JP5224966B2 - Voice transcription server - Google Patents

Description

  The present invention relates to a voice characterizing server and a voice characterizing method for simultaneously characterizing a call voice and distributing data to a plurality of callers who have a voice call.

  For example, voice collected by a microphone provided in an apparatus having an automatic recognition function or the like is converted into text. Furthermore, the received voice received from the communication means is converted into text, and the own voice and the received voice are converted into text and output, or the sender's voice is transmitted as it is from the receiver's mobile phone or converted into text and transmitted. There has been proposed a communication host device that enables selection to be performed.

In addition, voice data for each single sound uttered by the user transmitted from the user terminal is managed by single voice dictionary data for each user, and the voice data input and transmitted by the user terminal is separated for each single voice. It has been proposed to improve the accuracy of recognition. Further, in order to improve data search, it has been proposed to perform a search using a conventional filing key code when a filing key code as a search key matches an index code checksum.
JP 2000-244683 A JP 2006-39805 A JP 2005-49713 A JP-A-6-162096

  In the conventional technology for converting the voice during the call to a character, each communication host device must be provided with a voice recognition function, and voice recognition must be performed for each voice, which increases the load on the communication host device. . Further, since there is no mechanism for improving accuracy by storing voice data or similarity of conversation, it is impossible to improve the accuracy of speech recognition.

  In addition, in the conventional method of registering a voice for each single tone, the accuracy of voice conversion collected by a voice recognition method using HHM (Hidden Markov Mode) is improved. Voice data cannot be collected from everyday natural calls.

  Furthermore, the conventional search technique is a technique for performing a character search from a file in which a character string previously managed according to a predetermined rule is managed, and is not suitable for searching voice data having a larger capacity than character data. .

  Speech characteristics such as pronunciation and accent differ among individuals, and it is often difficult to perform speech recognition completely, and further, speech recognition in a noisy environment becomes even more difficult. In addition, language analysis or the like is required at the time of conversion to character data after speech recognition. For example, it is difficult to perform complete and accurate conversion even with Japanese original syntax analysis and kanji conversion processing.

  In order to increase the analysis rate of speech and language, it is necessary to accumulate speech data and text data, but it is limited to a specific environment and limited time, and it is difficult to accumulate ordinary daily conversation There are many. Therefore, even if the converted character data is used, there is a problem that the environment for extracting and providing the character data is limited.

  Therefore, an object of the present invention is to perform data distribution by simultaneously converting a call voice to a plurality of speakers who are making a voice call.

In order to solve the above-mentioned problem, the phonetic transcription server includes voice data receiving means for receiving voice data of a speaker who is talking through a communication line, and voice data of the speaker who is received by the voice data receiving means . Voice registration determination means for determining whether or not a checksum of the voice data is registered in a voice database in which the checksum of the voice data and character data converted from the voice data are associated with each other; and Distribution means for distributing character data corresponding to voice data with a matching checksum to the speaker , and voice for converting the voice data into character data when there is no voice data with the matching checksum in the voice database a character conversion unit, and a checksum of the audio data, character de converted from voice data by the voice character converting means By storing in the speech database and data in correspondence, configured with a voice registration means for registering the voice data.

  It is possible to convert the voice of a conversation using a mobile phone into text and simultaneously deliver the converted character data to a destination specified by the speaker who is talking.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings.

  FIG. 1 is a diagram for explaining the outline of the overall processing. 1 includes two or more exchanges 11, 12... Connected via a telephone line network, and a voice during a call transmitted from each of these exchanges 11, 12... And a voice characterizing server 5 for converting the character into a character. The mobile terminals 31, 32... And the terminal 50 that is installed at a service provider that requires voice recognition and can receive incoming calls are connected via the predetermined radio towers 21, 22. A PC (Personal Computer) 6 is connected to the voice characterizing server 5 through a network and receives a service provided by the voice characterizing server 5. The phonetic transcription server 5 is a computer device controlled by a CPU, and includes a storage area such as a main storage device and an auxiliary storage device, a communication device for connecting to a network, an input / output device, and a display device. Have.

  In FIG. 1, an outline of the entire process will be described in the case where the speaker A and the speaker B talk to each other using the mobile phones 31 and 32 that they own. Hereinafter, in the present embodiment, “speaker” means “caller” when talking to the other party using a mobile phone or the like, and is a service provided by the voice transcription server 5 via the Internet or the like. When using, it means “user”.

  For example, when the speaker A dials to the speaker B from the mobile phone 31 as the other party, it is received by the exchange 11 via the radio tower 21 having the location of the mobile phone 31 in the area, and further via the telephone network. Then, the office exchange 12 on the speaker B side receives the call through a predetermined communication procedure. The exchange 12 receives the incoming call at the mobile phone 32 via the radio tower 22 where the position of the mobile phone 32 of the speaker B is within the area. A voice call is started by receiving a dial transmission by the speaker A at the cellular phone 31 of the speaker B.

  The exchange 11 on the side of the speaker A continues communication until the line is disconnected, and also collects voice data of the speaker A from the started voice call via the communication line and converts it into character data. It transmits to the voice characterizing server 5 to convert. Similarly, the voice of the speaker B is transmitted from the office switch 12 on the speaker B side to the voice characterizing server 5. In the following, the processing for the voice data related to the speaker A will be described, but the same processing is performed for the voice data related to the speaker B.

