifth step of terminating the voice recording, When the DTMF tone indicating the end of the voice file is detected on the receiving telephone line 52, the telephone network matching unit 60 notifies the system control unit 72 through the call processing control unit 71, and the system control unit 72 calls the voice file. Method for automatically constructing a telephone voice database, characterized in that the recording and storage unit 75 controls to end voice recording. The method according to claim 2 or 3, The sixth step, And converting the recorded voice into a 16-bit linear PCM, and labeling the recorded voice with a name designated by the system controller 72 and storing the recorded voice in the telephone voice database 76. A computer-readable recording medium having recorded thereon a program for automatically constructing a telephone voice database using an existing voice database, A first step of making a local loop through a local telephone station 50 of a public network by dialing a transmission telephone line 51 from a transmission telephone line 51 of the telephone network matching unit 60; A second step of the voice file reproducing unit 74 reading and converting one voice file from the voice database 73; The telephone network matching unit 71 starts the reproduction of the voice file through the transmission telephone line 51, receives it from the receiving telephone line 52, and records the telephone voice file and starts recording the telephone voice file in the storage unit 75. Step 3; A fourth step of, if the end of the voice file is detected during the reproduction of the voice file, outputting a DTMF tone indicating the end of the voice file from the telephone network matching unit 60 to the transmission telephone line 51; If the DTMF tone indicating the end of the voice file is not detected on the incoming telephone line 52, repeating step 4 while continuing to record the telephone voice, and recording and storing the telephone voice file when the DTMF tone indicating the end of the voice file is detected. A fifth step of ending voice recording in the unit 75; A sixth step of converting the format of the recorded voice into a telephone voice database 76 when all of the voice playback and recording of one voice file are completed; And Check that all files in the voice database 63 have been converted into telephone voice files. If all files in the voice database 63 have been converted into telephone voice files, hang up the telephone to complete the telephone voice database construction operation. And a seventh step of returning to the second step to convert the data into a telephone voice file.