JP2006084965A - Voice data collecting device and program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To construct an enriched voice corpus in a short time at a low cost. <P>SOLUTION: A voice data collecting device includes a display section (112) which displays a text on a display device; a recording section (114) which starts sampling of voice signals from a microphone when a recording start is instructed while the text is being displayed and stores uttered voice data in a memory; a waveform display section (120) which displays voice waveforms based on the recorded uttered voice data when a recording completion instructing signal is received; and a preservation section (128) which stores the uttered voice data stored in the memory by relating the data to the text when a preservation instructing signal is received. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to an apparatus for recording a read-out voice of a predetermined text, and more particularly to an apparatus for efficiently collecting utterance voice data obtained by reading a text for a voice corpus.

  In speech-related technologies such as speech recognition and speech synthesis, statistical methods have become mainstream recently. In statistical methods, the speech corpus is important. In the statistical method, as the amount of speech data included in the speech corpus increases, processing with higher reliability becomes possible. Therefore, it is necessary to efficiently construct as large a speech corpus as possible.

  Conventionally, a speech corpus is constructed as follows. First, under the supervision of an expert, a speaker reads a series of texts continuously and records the speech. Furthermore, the audio data is manually separated by an expert for each utterance and stored as a separate file. A process of aligning each file stored in this manner and the text corresponding to the utterance is performed. According to the alignment result, information such as phoneme labels is attached to the audio data in a computer-readable format.

  When voice is digitized and used as basic data for statistical methods, noise must be avoided in the voice. In addition, when there is a large amount of text to be uttered, recording may be performed over a long period of time, and the sound level at the time of recording may change depending on the condition of the recording device and the physical condition of the speaker. As a result, the quality of speech accumulated in the speech corpus varies, which may be inappropriate as basic data for statistical methods. In addition, when a speaker makes a mistake in the text to be uttered, an error is mixed into the speech corpus that is a premise of the statistical method, which is a problem.

  For this reason, in the past, an expert carefully observed the recording status of the speaker, and while looking at the display of the specified measuring device, noise was mixed into the speech, the speech level deviated from the appropriate range, You need to check that you haven't read the text by mistake. And if appropriate recording was not performed, the speaker was made to repeat the recording from the beginning of the utterance.

  However, such a method has a problem that the burden on the supervisor becomes large during recording. For this reason, only one speaker can be recorded at a time, and when collecting voice data of multiple speakers, it is necessary to increase the number of supervisors or increase the time required for recording. . As a result, there is a problem that it is difficult to construct a complete speech corpus at a low cost in a short time.

  Therefore, one object of the present invention is to provide an audio data collection apparatus and a program therefor that can build a complete audio corpus in a short time and at a low cost.

  Another object of the present invention is to provide an audio data collection device and a program therefor that can reduce the burden on the supervisor and can construct a complete audio corpus in a short time at low cost.

  An audio data collection device according to a first aspect of the present invention is connected to a display device, a predetermined input device operable by a user, and a microphone, and collects audio data of an utterance corresponding to a predetermined text. A device for displaying text to be spoken on a display device, and responding to reception of a predetermined recording start instruction signal when the text to be spoken is displayed on the display device Then, in response to the voice recording means for starting sampling of the voice signal from the microphone and storing the sampled speech voice data in the first storage device, and a predetermined recording end instruction signal, Waveform display means for generating a speech waveform based on the utterance speech data stored in the storage device and displaying it on the display device, and a predetermined storage instruction from the input device In response to receiving signals, including speech data stored in the first storage device, and a storage means for storing in the second storage device in association with predetermined text.

  The text to be uttered is displayed on the display device. When the user speaks while watching the text, the voice is sampled and recorded. Furthermore, the voice waveform of the recorded utterance is displayed on the screen. The user can confirm the recording state by viewing this waveform. Therefore, by using this apparatus, it is possible to collect speech utterances for texts in a favorable form by user operations. Supervision by the supervisor is minimal.

