CN102881285B - A kind of method of prosodic labeling and prosodic labeling equipment - Google Patents

Abstract

The embodiment of the present invention discloses a method of prosodic labeling and a special labeling device, the method includes: a receiving module, used to receive the audio data identification to be labeled, the audio data identification includes the batch and/or data of the audio data The entry number; the playback module, used to play the corresponding audio data to the tagger according to the audio data identification; the record module, used to record the tagging process with the tagger when it is detected that the tagger has triggered the tagging process Information related to the labeling behavior of the labeler; a generating module configured to generate audio labeling information of the audio data according to the information related to the labeling behavior of the labeler. Through the embodiments of the present invention, the labeled audio annotation data can be made more accurate, and then the accuracy or fluency of speech synthesis using the audio annotation data can meet actual requirements. The special labeling device provided by the embodiments of the present invention is also more suitable for blind people.

Description

A prosodic labeling method and prosodic labeling device

technical field

The present invention generally relates to the technical field of speech data processing, in particular to a prosodic labeling method and special labeling equipment.

Background technique

The sound resource library with prosodic annotation is an indispensable source of knowledge training in the field of speech recognition or text to speech (Text To Speech, TTS).

At present, when performing prosodic labeling in existing technologies, one method is to use raw corpus and punctuation information to generate a statistical probability model, and then perform prosodic labeling based on the model; however, because the generated statistical probability model is not accurate enough, the result of prosodic labeling is also poor. Not accurate enough; there is another existing technology that uses the user's real voice as training data to generate regular prosodic information according to the vocalization rules and the statistical phoneme length; but the user will appear tired, which will make the generated regular prosodic information Not generic enough.

In short, prosodic labeling using existing technologies cannot effectively generate accurate audio labeling information, and furthermore, the accuracy or fluency of speech synthesis does not meet actual needs.

Contents of the invention

In view of this, the embodiment of the present invention provides a prosodic labeling method and a special labeling device, which can conveniently generate accurate audio labeling information, and then make the accuracy or fluency of speech synthesis meet actual needs.

According to an aspect of an embodiment of the present invention, there is provided a prosodic tagging device, including: a receiving module, configured to receive an audio data identifier to be tagged, where the audio data identifier includes a batch and/or data entry number of the audio data; A module for playing corresponding audio data to the tagger according to the audio data identifier; a recording module for recording the tagging behavior related to the tagger when it is detected that the tagger triggers the tagging process information; a generating module, configured to generate the audio annotation information of the audio data according to the information related to the tagging behavior of the tagger, and the information related to the tagging behavior of the tagger is specifically trigger time point and pause duration information .

According to another aspect of the embodiments of the present invention, there is provided a method for prosodic labeling, including: receiving an audio data identifier to be labeled, the audio data identifier including the batch and/or data entry number of the audio data; The data identifier plays corresponding audio data to the annotator; when it is detected that the annotator triggers the annotation process, record the information related to the annotator’s annotating behavior; according to the information related to the annotator’s annotating behavior The audio annotation information of the audio data is generated, and the information related to the annotation behavior of the annotator is specifically trigger time point and pause duration information.

In addition, according to another aspect of the embodiments of the present invention, a storage medium is also provided. The storage medium includes machine-readable program code. When the program code is executed on the information processing device, the program code causes the information processing device to execute the above-mentioned prosodic tagging method according to the present invention.

In addition, according to yet another aspect of the embodiments of the present invention, a program product is also provided. The program product includes machine-executable instructions, and when the instructions are executed on the information processing device, the instructions cause the information processing device to execute the above-mentioned prosodic tagging method according to the present invention.

According to the above-mentioned method of the embodiment of the present invention, by playing the audio data to the annotator multiple times, the audio data annotation process can be triggered after the annotator is familiar with the audio data, and by repeating the annotation, it is possible to use this The audio annotation data obtained in the embodiment is more accurate, and thus the accuracy or fluency of speech synthesis using the audio annotation data can meet actual requirements. Moreover, it is also possible to perform credible weight processing on all audio annotation information marked by a certain annotator, so that the accuracy and accuracy of the audio annotation information can be further evaluated, thereby laying the foundation for subsequent applications such as speech synthesis.

Further aspects of the embodiments of the present invention are given in the description section below, wherein the detailed description serves to fully disclose preferred embodiments of the embodiments of the present invention without imposing limitations thereon.

Description of drawings

The above and other objectives and advantages of the embodiments of the present invention will be further described below in conjunction with specific embodiments and with reference to the accompanying drawings. In the drawings, the same or corresponding technical features or components will be indicated by the same or corresponding reference numerals.