  Whether or not the voice characterizing server 5 agrees that the voicer A accumulates additional information such as voice data, voicer name, date and time, place, and context before and after as voicer information for voice processing. The speaker A is authenticated using the speaker DB (database) 41 (step S1).

  Only when it is confirmed that the speaker A is authenticated and agrees, the following processing is continuously performed. If the authentication of the speaker A fails, or if the authentication is successful but not agreed, the processing related to the voice data of the speaker A is terminated without executing the following processing. The phonetic transcription server 5 notifies the central office switch 11 of the authentication result including the agreement / denial agreement. Only when the authentication is confirmed and the consent is confirmed, one session relating to the speaker A is established between the central office switch 11 and the voice conversion server 5, and the voice data obtained by digitizing the voice of the speaker A from the central office switch 11 is the voice. The voice processing is executed by being transmitted to the characterizing server 5. With such authentication, it is possible to execute voice processing in consideration of privacy issues. It should be noted that when other person uses the cellular phone of the speaker A, it is possible to set so that voice collection is not performed before the call in order to cope with a case where the voice is not desired to be collected. May be. If a setting is made so that voice collection is not performed before a call, voice collection is not performed even if the above-described consent is confirmed at the time of authentication. That is, the authentication result indicating the disapproval may be notified to the central office switch 11.

  When the voice conversion server 5 receives voice data from the central office 11, the voice conversion server 5 converts the size of the voice analog signal into a digital signal expressed as binary numerical data, and uttered syllables with a portion where there is no voice as a delimiter. The audio data divided for each frame is stored in the audio data file 42. The part having no sound is defined as a part having a predetermined intensity or less in the intensity of the sound represented by a digital value. At that time, the voice characterizing server 5 calculates a value obtained by adding all the voice digital signals stored for each voice data file 42 as a checksum and adds the calculated value to the voice data file 42 (step S2).

  Then, the phonetic transcription server 5 checks whether or not voice data in which the calculated checksum and additional information such as a speaker name match are registered in the voice DB 43 by searching the voice DB 43 (step). S3). By using the checksum, the search process can be performed at a higher speed than the matching search using the audio digital signal.

  In step S3, if the audio data is not registered in the audio DB 43, the audio data is divided by a predetermined delimiter method to be described later, a checksum is calculated for each divided audio data, and the calculated checksum is similarly calculated. It is confirmed by searching the voice DB 43 whether or not voice data matching the additional information such as the speaker name is registered in the voice DB 43.

  Then, the phonetic transcription server 5 determines whether it has been registered by the search (step S4). When the audio data is divided, registration determination processing is performed for each divided audio data. The phonetic character server 5 stores the character data read out by searching the voice DB 43 together with the speaker ID of the speaker A in the transmission standby DB 47 (step S5).

  The phonetic transcription server 5 stores in the transmission standby DB 47 until the conversion result transmission timing condition for transmitting the conversion result to the character data stored in the speaker information of the speaker A is satisfied in the speaker DB 41. Keep it. If the conversion result transmission timing condition is satisfied, the phonetic character server 5 transmits the character data according to the calling method stored in the speaker information of the speaker A from the speaker DB 41 (step S6).

  On the other hand, if the corresponding character data cannot be retrieved from the voice DB 43 even if the voice data is divided, that is, if the voice data is not registered, the voice characterizing server 5 uses the voice recognition function and the language analysis function. The data is converted into character data, and the voice data, the character data, and the additional information are registered in the voice DB 43 in association with the speaker ID of the speaker A (step S7). Of the voice data stored in the voice data file 42, only the voice data that could not be retrieved from the voice DB 43 is converted into character data. Additional information includes items such as a checksum. The data structure of the voice DB 43 will be described later.

  Then, the phonetic transcription server 5 determines whether or not to perform calibration by the operator from the state of the operator calibration flag of the speaker DB 41 based on the accuracy of speech recognition and the contract conditions of the speaker A (steps S8 and S9). When the operator calibration flag is ON (value “1”), the phoneticization server 5 sends the calibration request information including the speaker ID of the speaker A, voice data, converted character data, additional information, etc. to the calibration request queue 49. (Step S10). One or a plurality of operators calibrate as necessary and register in the voice DB 43, and the registered calibration request information is deleted from the calibration request queue 49 (step S11).

  In addition, when the operator specifies the calibration of the character data based on the voice data already registered in the voice DB 43 in the contract condition of the speaker A, the operator calibrates the character data stored in the voice DB 43 at a predetermined timing. You may make it do. When correcting the character data stored in the speech DB 43 by calibration, a calibration pattern by the operator may be accumulated. Moreover, old data may be deleted for the purpose of improving the search efficiency of the voice DB 43. The calibration operation also includes a method that allows the operator to customize options for the speech recognition function and the language analysis function in order to increase conversion accuracy.

  Since the speech recognition function and the language analysis function are not completely capable of analysis, the speech data is converted by mutual complementation of the speech recognition function and the language analysis function, and information such as calibration work and calibration patterns by humans such as operators is added. Thus, it is possible to feed back at the time of the next voice processing and increase the voice recognition and language analysis rate. As a result, for example, even if the voice data is recognized as “Suizokaniku” by the voice recognition function, it will be converted to “Go to the aquarium” in the same way even if it is recognized as “Going to Saizokan”. Can be.