  Preferably, the voice data collection device starts sampling the voice signal from the microphone in response to receiving the recording start instruction signal when the voice waveform is displayed on the display device, and the sampled speech Voice re-recording means for replacing the voice data stored in the first storage device with the voice data is further included.

  By generating the recording start instruction signal when the voice waveform is displayed, the voice data already recorded can be replaced with new voice data. It is possible to record utterances for the same text repeatedly until a favorable recording is obtained. As a result, it is possible to easily collect well-recorded speech voice data for text.

  More preferably, the audio data collection device is further connected to a speaker, and the audio data collection device is responsive to receiving a predetermined reproduction instruction signal when the audio waveform is displayed on the display device, Reproducing means for reproducing the utterance voice from the utterance voice data stored in the first storage device and giving the utterance voice to the speaker is further included.

  The recorded voice waveform is reproduced by the reproducing means. The user can easily determine whether or not recording has been successfully performed using the reproduced voice.

  More preferably, the voice data collection device further includes level judgment means for judging whether or not the speech voice data stored in the first storage device is within a predetermined signal level range, and the waveform display means In response to the recording end instruction signal, a voice waveform is generated based on the utterance voice data stored in the first storage device, and the signal level is within a predetermined signal level range according to the determination result by the level determination means. And means for displaying on the display device together with level determination information visually indicating whether or not there is.

  It is determined whether the recorded waveform is at an appropriate level, and the result is visually displayed. The user can determine whether the recording level is appropriate and can re-record if necessary. As a result, the collected utterance voice data has an appropriate level, and the quality of the utterance voice data is improved.

  The voice data collection device may further include an utterance portion detection means for detecting an utterance portion of the utterance voice data stored in the first storage device, and the waveform display means outputs a predetermined recording end instruction signal. In response, an utterance portion that generates a speech waveform based on utterance speech data stored in the first storage device and visually indicates the utterance portion of the speech waveform according to the detection result by the utterance portion detection means A means for displaying on the display device together with the marker may be included.

  The speech part and the non-speech area are separated and displayed together with the speech waveform. For example, even when noise is mistakenly recognized as an utterance or a portion that should originally be an utterance is not detected as an utterance portion, those errors can be easily confirmed. It becomes possible to re-record as necessary, and the quality of the collected voice data is improved.

  Preferably, the voice data collection device changes the utterance part for changing the utterance part detected by the utterance part detection means in response to the instruction to change the position of the utterance part marker given from the input device. Means are further included.

  If the utterance is detected in error, it can be corrected. It is possible to prevent errors in the extraction of utterances and alignment with text.

  More preferably, the voice data collection device performs alignment in a predetermined voice unit between the utterance voice data stored in the first storage device and the utterance target text displayed on the display device. Alignment means for generating alignment data indicating the results is further included, and the storage means is uttered voice data and alignment data stored in the first storage device in response to receiving the storage instruction signal from the input device. Is stored in the second storage device in association with the predetermined text.

  It is possible to automatically align the text and the speech voice data in a predetermined voice unit and save the alignment data. Aligned speech data can be created with this device alone. The speech data and alignment data can be used for learning an acoustic model.

  In response to receiving the save instruction signal from the input device, the save means stores the utterance voice data stored in the first storage device in the second storage device in association with a predetermined text. Means for displaying the text to be uttered on the display device may be included.

  The texts to be uttered are processed one by one, and utterance voice data for them is stored in the second storage device. There is no need to record the whole sentence at once and separate it later manually by conventional methods.

  The sound data collection program according to the second aspect of the present invention, when executed by a computer, causes the computer to operate as one of the sound data collection devices described above.

[Constitution]
FIG. 1 is a block diagram showing a configuration of an audio data collection system 30 including an audio data collection device 42 according to an embodiment of the present invention. Referring to FIG. 1, a voice data collection system 30 reads a text file from a text storage device 40 storing a speech file in which a plurality of texts read by a speaker are recorded, and reads out a text file from the text storage device 40. The voice data collection device 42 according to the present embodiment and the voice data file output by the voice data collection device 42 for creating a voice data file one sentence at a time by presenting each speaker to the speaker and the voice data file output from the voice data collection device 42 are stored. And an audio data file storage device 44.