Fig. 1 is a flow chart showing the first method of prosodic labeling provided as an embodiment of the present invention;

Fig. 2 is a flowchart showing step S102 as the first method embodiment;

Fig. 3 is a flowchart showing step S103 as the first method embodiment;

Fig. 4 is a flow chart showing a second prosodic tagging method provided as an embodiment of the present invention;

Fig. 5 is a flow chart showing a third prosodic tagging method provided as an embodiment of the present invention;

FIG. 6 is a flow chart showing step S506 as a third method embodiment;

FIG. 7 is a flow chart showing step S507 as a third method embodiment;

Fig. 8 is a schematic diagram showing a prosodic tagging device provided as an embodiment of the present invention;

Fig. 9 is a schematic diagram showing an interface as a dedicated labeling device in the prosody device embodiment;

Fig. 10 is a schematic diagram showing the recording module 603 in the prosodic device embodiment;

FIG. 11 is a block diagram showing an exemplary structure of a personal computer as an information processing device employed in an embodiment of the present invention.

Detailed ways

Embodiments of the present invention will be described below with reference to the drawings.

Specifically, referring to FIG. 1, the first method of prosodic labeling provided by the embodiment of the present invention may include:

S101: Receive an audio data identifier to be marked, where the audio data identifier includes a batch and/or data entry number of the audio data.

In the embodiment of the present invention, the audio data identification to be marked can be represented by the batch or data entry number of the audio data, so that the audio data to be marked can be determined by selecting the batch and/or data entry number to be marked. data.

Wherein, the audio data to be marked can be downloaded from the Internet by using the wireless Internet access module, or can be downloaded from the local area by using the USB module. There are many ways to obtain the audio data to be marked, which is not limited in the embodiment of the present invention.

S102: Play corresponding audio data to the tagger according to the audio data identifier.

After the audio data identification is selected, the audio data to be marked can be determined, and at this time, the selected audio data is played to the marker. It should be noted that the annotator here can be realized by a natural person, or by an entity with an annotation function.

Specifically, as shown in FIG. 2, the step S102 may include in practical applications:

S201: Play the corresponding audio data to the annotator for the first time according to the audio data identifier.

In the process of playing the audio data, the way of playing repeatedly can be adopted, which can improve the accuracy of labeling. The corresponding audio data is therefore first played to the annotator for the first time.

S202: After pausing for a first predetermined period of time, play the audio data to the annotator for the second time.

The first predetermined time period can be set to two seconds, so that the labeler can be given a period of time to get used to it, and the attention of labeling can be improved.

S203: After pausing for a second predetermined period of time, play the audio data to the annotator for the third time.

The second predetermined time period may be set to three seconds, which has the same effect as the first predetermined time period.

Returning to FIG. 1 , in step S103 : if it is detected that the annotator triggers the annotating process, record information related to the annotator's annotating behavior.

In the case of playing the audio data to be labeled twice to the labeler, it is detected whether the labeler has triggered the labeling process, and if the labeler has triggered the labeling process, the information related to the labeling behavior of the labeler is recorded, and the labeler The information related to the labeling behavior can specifically be the trigger time point and pause duration information, that is, the time point when the labeler triggers the label button and the duration of each trigger. If the annotator does not trigger the annotation process, no other related processing is performed, and the audio data continues to be played until the annotator triggers the annotation process before recording annotation information, such as initial trigger time point and initial pause duration information etc.

Specifically, as shown in FIG. 3, the step S103 may include in practical applications:

S301: In the process of playing the audio data for the third time, detect whether the tagger triggers the tagging process, if yes, go to step S302, if not, go to step S304.

It should be noted that detecting whether the tagger triggers the tagging process can be achieved by detecting whether the tagger triggers the tagging button, and the specific introduction can refer to the next embodiment.

S302: Record information about an initial trigger time point and an initial pause duration.

Record the initial trigger time point and initial pause duration information. It should be noted that the initial trigger time point and initial pause duration information recorded in S302 can be considered as an analog process, because the trigger recorded during the fourth playback of audio data The time point and pause duration information are used as the final audio annotation data.

S303: When the audio data is played for the third time, pause for a third predetermined period of time, and record the information of the final trigger time point and the final pause duration during the process of playing the audio data to the annotator for the fourth time.

Specifically, the third predetermined time period may be set as one second. It should be noted that the final trigger time point and final pause duration information recorded in S303 can also be compared with the initial trigger time point and initial pause duration information in S302, or the average value thereof can be taken as the final audio annotation data. It is feasible, and can be adaptively adjusted according to the actual situation or user needs.

S304: Continue playing the audio data without performing any other related processing.