  In addition, an interface is provided for the speaker A to check the character data converted from his / her voice data stored in the voice DB 43 after authentication from his / her mobile phone 31 or PC 6, and the already converted character data is calibrated. You may make it do.

  The character data corrected by the operator in step S11 is stored in the transmission standby DB 47 together with the speaker ID of the speaker A, and the transmission timing condition stored in the speaker information of the speaker A is satisfied in the speaker DB 41. Then, the character conversion server 5 transmits the character data according to the transmission method stored in the speaker information of the speaker A from the speaker DB 41 (step S12).

  The character data stored in the transmission standby DB 47 in association with the speaker ID of the speaker A is the speaker information of the speaker A when the speaker A is authenticated by requesting a service that provides character data. Character data is provided based on the transmission timing and transmission method stored in the. The transmission timing is specified in real time, in units of n characters, after the call is completed, or the like. As a transmission method, information provision via the Web, provision via electronic mail, or the like is designated. After the call by the speaker A is completed, the character data converted from the voice data by the call is transmitted to the speaker A by e-mail. It is also possible for the utterer A to designate the e-mail of the utterer B and to transmit the character data to the utterer B.

  In the above description, the processing for the voice data of the speaker A has been described, but the same applies to the voice data of the speaker B who is the other party. The same processing is also performed when the speaker A or B dials the device 33 having an incoming function of a service provider that requires voice recognition.

  Moreover, although the example which collects audio | voice data using the mobile telephones 31 and 32 was demonstrated, the installation-type telephone installed in the predetermined position which does not go through a network like a public telephone and a home telephone, IP telephone using a network Etc. are also applicable.

  FIG. 2 is a flowchart for explaining a voice registration process in which voice data is converted into character data and registered. In FIG. 2, the phonetic transcription server 5 acquires the telephone numbers or user IDs of the sender and receiver (step S51). For example, when the speaker A and the speaker B make a call, the phonetic transcription server 5 acquires the telephone number of the speaker A and the telephone number of the speaker B from the exchanges 11 and 12 on the respective sides. On the other hand, when a user who has been registered as a speaker accesses the phonetic character server 5 to receive provision of converted character data, the phonetic server 5 obtains the speaker ID from the user.

  The phonetic transcription server 5 searches and acquires speaker information from the speaker DB 41 using the telephone number or speaker ID acquired in step S51 (step S52). The phonetic transcription server 5 determines whether or not the speaker information can be acquired and the use of the voice data to text data conversion service is desired (step S53). If the speaker information could not be acquired, or if it was acquired but the user did not wish to use the voice data to character data conversion service, the voice characterizing server 5 is not registered by the speaker (or user). This voice registration process is terminated.

  On the other hand, when the speaker information can be acquired, the phonetic transcription server 5 determines that the speaker (or user) has been registered, and determines whether or not it is in a call state (step S54). For example, when the information acquired in step S51 is a telephone number, if the notification of call termination is received from the corresponding central office switch 11 or 12, the determination of the call state proceeds to step S54-2. .

  On the other hand, if the call end notification has not been received, it is determined that the call is in progress, audio data is taken in (step S55), and analog audio data is converted to digital audio data (A / D conversion) (step S55). S56). If the voice data received from the central office 11 or 12 is provided digitally, step S56 is omitted.

  The voice characterizing server 5 analyzes the voice data, divides the part without the voice into frames, and creates the voice data file 42 for each frame (step S57). The voice characterizing server 5 calculates the checksum by adding the digital values of the voice data for each voice data file 42 and sets the checksum in the voice data file 42 (step S58). Then, for each voice data file 42, the phonetic transcription server 5 registers the same voice data in the voice DB 43 using the speaker ID of the speaker information acquired in step S 52 and the checksum calculated in step S 58. It is confirmed by searching the voice DB 43 whether or not there are (steps S59 and S60).

  When it is confirmed that the voice data is registered in the voice DB 43 for all the voice data files 42, the process proceeds to step S60-2, and after performing the character data queuing process shown in FIG. 5, the process returns to step S54 and described above. Similar processing is repeated.

  On the other hand, when there is a file in which the voice data registration could not be confirmed in the voice data file 42, the voice characterizing server 5 determines whether or not the voice data of the voice data file 42 is a subdivided file. Is determined (step S61). If the file has been subdivided, the process proceeds to step S62. On the other hand, when the audio data file 42 stores the audio data that has not been subdivided, the audio characterizing server 5 subdivides the audio data corresponding to one stored frame by a predetermined dividing method to generate the audio data. A data file 42 is created (step S61-2). In this case, a flag indicating subdivision may be set in a predetermined storage area of the audio data file 42. As the predetermined separation method, for example, there is a method of dividing in a predetermined time axis, number of bytes, or character unit. After that, the phonetic character server 5 returns to step S58 and repeats the same processing as described above, such as calculating the checksum for the voice data file 42 by subdivision by the method described above.

  If the registration of the voice data cannot be confirmed even after the subdivision, the voice characterizing server 5 uses the existing voice recognition function to convert the voice data into characters and creates character data from the voice data (step S62). Then, it is registered in the voice DB 43 in association with the speaker ID (step S63).