  As will be described later, the voice data collection system 30 can be realized by a computer. In this case, the text storage device 40 and the voice data file storage device 44 are realized by a nonvolatile storage device such as a hard disk. Both may be realized by the physically same storage device.

  As the utterance text file, a plain text file is used in the present embodiment. One file contains a plurality of utterance texts. Each utterance text is separated by a line feed code. When reading text, the text in the file can be taken out one by one by reading the text from the file until a line break is encountered.

  The voice data collection system 30 is further connected to the voice data collection device 42. When the voice data collection device 42 presents the text to be spoken next to the speaker or the voice waveform of the recording result. Connected to a monitor 46 to be used and a voice data collecting device 42 and used when a speaker records voice, and connected to a voice data collecting device 42, the voice data collecting device 42 records the speaker. A speaker 50 used for reproduction and an input device 52 composed of a mouse, a keyboard, and the like, which are used by a speaker to give instructions to the voice data collection device 42 are included.

  FIG. 2 shows a more detailed configuration of the voice data collection device 42 as a functional block diagram. Referring to FIG. 2, the audio data collection device 42 is connected to the bus 72 and the input device 52, and controls the operation sequence of each functional unit described below in accordance with an instruction given by the user via the input device 52. The sequence controller 70, the memory 76 and the memory 86, both connected to the bus 72, and the bus 72, and the text file designated by the user is read from the text storage device 40 and sent via the bus 72. A load module 74 for loading into the memory 76.

  The voice data collection device 42 further includes a display unit 78 for outputting information generated by each functional unit of the voice data collection device 42 to the monitor 46 according to an instruction from the sequence control unit 70, and a sequence control unit 70. And a recording unit 80 for recording the voice from the microphone 48 and storing it in the memory 86 together with the corresponding text. These are all connected to the bus 72.

  The audio data collecting device 42 is further connected to the recording unit 80 and the memory 86, and the function of detecting the audio part and the volume of the audio out of a predetermined range out of the audio signals recorded by the recording unit 80 and stored in the memory 86. A volume check unit 82 for performing a function for determining whether or not the recording is performed, and a recording unit 80, a volume check unit 82, and a bus 72, between the voice recorded by the recording unit 80 and the corresponding text. A Viterbi alignment unit 84 that performs Viterbi alignment and associates the speech with the phonemes of the text stored in the memory 76 to create alignment data.

  The voice data collection device 42 further responds to a correction input of a marker for a utterance part, which will be described later, given by the user via the input device 52 and the sequence control unit 70, and the utterance position in the voice information stored in the memory 86. In response to receiving an audio reproduction instruction from the adjustment unit 88 and the sequence control unit 70, the audio data is read from the memory 86, converted into an analog audio signal, and sent to the speaker 50. In response to an instruction to store audio data from the reproduction unit 90 and the sequence control unit 70, the audio data, corresponding text data, alignment data, and other related data held in the memory 86 are audio A storage processing unit 92 for writing to the data file storage device 44.

  FIG. 3 is a flowchart showing an operation sequence of the audio data collection device 42 realized by the sequence control unit 70. The sequence control unit 70 controls each part of the voice data collection device 42 so that the operation status of the voice data collection device 42 changes according to the flowchart shown in this figure. The operation of the voice data collection device 42 can also be explained with this figure.

  With reference to FIGS. 3 and 2, according to this sequence, first, in step 110, the load module 74 is controlled to read the utterance text file from the text storage device 40 and load it into the memory 76. In step 112, the next utterance text (first utterance text immediately after loading) is read from the memory 76, and the display unit 78 is controlled and displayed on the monitor 46.

  Subsequently, in response to the recording instruction from the user, the process proceeds to step 114 where the recording unit 80 is controlled to sample the audio signal given in the form of an electric signal from the microphone 48 and record it in the memory 86 in a predetermined data format. . This sampling is performed with a frame length of 30 milliseconds and shifting the frame position every 10 milliseconds.

  When the recording is completed, the recorded data stored in the memory 86 is checked in step 116, and the volume check unit 82 determines whether or not the utterance level exceeds the appropriate level. Further, the volume check unit 82 determines where the utterance portion is in the recorded data based on the amplitude of the voice waveform. In step 118, the Viterbi alignment unit 84 associates the recorded data with the corresponding utterance text, and the alignment information is attached to each frame of the audio data as a label and stored in the memory 86.

  Next, in step 120, the display unit 78 is controlled to display the waveform of the audio data stored in the memory 86 on the monitor 46 together with the corresponding text. At this time, if a part with an excessive audio level is detected by the volume check, the audio waveform is clipped at an appropriate level and a frame indicating the appropriate level is displayed on the monitor 46 in red. In addition, a marker indicating the utterance area detected by the volume check unit 82 in the voice waveform is displayed on the monitor 46. After that, it waits for an instruction from the user.

  Although not shown in FIGS. 2 and 3, the audio data collection device 42 can switch between two operation modes: a reproduction mode and a speech area correction mode. In the playback mode, audio playback by the playback unit 90 can be performed on the part with the marker. In the utterance area correction mode, the utterance area can be corrected by correcting the marker position.

  In step 122, it is determined what the input from the user is. If the operation mode is the reproduction mode and a marker is input, the process proceeds to step 124 where the designated marker portion of the audio waveform is reproduced using the reproduction unit 90 and the speaker 50. Thereafter, the process returns to step 120 to display the waveform and wait for user input.

  In step 122, if the operation mode is the speech area correction mode and the marker is input, the process proceeds to step 126, where the voice data stored in the memory 86 is corrected using the adjustment unit 88 and designated. Match to the utterance area. Thereafter, the process proceeds to step 124. Subsequent processing is as described above.

  If a later-described save (SAVE & NEXT) button is pressed in step 122, the voice data file storage device 44 collects the voice data, corresponding text, alignment data, labels, etc. held in the memory 86 together. Write to. Thereafter, in step 130, it is determined whether or not the processing for all utterance texts in the utterance text file loaded in the memory 76 has been completed. If it has been completed, the process ends. If not completed, the process returns to step 112, and the next utterance text is read from the memory 76 and displayed. Thereafter, the above processing is repeated for each utterance text.

  If it is determined in step 122 that the “record” button has been pressed, the process returns to step 114 to repeat recording using the same utterance text again.

  The above is the contents of the operation sequence of each unit of the audio data collection device 42 realized by the sequence control unit 70. From the above description, the general operation of the audio data collection device 42 may also have been clarified.

[Operation]
The operation of this apparatus is as described with reference to FIG. Therefore, details thereof will not be repeated here.

[Realization by computer]
-Control structure of computer program-
The voice data collection device 42 described above can be realized by computer hardware and a computer program executed on the computer. FIG. 4 and FIG. 5 show a schematic control structure of a program for this purpose in a flowchart format. This program adopts a GUI (graphical user interface). Therefore, when a user operates a GUI component (an object such as a button or a menu item) displayed on the monitor 46, a program (method) defined in advance for the operation of the GUI component is executed. A mechanism for calling and executing a corresponding method by a user operation includes an OS (operating system), a user program, a module installed separately from the OS or the user program, and dynamically called when the program is executed, and the like. In some cases, it is realized by cooperation with a virtual computer environment or the like operating on the OS.

  Referring to FIG. 4, first, in step 150, a file open dialog is displayed. Here, a file open dialog prepared by the OS may be called. So-called filter processing may be performed so that only text attribute files are displayed in the dialog. Filter processing is a function provided by many OSs.