  Then, the phonetic transcription server 5 analyzes the character data to determine whether or not the voice recognition is incomplete, and checks the operator calibration flag of the speaker information acquired in step S52 to provide a calibration service. It is determined whether or not it is desired (step S64). When the voice recognition is incomplete and the proofreading service is desired, the phonetic transcription server 5 sends the proofreading request information including the utterer ID, the voice data, the converted character data, and the additional information to the proofreading request queue 49. (Step S65), the character data queuing process is executed (step S65-2), the process returns to step S54, and the same process as described above is repeated. When the speech recognition is completely performed, or when the speech recognition is incomplete but the calibration service is not desired, the speech characterizing server 5 does not perform step S65 and performs the character data queuing process. After the execution (step S65-2), the process returns to step S54 and repeats the same processing as described above.

  FIG. 3 is a diagram illustrating an example of a method for quantizing analog audio data. In FIG. 3A, the analog audio waveform 2p is set as a function of time F (t), and then time point sequences T0, T1, T2,... Tn are taken along the time axis, and the peak values at each store are taken. Sampling for reading F (tk) is performed. A value obtained as a result of the sample is called a sample value. Next, in FIG. 3B, since the crest value as the sample value is a continuous quantity (analog), there is generally a value after the decimal point. However, the crest value is approximated by an integer value closest to the crest value. Integer conversion is performed. This is called quantization.

  For example, the waveform 2p of the analog voice has a peak value of 1, 9, 13, 13, 10, 6, 6, 6, 7, 5, 1 at times T0, T1, T2,. Digitized by value.

  By sampling and quantizing as described above, the original analog speech waveform 2p can be expressed as a set of appropriate integer values. By replacing the integer values with an electric pulse train and performing A / D conversion, the original analog speech waveform 2p can be expressed. The voice waveform can be handled as digital voice data in which corresponding electrical pulses are collected.

  FIG. 4 is a diagram for explaining a process of dividing audio data. In the example of the audio data shown in FIG. 4A, as shown in FIG. 4B, the audio data is divided for each frame with a portion where there is no audio as a delimiter. The frames A1 to An obtained by dividing the audio data are stored in the respective audio data files 42. For example, “Good morning”, “It is Yoshioka”, etc. are stored in the audio data file 42, respectively. The determination of the portion without sound may be made when the quantized digital value is equal to or smaller than a predetermined value as shown in FIG.

  A checksum obtained by adding the digital values quantized for each frame A1 to An is set in each audio data file 42.

  Next, processing for transmitting character data will be described with reference to FIGS. FIG. 5 is a flowchart for explaining the character data queuing process called from the voice registration process shown in FIG. In FIG. 5, the phonetic transcription server 5 acquires the speaker information from the speaker DB 41 using the speaker ID (step S71). Then, the voice characterizing server 5 determines whether or not there is a transmission request for character data (step S72). When there is no character data transmission request, the voice characterizing server 5 ends the character data queuing process and returns to the voice registration process (FIG. 2). On the other hand, when there is a transmission request for character data, the phonetic character server 5 analyzes the transmission timing specified by the speaker information (step S73).

  When the transmission timing is “(a) transmission after a specified time from the utterance”, the phonetic transcription server 5 adds the time set to the current time to determine the transmission time (step S74). If the transmission timing is “(b) Immediate transmission”, the voice characterizing server 5 sets the transmission time to the current time (step S75). When the transmission timing is “(c) after the call ends”, the phonetic transcription server 5 sets the transmission time to the current time when the call ends, and sets it to NULL when the call is not ended. (Step S76). When the transmission timing is “(d) When the specified number of characters has been reached”, the voice characterizing server 5 determines that the total number of characters to be transmitted and the number of characters set in the transmission standby DB 47 is the number of characters set in advance by the user. If it has reached, the transmission time is set to the current time (step S77).

  After setting the transmission time, the phonetic transcription server 5 refers to the speaker information and determines whether or not to feed back characters that cannot be completely recognized (step S78). If feedback is not specified in the speaker information, the phonetic transcription server 5 proceeds to step S80. On the other hand, if feedback is designated, the phonetic character server 5 changes the font or color different from the recognized character to the unrecognizable character and highlights it so as to stand out (step S89). In addition, the voice characterizing server 5 also transmits the voice data of that portion if necessary.

  Furthermore, the phonetic transcription server 5 refers to the speaker information and determines whether or not there is an encryption designation (step S80). If there is no designation of encryption, the phonetic transcription server 5 proceeds to step S82. On the other hand, if there is an encryption designation, the phonetic character server 5 encrypts the character data by a predetermined method (step S81).

  Thereafter, the voice characterizing server 5 queues the transmission waiting DB 47 by storing the transmission information including the character data, the transmission time, and the speaker information (step S82), and is executed from the voice registration process. This character data queuing process is terminated, and the process returns to the caller's voice registration process.

  FIG. 6 is a flowchart for explaining the character data transmission process. The character data transmission process shown in FIG. 6 is repeated until the phonetic character server 5 is stopped at a predetermined interval. The phonetic server 5 searches for transmission information related to character data queued in the transmission standby DB 47 for transmission information whose transmission time is the same as or has elapsed (step S91). The phonetic transcription server 5 identifies the transmission method specified by the speaker information for each searched transmission information (step S92).