  Subsequently, in step 152, it is determined whether or not the user cancels the opening of the file in the file open dialog and selects the end of the process. If exit is selected, the program is terminated. In other cases, that is, when a file is selected, the process proceeds to step 154.

  In step 154, the file specified in the dialog is loaded into the memory 76. In step 156, an attempt is made to read the utterance text from the file loaded in the memory 76. In step 158, as a result of the processing in step 156, it is determined whether an EOF (End Of File) mark indicating the end of the file has been read. When the EOF is read, it means that there is no utterance text to be processed, and the process is terminated. If the utterance text has been successfully read, the process proceeds to step 160.

  In step 160, the save button is disabled so that it cannot be operated. In addition, a recording button is made available so that the user can instruct recording. Further, the operation mode of the audio data collection device 42 is set to the reproduction mode.

  Thereafter, in step 162, an initial recording screen is displayed. On this screen, the read utterance text is displayed and a user operation is waited for.

  Subsequently, referring to FIG. 5, it is determined in step 164 whether any event has occurred. Here, an event refers to a notification given to this program from the OS or the like when a user operates one of the operable GUI components or a module issues a message. Here, it is assumed that any of the recording (START) button operation, recording end (STOP) button operation, mode switching operation, marker input operation, and save (SAVE & NEXT) button operation occurs as an event. To do. Control branches to steps 170, 180, 200, 202, and 210, respectively, corresponding to these events.

  In step 170, the recording button is toggled to the recording end button on the screen. That is, the display of the recording button is changed to the recording end button, and the function is switched from the recording button to the recording end button. This process disables the record button and enables the record end button.

  Subsequently, recording is started in step 172, and the process returns to step 164. Specifically, the recording is performed by calling a function prepared by the OS or the like for recording using an API (Application Programming Interface).

  If the recording end button is operated, in step 180, the recording end button is toggled to the recording button. In step 182, the recording is terminated. That is, in step 172, the function called by the recording API is stopped using another API.

  In step 184, the level of the recorded voice data is checked, and the utterance area is detected and a label indicating the utterance area is assigned to the voice data. Subsequently, at step 186, the level of the audio data is examined to determine whether or not there is a part that deviates from an appropriate level range. If there is a deviation, a volume check label is attached to the frame of the audio data.

  In step 188, Viterbi alignment is performed between the text corresponding to the utterance and the voice data, and it is determined which part of the voice data corresponds to which phoneme of the text. Depending on the determination result, each frame of audio data is labeled to indicate the corresponding phoneme. This alignment process requires a learned acoustic model or the like using the speaker's voice as learning data, but a known algorithm can be used for the alignment itself. Furthermore, in this embodiment, since the text corresponding to the voice data is known, alignment can be performed more easily.

  In step 190, a voice waveform obtained by visualizing the recorded voice data is displayed on the monitor 46. At this time, the display of other parts of the screen is also refreshed. At the same time, a marker indicating the speech area is displayed. If there is a part that deviates from the proper level as a result of the volume check, the entire waveform is clipped with the proper level as the upper limit or lower limit, and a rectangle indicating the proper level is displayed so as to surround the waveform. Thereafter, the process returns to step 164.

  If it is determined in step 164 that the event is a mode switching event, the operation mode is toggled to the correction mode if the operation mode is the reproduction mode, and to the reproduction mode if the operation mode is the correction mode. Thereafter, the process returns to step 164.

  If it is determined in step 164 that the event is a marker input, it is determined in step 202 whether or not the operation mode of the audio data collection device 42 is the playback mode. If it is in the reproduction mode, the part indicated by the marker in the audio data is reproduced in step 206 and the process returns to step 164. If the playback mode is not set, the marker indicating the speech area is corrected in step 204 and the process proceeds to step 206. In step 206, the marker portion, that is, the speech area is reproduced, and the process returns to step 164.