  When the transmission method indicates “(a) save character data in file”, the phonetic character server 5 saves the character data in a file and sends it to the destination specified by the speaker information (step S93). When the transmission method indicates “e-mail”, the voice characterizing server 5 transmits the contents of the call by e-mail as characters. When the transmission method indicates “(c) RSS (Rich Site Summary)”, the voice characterizing server 5 transmits characters by RSS (step S95). In this case, the phonetic transcription server 5 may perform screen display settings. When the transmission method indicates “(d) electronic bulletin board, CHAT, etc.”, the voice characterizing server 5 transmits character data to a specific server (step S96). In this case, a qualified person who can access with a specific server can refer in real time.

  After providing the character data by the specified transmission method, the phonetic character server 5 returns to step S91 and repeats the above-described processing until the phonetic character server 5 is stopped.

  Next, character data proofreading processing by a user registered as a speaker will be described. In this case, the phonetic character server 5 operates as a Web server and provides a service for the user to update the character data. FIG. 7 is a flowchart for explaining the proofreading process of the character data stored in the voice DB by the user. In FIG. 7, when a user registered as a speaker accesses a service for proofreading character data provided by the phonetic character server 5 from the browser of the PC being used (step S101), the phonetic character server 5. Requests the speaker ID and password from the user, and performs user authentication using the speaker DB 41 using the speaker ID and password acquired from the user (step S102).

  Next, using the speaker ID, the phonetic character server 5 searches the voice DB 43 for previously registered character data (step S103), and displays the list 8a on the browser in the order specified by the user (step S103). Step S104). The order in which the list 8a is displayed includes, for example, the order of registration in the speech DB 43, the order of recently used character data, the order of character data for which speech recognition has been incomplete. If the character data includes special characters such as “**”, it indicates that the speech recognition is incomplete. The order may be designated from a screen displaying a list on the browser, or a desired order may be acquired from the user in advance after user authentication.

  From the screen displaying the browser list 8a, the user clicks the play icon that is paired with the character converted by the voice recognition to play the voice, and compares it with the recognized character. If there is, the proofread character is input to the proofread character string (step S105). If the user can update the text after proofreading, the user clicks the update button 8b (step S106).

  In response to these operations by the user, the character data calibrated from the browser is transmitted to the voice characterizing server 5, and the voice characterizing server 5 updates the voice DB 43 with the character data based on the character string input by the user. .

  FIG. 8 is a flowchart for explaining the proofreading process by the operator of character data after automatic recognition registered in the proofreading request queue. 8, when the character data to be proofread is registered in the proofreading request queue 49, the phonetic transcription server 5 searches for a standby operator terminal among a plurality of operator terminals (step S201). The phonetic transcription server 5 determines whether or not there is an operator terminal in a standby state by the search (step S202). If there is no operator terminal in a standby state, the voice characterizing server 5 returns to step S201 and waits for new character data to be registered in the calibration request queue 49.

  On the other hand, when there is an operator terminal in the standby state, the phonetic transcription server 5 selects any one of the operator terminals in the standby state and defines that terminal as a use state (step S203). Then, the voice characterizing server 5 outputs an alarm for alerting the operator terminal, and when the voice to be calibrated is reproduced by the operator's headset, one or more character conversion candidates for the voice are further displayed on the operator terminal. (Step S204).

  The phonetic transcription server 5 determines whether one conversion candidate has been selected by the user (step S205). If the user selects a conversion candidate, the phonetic transcription server 5 proceeds to step S207 to update the voice DB 43 with the selected conversion candidate. On the other hand, when the user does not select a conversion candidate, the phonetic character server 5 accepts character input from the keyboard of the operator terminal instead of selecting the conversion candidate, or uses a voice recognition function and a language analysis function. The voice re-produced by the operator is converted into a character string (step S206).

  The voice characterizing server 5 updates the voice DB 43 with the character string calibrated by selection or input acquired from the operator terminal as character data (step S207), defines the operator terminal in a standby state (step S208), and It returns to step S201 and repeats the same process mentioned above.

  FIG. 9 is a diagram illustrating a table configuration example of a speaker DB registered as a user. In FIG. 9, a speaker DB 41 includes a speaker ID, a speaker password, a speaker registration date, an address, a telephone number, a mobile phone number, a PC identification code, a voice-to-character conversion service use flag, an operator calibration flag, and encryption support. It has items such as a flag, a transmission timing flag, a prescribed number of transmission timings, a transmission method, a transmission destination, and a feedback flag.

  The speaker ID is an ID for identifying a speaker as a user who uses the service provided by the phonetic transcription server 5. For example, a cellular phone number of the speaker such as “0800010001” is set. The speaker password is a character string for authentication set by the user when registering for use in the phonetic transcription server 5. The speaker registration date and time indicates the date and time of registration, for example, “0803281030”. The name of the speaker is set at the time of registration such that the name of the speaker as a user is “Fujitsu Taro”, for example. The address is set by the speaker at the time of registration, for example, “Lake Yamanaka”. The telephone number and the mobile phone number are set at the time of registration, for example, “042xxxxxxxx” and “0800010001”. As the PC identification code, for example, an IP address such as “01.109.xx.xx” or a MAC address is set.

  In the voice-to-character conversion service use flag, “1” is set when using the conversion service from voice data to character data, and a value other than “1” such as “0” is set when not using. Is done. In the operator calibration flag, “1” is set when the calibration service by the operator is used, and a value other than “1” such as “0” is set when the service is not used. In the encryption correspondence flag, “1” is set when the character data is encrypted, and a value other than “1” such as “0” is set when the character data is not encrypted.

  The transmission timing flag is set to “1” when transmitting a specified time after the utterance, set to “2” when transmitting immediately, and set to “3” when transmitting after the call ends. When the predetermined number of characters is reached, “4” is set when transmitting. The value set as the specified number of transmission timings indicates the specified time when the transmission timing flag is “1”, and indicates the number of characters when “4”.

  The transmission method is designated by any one of “FILE”, “MAIL”, “RSS”, and “SERVER”. Further, it is possible to set a transmission means such as a fax. As the transmission destination, a destination corresponding to the transmission method is set. For example, when the transmission method is “MAIL”, one or more e-mail addresses are designated. A plurality of transmission destinations such as the e-mail addresses of the caller and the other party may be set. The feedback flag is set to “1” when performing feedback when speech recognition is not completely completed, and is set to a value other than “1” such as “0” when not performing feedback.

  The telephone number, the mobile phone number, and the PC identification code are device specifying information for specifying the device used by the speaker.

  FIG. 10 is a diagram illustrating a table configuration example of a voice DB that stores character data converted from voice data. In FIG. 10, the voice DB 43 includes items such as a speaker ID, checksum, voice data information, recognized characters, post-proofreading characters, latest reference date, reference count, voice data registration date, calibration date and time.

  The speaker ID is a speaker ID registered in the speaker DB 41 as a user. When a mobile phone number is registered as a speaker ID in the speaker DB 41, the mobile phone number is set. In the checksum, a total value of digital values of the audio data is set. In the audio data information, information indicating the storage destination of the digitized audio data is set. If saved in a file, the file name is set.

  A character string automatically recognized from voice data is set as character data in the recognized character. In the post-proofreading character, a character string proofread by a user registered as an operator or a speaker is set as character data, and is blank when not proofreading. When registration of voice data is retrieved by the speaker ID and checksum, the character data set in the proofread character is used preferentially over the character data set in the recognized character, and the proofread character is blank. Use the character data of the recognized character.

  The latest reference date / time indicates the last date / time when the character data recognized or calibrated from the voice data stored in the storage destination of the voice data information is used. The reference count indicates the number of times this character data is used. The voice data registration date and time indicates the date and time when the voice data was converted into character data and registered. The calibration date and time indicates the date and time when the operator or user calibrated the character data stored in the recognized character.

  Next, a usage form to which the above-described phonetic transcription server 5 is applied will be described. FIG. 11 is a diagram illustrating an application example in a usage mode in which a mobile phone makes a call. In FIG. 11, the voice signals of the voice calls made by the speakers A and B using the mobile phones 2a and 2b are received by the mobile base stations 3a and 3b, and the relay exchanges 4a are connected via the telephone network 7a. And 4b are transmitted to both mobile phones 2a and 2b. The relay exchanges 4a and 4b are connected to the voice characterizing server 5 by a communicable data line network. After the voice characterizing server 5 confirms the authentication with a mobile phone number or the like, it further uses the voice → character conversion service use flag After confirming the use of the service, the voice data and the character data converted by the automatic recognition are stored in the voice characterizing server 5. The voice characterizing server 5 transmits character data to a transmission destination designated at a predetermined transmission timing.

  FIG. 12 is a diagram showing an application example in a usage mode in which a telephone call is made with an IP telephone. In FIG. 12, the voice signals of the voice calls made by the speakers A and B using the IP telephones 2c and 2d are transmitted and received via the IP network 7c by the wireless antennas 3c and 3e connected to an optical fiber or a digital dedicated line. . The voice characterizing server 5 connected to the IP network 7c authenticates and confirms the IP telephones 2c and 2d transferred from the routers forming the IP network 7c, respectively, and further uses the voice → character conversion service use flag. After confirming the use of the service, the voice data from the IP telephones 2c and 2d and the character data converted by the automatic recognition are stored in the voice characterizing server 5.

  The same applies when the speaker A or B makes a call from an IP telephone set in a public telephone box or a convenience store. In such a usage mode, authentication or the like may be performed using the PC identification code of the speaker DB 42 shown in FIG.

  FIG. 13 is a diagram illustrating an application example in a usage mode in which a telephone call is made by an IP phone via a P2P network. In FIG. 13, a PC terminal 62 that is a general node is connected to a P2P network 67 composed of a plurality of super nodes 61 after being authenticated by the phonetic transcription server 5 via the Internet 68, and via the super node 61. Connected to the other party's PC terminal 62, the IP telephone call is started.

  After the call is started, each PC terminal 62 transfers the voice data during the call to the voice characterizing server 5, thereby storing the voice data and the character data converted by the automatic recognition in the voice characterizing server 5.