  If it is determined in step 164 that the save button has been operated, the text and the aligned and labeled audio data are stored in the audio data file storage device 44 in step 210. Thereafter, the process returns to step 156 in FIG. 4, the next text is read, and the processes after step 158 are repeated.

  The above is the schematic control structure of the computer program for realizing the audio data collection device 42 according to the present embodiment by a computer.

-Computer hardware-
FIG. 6 shows the external appearance of a computer system 330 that implements the audio data collection system 30 according to this embodiment, and FIG. 7 shows the internal configuration of the computer system 330.

  Referring to FIG. 6, the computer system 330 includes a computer 340 having an FD (flexible disk) drive 352 and a CD-ROM (compact disk read only memory) drive 350, a keyboard 346, a mouse 348, and a monitor 342. , And microphone 370 and speaker 372. Of these, the keyboard 346 and the mouse 348 correspond to the input device 52 shown in FIGS. 1 and 2. The monitor 342, microphone 370, and speaker 372 correspond to the monitor 46, microphone 48, and speaker 50 shown in FIGS.

  Referring to FIG. 7, in addition to FD drive 352 and CD-ROM drive 350, computer 340 includes CPU (central processing unit) 356 and bus 366 connected to CPU 356, FD drive 352 and CD-ROM drive 350. And a read only memory (ROM) 358 for storing a boot-up program and the like, and a random access memory (RAM) 360 connected to the bus 366 for storing a program command, a system program, work data, and the like. The computer system 330 further includes a printer (not shown).

  Although not shown here, the computer 340 may further include a network adapter board that provides a connection to a local area network (LAN).

  A computer program for causing the computer system 330 to operate as the voice data collection system 30 (and the voice data collection device 42) is stored in a CD-ROM 362 or FD 364 inserted in the CD-ROM drive 350 or FD drive 352. And further transferred to the hard disk 354. Alternatively, the program may be transmitted to the computer 340 through a network (not shown) and stored in the hard disk 354. The program is loaded into the RAM 360 when executed. The program may be loaded directly into the RAM 360 from the CD-ROM 362, from the FD 364, or via a network.

  This program includes a plurality of instructions that cause the computer 340 to operate as the audio data collection device 42 of this embodiment. Some of the basic functions required to perform this method are provided by an OS or third party program running on the computer 340 or various toolkit modules installed on the computer 340. Therefore, this program does not necessarily include all functions necessary for realizing the system and method of this embodiment. If this program includes only the instructions for realizing the above-described voice data collecting device 42 by calling an appropriate function or “tool” in a controlled manner so as to obtain a desired result, Good. The operation of computer system 330 is well known and will not be repeated here.

-GUI screen-
FIG. 8 shows a GUI screen 180 displayed on the audio data collection system 30 in the present embodiment. Referring to FIG. 8, this GUI screen 180 includes a file menu 190 arranged in the menu area, a save button 192 operated when saving the file, a level meter 194 indicating the recording level, and recording / recording. And an end button 196. The recording / recording end button 196 is toggled by the program to be a recording button when recording is possible and to be a recording end button during recording. Also, by clicking on the file menu 190, an operation mode switching pull-down menu including items of both the reproduction mode and the correction mode is displayed. By selecting either one, the operation mode of the audio data collection device 42 toggles between the reproduction mode and the correction mode.

  The GUI screen 180 further includes an ID (identification name) display area 198 of the speech text being processed, a display area 200 of the speech text being processed, a display area 202 of the file name where the speech text is being processed, and a recording. A waveform display area 204 that displays the waveform 206 of the voice data with respect to the time axis together with the level, and a storage area marker 208 that indicates a portion of the voice data that is stored as speech data. The storage area marker 208 includes an utterance area 210 detected by the volume check unit 82 and a silent area 212 having a predetermined length before and after the utterance area 210. In the present embodiment, the length of each of the two silent areas 212 is selected to be half the length of the speech area 210.

  In the present embodiment, the save destination file name is a predetermined character string (this can be specified by the operator. In the case of FIG. 8, “speechfile_demo”), the speech text ID is added, and the sampling rate The name of the file with the extension added is the saved file name.