  FIG. 14 is a diagram illustrating a functional configuration example of the phonetic transcription server. In FIG. 14, the phonetic transcription server 5 is a computer device including a CPU, a memory, a storage device, a display unit, an output unit, an input unit, a communication unit, an external storage device I / F, etc., and the CPU executes a program. A voice data reception processing unit 501, a user authentication and service use confirmation unit 502, a voice data A / D conversion unit 503, a frame division unit 504, a voice data registration confirmation unit 505, and character data. A conversion unit 506, an operator calibration unit 509, a user calibration unit 510, a character data distribution unit 511, a display processing unit 521, an input / output processing unit 522, a communication control unit 523, and an installer 524 are included. Further, the speaker DB 41, the voice data file 42, the voice DB 43, and the transmission standby DB 47 are held in the storage device. The audio data file 42 is a data file in which audio data is registered in the audio DB 43.

  The display processing unit 521 controls display of data on the display unit. The input / output processing unit 522 controls input / output of data to / from the input unit and the output unit. The communication control unit 523 controls data communication performed via the network. The installer 524 installs the program from the recording medium 520 in which the program according to the present invention is recorded via the external storage device I / F. The recording medium 520 may be a computer readable medium.

  The voice data reception processing unit 501 corresponds to steps S51 and S52 in FIG. 2. When voice data is received by the communication control unit 523, the voice data reception processing unit 501 searches the speaker DB 41 using the telephone number transmitted together with the voice data. The speaker information is acquired, and the voice data and the speaker information are stored in the working storage area. After the authentication, the stored voice data is taken in, A / D converted as necessary, and converted into character data. Further, when the speaker ID is notified from the communication control unit 523, the voice data reception processing unit 501 searches the speaker DB 41 using the speaker ID, acquires the speaker information, and obtains the speaker information. Store in the working storage area.

  The user authentication and service usage confirmation unit 502 corresponds to steps S53 to S54 in FIG. 2, and uses the speaker information acquired by the voice data reception processing unit 501 to access the speaker during a call or via the Internet. In addition to authenticating the user, the use of the voice data to character data conversion service is confirmed by referring to the voice to character conversion service use flag of the speaker information.

  The audio data A / D converter 503 corresponds to step S56 in FIG. 2, and converts analog audio data into digital audio data according to a predetermined algorithm shown in FIG. The frame dividing unit 504 corresponds to step S57 in FIG. 2, divides the audio data converted into digital data into frames, and calculates the checksum by adding the digital values for each divided frame.

  The audio data registration confirmation unit 505 corresponds to steps S58 to S60 in FIG. 2, and confirms registration of the audio data using the checksum calculated by the frame division unit 504. When the voice data is registered in the voice DB 43, it means that the converted character data is registered in the voice DB 43.

  The character data conversion unit 506 corresponds to steps S62 and S63 in FIG. 2 and is converted into character data using the speech recognition function 507 and the language analysis function 508 and then registered in the speech DB 63.

  The operator calibration unit 509 corresponds to steps S201 to S208 in FIG. 8, and is executed when it is determined that the operator wants calibration by referring to the operator calibration flag of the speaker information.

  The user calibration unit 510 corresponds to steps S101 to S107 in FIG. 7, and the voice registered in the voice DB 42 when a user who accesses through the network is authenticated by the user authentication and service usage confirmation unit 502. The data can be viewed, the proofreading of the converted character data is permitted, and the voice DB 42 is updated with the character data confirmed by the user.

  The character data distribution unit 511 corresponds to steps S71 to S82 in FIG. 5 and steps S91 to 96 in FIG. 6, and refers to the transmission timing flag of the speaker information, the prescribed number of transmission timings, the transmission method, the transmission destination, and the like. Then, the converted character data is transmitted.

  As described above, by using the phonetic character conversion server 5, even when the call is not clear due to noise or hearing loss, for example, it becomes easy to understand the call by providing converted characters.

  The voice characterizing server 5 searches the voice DB 43 with the checksum of the received voice data before converting the character data and confirms the registration of the voice data in the voice DB 43. The character data stored corresponding to the character data can be provided, and the voice data can be converted to the character data at a higher speed. Further, since the search is performed using the checksum value associated with the speaker ID, it is not necessary to analyze in detail the characteristics of the individual speaker's voice for conversion into character data.

  By allowing the operator to calibrate the character data with respect to the voice data during a call, it is possible to provide character data with higher accuracy in a timely manner. In addition, since it is possible to calibrate the character data directly by the speaker after the call, it is possible to provide more accurate character data.

  In addition, since the voice call server 5 can convert a telephone call into character data, the speaker has a voice recognition function and a language analysis function only by using a normal telephone or his / her own mobile phone. It is not necessary to provide a special device separately, and the place where such a special device is installed is not limited.

  In the embodiment, the example of conversion into Japanese has been described, but the present invention is not limited to Japanese, and English and other languages are also possible.