  Hereinafter, the state of the GUI screen 180 in each aspect of the operation will be described. FIG. 9 is an example of a screen displayed in step 162 of FIG. The file menu 190 is enabled, and the record / record end button 196 is set as a record button. In the ID display area 198, the display area 200, and the storage destination file name display area 202, the ID of the utterance text currently processed, the utterance text, and the file name in which the utterance text to be processed are stored are displayed. Has been.

  FIG. 10 is a display example of the GUI screen 180 during recording. During recording, the recording / recording end button 196 functions as a recording end button and does not function as a recording button. During recording, the level meter 194 indicates the recording level in real time.

  As shown in FIG. 9, when recording is possible, the record / record end button 196 is a record button and does not function as a record end button. Also, as shown in FIG. 10, during recording, the recording / recording end button 196 functions as a recording end button and does not function as a recording button. Therefore, the program should be processed on the assumption that these buttons are operated only at appropriate times.

  FIG. 11 is a display example of the GUI screen 180 at the end of recording. The recording / recording end button 196 is again a recording button. When the marker of the utterance area 210 is moved on this screen, the voice of the marked area is reproduced in the reproduction mode according to the operation mode, and the marker itself in the utterance area 210 is updated to the input range in the correction mode. At the same time, the sound in that area is reproduced.

  FIG. 12 is a display example of the GUI screen 180 when there is a portion 220 with an excessive audio level. As shown in FIG. 12, when the audio level becomes excessive, the audio waveform is clipped at an appropriate level, and a rectangle 222 indicating the range of the appropriate level is displayed in red. Therefore, the operator can immediately understand that the recording level is not appropriate, and can press the recording / recording end button 196 again to start recording again.

  FIG. 13 is a display example of the GUI screen 180 when noise 232 exists in addition to the speech sound 230. Thus, when noise has been recorded, it is displayed in an easy-to-understand manner, so that the user can easily determine whether or not to record again. Also, if the noise is outside the storage area, it is easy to see that there is no need to re-record, saving time and effort during recording.

  As described above, according to the voice data collection system 30 and the voice data collection device 42 according to the present embodiment, it is possible to interactively proceed with the speaker while displaying the utterance text during voice recording. it can. Since the voice waveform is displayed together with the display of the automatically detected speech area, it can be easily corrected when the automatic detection is wrong. In addition, since the voice can be immediately played back in units of utterances or in designated areas in the utterance, it is possible to easily confirm utterance mistakes. Waveforms are displayed so that noise can be checked immediately. In addition, it is displayed in a manner that can immediately confirm whether or not the sound level at the time of recording is appropriate. Therefore, it can be immediately determined whether or not the recording has been properly performed, and high-quality audio data can be obtained.

  According to the voice data collection system 30 and the voice data collection device 42 described above, since it is possible for the speaker to easily determine whether or not proper recording has been performed, it is not necessary for the supervisor to always pay attention to the recording. The burden on the supervisor is greatly reduced. For this reason, for example, recording by a plurality of speakers can be progressed while being supervised by a single supervisor using a plurality of devices, and the cost and time for creating a speech corpus can be saved.

  The embodiment disclosed herein is merely an example, and the present invention is not limited to the above-described embodiment. The scope of the present invention is indicated by each claim in the claims after taking into account the description of the detailed description of the invention, and all modifications within the meaning and scope equivalent to the wording described therein are intended. Including.

It is a block diagram which shows schematic structure of the audio | voice data collection system 30 which concerns on one embodiment of this invention. It is a figure which shows schematic structure of the audio | voice data collection apparatus 42 in a block diagram format. It is a figure which shows the processing sequence implement | achieved by the sequence control part 70 in a flowchart format. 4 is a flowchart of a program that realizes an audio data collection device. 4 is a flowchart of a program that realizes an audio data collection device. 1 is an external view of a computer system 330 that implements an audio data collection system 30. FIG. FIG. 7 is a block diagram of the computer system 330 shown in FIG. 6. 4 is a diagram showing a configuration of a GUI screen 180. FIG. It is a figure which shows the example of a display screen before a recording start. It is a figure which shows the example of a display screen during recording. It is a figure which shows the example of a display screen after completion | finish of recording. It is a figure which shows the example of a display screen when a recording level is inadequate. It is a figure which shows the example of a display screen when there exists noise.