Regarding the above description, the following items are further disclosed.
(Appendix 1)
Voice data receiving means for receiving voice data of a speaker during a call via a communication line;
A voice character conversion means for converting the voice data into character data;
A voice characterizing server having distribution means for distributing the character data to the speaker;
(Appendix 2)
Voice registration means for registering the voice data by storing the checksum of the voice data and the character data converted from the voice data in the voice database in association with each other;
Voice registration judging means for judging whether or not the voice data of the speaker received by the voice data receiving means is registered in the voice database by matching the checksum of the voice data;
The voice characterizing server according to claim 1, wherein the distribution means distributes character data corresponding to the voice data having the matched checksum to the speaker.
(Appendix 3)
A frame unit dividing unit that divides the audio data received by the audio data receiving unit into frames for each generated syllable;
Checksum calculation means for calculating a checksum for each frame,
The phonetic transcription server according to claim 2, wherein the voice registration determining unit determines registration of the voice data in the voice database using the checksum calculated for each frame.
(Appendix 4)
When the audio registration determining means determines that the audio data is unregistered, the frame-by-frame audio data is subdivided by a predetermined dividing method, and the checksum calculating means The voice characterizing server according to supplementary note 3, wherein the checksum is calculated and the voice registration judging means judges whether or not voice data having the same checksum is registered.
(Appendix 5)
When the voice data is not registered in the voice database, the voice character conversion means converts the voice data into character data using a voice recognition function,
Calibration means for enabling conversion error due to the voice recognition function to be calibrated by an operator or the speaker.
The phonetic transcription server according to any one of appendices 2 to 4, further comprising: updating means for updating the voice database with the character data in which the conversion error is corrected.
(Appendix 6)
The distribution means can distinguish character data recognized by the voice recognition function from character data that could not be recognized when transmitting character data converted from the voice data of the speaker at a predetermined transmission timing. The phonetic transcription server according to appendix 5, which feeds back as follows.
(Appendix 7)
A speaker authentication means for authenticating the speaker using device specifying information for specifying a communication device to be received together with the voice data;
The phonetic transcription server according to any one of appendices 1 to 6, wherein when the authentication by the speaker authenticating unit is successful, the phonetic character converting unit and the distributing unit are validated.
(Appendix 8)
Speaker information management means for storing and managing speaker information including the device specifying information and the character data transmission method in a speaker database in association with a speaker ID for identifying the speaker;
The phonetic transcription server according to appendix 7, wherein the distribution means distributes the character data by means specified by the transmission method.
(Appendix 9)
A computer that functions as a phonetic transcription server
A voice data reception procedure for receiving voice data of a speaker during a call via a communication line;
A frame unit dividing procedure for dividing the audio data received by the audio data receiving procedure into frames for each generated syllable;
A voice character conversion procedure for converting the voice data into character data in units of frames;
A distribution procedure for distributing the character data to the speaker and a voice conversion method to be executed.

  The present invention is not limited to the specifically disclosed embodiments, and various modifications and changes can be made without departing from the scope of the claims.

It is a figure for demonstrating the outline | summary of a whole process. It is a flowchart for demonstrating the audio | voice registration process which converts and registers audio | voice data into character data. It is a figure which shows the example of a method of quantizing analog audio | voice data. It is a figure for demonstrating the process which divides | segments audio | voice data. It is a flowchart for demonstrating the character data queuing process called from the audio | voice registration process shown in FIG. It is a flowchart figure for demonstrating a character data transmission process. It is a flowchart for demonstrating the proofreading process of the character data preserve | saved by the voice DB by a user. It is a flowchart for demonstrating the proofreading process by the operator of the character data after the automatic recognition registered into the proofreading request queue. It is a figure which shows the example of a table structure of speaker DB registered as a user. It is a figure which shows the example of a table structure of audio | voice DB which preserve | saves the character data converted from audio | voice data. It is a figure which shows the example of application in the utilization form which talks with a mobile telephone. It is a figure which shows the example of application in the utilization form which makes a telephone call with an IP telephone. It is a figure which shows the example of application in the utilization form which makes a telephone call by IP telephone via a P2P network. It is a figure which shows the function structural example of a phonetic transcription server.

Explanation of symbols

5 Spoken text server 6 PC
11, 12 Office switch 21, 22 Radio tower 31, 32 Mobile phone 41 Speaker DB
42 Voice data file 43 Voice DB
47 Transmission standby DB
100 Speech transcription system

Claims (3)

Voice data receiving means for receiving voice data of a speaker during a call via a communication line;
Whether the voice data of the speaker received by the voice data receiving means is registered in a voice database in which a checksum of the voice data is associated with character data converted from the voice data. Voice registration judgment means for judging by the checksum match,
Distribution means for distributing character data corresponding to the voice data with the matched checksum to the speaker;
When there is no voice data having the same checksum in the voice database, voice character conversion means for converting the voice data into character data;
Voice registration means for registering the voice data by storing the checksum of the voice data and the character data converted from the voice data by the voice character conversion means in correspondence with each other in the voice database;
A phonetic transcription server. A frame unit dividing unit that divides the audio data received by the audio data receiving unit into frames for each generated syllable;
Checksum calculation means for calculating a checksum for each frame,
The voice registration determination means, the speech text of the server according to claim 1, wherein for determining the registration of the voice data to the voice database by using the checksum calculated in the frame. When the audio registration determining means determines that the audio data is unregistered, the frame-by-frame audio data is subdivided by a predetermined dividing method, and the checksum calculating means 3. The voice characterizing server according to claim 2 , wherein a checksum is calculated and the voice registration judging means judges whether or not voice data having the same checksum is registered.

Images