Explanation of symbols

  30 voice data collection system, 40 text storage device, 42 voice data collection device, 44 voice data file storage device, 46 monitor, 48 microphone, 50 speaker, 52 input device, 70 sequence control unit, 72 bus, 74 load module, 76 Memory, 78 Display section, 80 Recording section, 82 Volume check section, 84 Viterbi alignment section, 86 Memory, 88 Adjustment section, 90 Playback section, 92 Storage processing section

Claims (9)

A voice data collection device for collecting voice data of an utterance corresponding to a predetermined text, connected to a display device, a predetermined input device operable by a user, and a microphone;
Text display means for displaying the text to be spoken on the display device;
In response to receiving a predetermined recording start instruction signal when the text to be uttered is displayed on the display device, sampling of the voice signal from the microphone is started, and the sampled utterance voice data is obtained. Voice recording means for storing in the first storage device;
In response to a predetermined recording end instruction signal, waveform display means for generating a voice waveform based on the utterance voice data stored in the first storage device and displaying it on the display device;
In response to receiving a predetermined storage instruction signal from the input device, the speech data stored in the first storage device is stored in the second storage device in association with the predetermined text. And a voice data collection device including a storage means. In response to receiving a recording start instruction signal when a voice waveform is displayed on the display device, sampling of the voice signal from the microphone is started, and the first utterance voice data is used as the first utterance voice data. 2. The voice data collection device according to claim 1, further comprising voice re-recording means for replacing utterance voice data stored in the storage device. The audio data collection device is further connected to a speaker;
In response to receiving a predetermined playback instruction signal when a voice waveform is displayed on the display device, the voice is reproduced from the voice data stored in the first storage device, and The audio data collection device according to claim 1, further comprising reproduction means for giving to a speaker. Further comprising level determination means for determining whether the speech data stored in the first storage device is within a predetermined signal level range;
The waveform display means generates a speech waveform based on the utterance voice data stored in the first storage device in response to the recording end instruction signal, and a signal according to the determination result by the level determination means The sound according to any one of claims 1 to 3, comprising means for displaying on the display device together with level determination information visually indicating whether or not a level is within the predetermined signal level range. Data collection device. Utterance part detection means for detecting an utterance part of the utterance voice data stored in the first storage device,
The waveform display means generates a speech waveform based on speech data stored in the first storage device in response to a predetermined recording end instruction signal, and according to a detection result by the speech portion detection means. The voice data collection device according to any one of claims 1 to 3, further comprising means for displaying on the display device together with an utterance portion marker that visually indicates an utterance portion of the voice waveform. The utterance part changing means for changing the utterance part detected by the utterance part detecting means in accordance with the change instruction in response to the instruction to change the position of the utterance part marker given from the input device. 5. The voice data collection device according to 5. The speech data stored in the first storage device and the speech target text displayed on the display device are aligned in a predetermined speech unit, and alignment data indicating the result is generated. Further comprising alignment means for
In response to receiving the save instruction signal from the input device, the save means associates the speech data and the alignment data stored in the first storage device with the predetermined text in a second The voice data collection device according to claim 1, comprising means for storing in a storage device. In response to receiving the save instruction signal from the input device, the saving means stores the speech data stored in the first storage device in the second storage device in association with the predetermined text. The voice data collection device according to claim 1, further comprising means for causing the display device to display a text to be spoken next. An audio data collection program that, when executed by a computer, causes the computer to operate as the audio data collection device according to any one of claims 1 to 8.

Images