Returning to FIG. 1 , in step S104 : generating audio annotation information of the audio data according to the information related to the annotation behavior of the annotator.

In this embodiment, the final trigger time point and the final pause duration information recorded during the fourth playing of the audio data may be used as audio annotation information.

It should be noted that the predetermined time periods mentioned in the embodiments of the present application can be adjusted adaptively according to different actual needs, and are not limited in the embodiments of the present application.

Through the above-mentioned first method of prosodic labeling, by playing the audio data to the labeler multiple times, the labeling process of the audio data can be triggered after the labeler is familiar with the audio data, and by repeated labeling, it can be used. The audio annotation data obtained in this embodiment is more accurate, and thus the accuracy or fluency of speech synthesis using the audio annotation data can meet actual requirements.

Specifically, referring to FIG. 4, the embodiment of the present invention provides another method for prosodic labeling, which may include:

S401: Receive an audio data identifier to be annotated through an interface of a dedicated annotating device, where the audio data identifier includes a batch and/or data entry number of the audio data.

In this embodiment, specifically, the audio data identifier to be marked may be received through an interface of a special device, and the audio data identifier may also include a batch and/or data entry number of the audio data.

S402: Play corresponding audio data to the tagger according to the audio data identifier.

This step is similar to the first method embodiment of prosodic labeling, and will not be repeated here.

S403: Detect whether the tagging process is triggered by detecting whether the tagger triggers the tagging button on the interface of the dedicated tagging device.

In this embodiment, whether the tagging process is triggered is detected specifically by detecting whether the tagger triggers the tagging button on the interface of the special tagging device. If the trigger information of the labeler on the label button can be received, it is considered that the labeler has triggered the labeling process, so whether to start labeling can be detected by detecting the trigger information of the label button.

S404: Record information related to the tagging behavior of the tagger when it is detected that the tagging process is triggered by the tagger.

S405: Generate audio annotation information of the audio data according to the information related to the annotation behavior of the annotator.

S406: Save the recorded audio annotation information in a data format that can be used for network transmission.

After generating the audio annotation information of the audio data, the audio annotation information is also saved as a data format that can be used for network transmission, such as a file in Extensible Markup Language (XML, Extensible Markup Language) format, and the audio annotation can be The information is stored in the memory, and each piece of sound data can only keep the latest annotation record.

Through this embodiment, if the audio data to be marked is updated, the marking process can be repeated through network download or local download, and the final saved audio marking information can be exported through the USB interface or uploaded to the server through the network. This embodiment can conveniently generate accurate audio annotation information, and then make the accuracy or fluency of speech synthesis meet actual needs, and can also conveniently monitor whether annotation is required by detecting the trigger information of the annotation button , and can conveniently realize the sharing and publishing of audio annotation information on the network.

Specifically, referring to FIG. 5, the embodiment of the present invention provides a third method for prosodic labeling, which may include:

S501: Receive an audio data identifier to be marked, where the audio data identifier includes a batch and/or data entry number of the audio data.

S502: Play corresponding audio data to the tagger according to the audio data identifier.

S503: Record information related to the tagging behavior of the tagger when it is detected that the tagging process is triggered by the tagger.

S504: Generate audio annotation information of the audio data according to the information related to the annotation behavior of the annotator.

S505: Extract at least two audio annotation information samples from the audio annotation information set of any annotator.

In this embodiment, after the audio annotation information is generated, for any annotator, credible weight processing may be performed on all the audio annotation information to detect the credibility of the audio annotation information. Firstly, at least two audio annotation information samples need to be extracted from any annotator's audio annotation information set.

S506: Obtain a standard reference duration of any annotator according to the at least two audio annotation information samples.

In this step, it is necessary to calculate the standard reference duration of any annotator based on at least two extracted audio annotation information samples.

Referring to FIG. 6, specifically, the step S506 may include in practical applications:

S601: Obtain a minimum value of sub-label duration in each audio label information sample.

In practical applications, assuming that N pieces of audio annotation information samples are extracted, and each annotation information has M sub-annotation durations, this step obtains the minimum value Min(T ₁ , T ₂ ... T _N ), to obtain N minimum hour length information.

S602: Calculate the standard deviation value of the sub-annotation duration in each piece of audio annotation information according to the obtained minimum value.

Calculate the standard deviation E of the sub-tagging duration according to the N smallest sub-tagging duration information. In this step, an existing calculation formula can be used to calculate the standard deviation value, and the amounts will not be listed here.

S603: Use the smallest sub-annotation duration in the audio annotation information corresponding to the smallest standard deviation value as the standard reference duration of any annotator.

Because there are N calculated standard deviation values, the minimum sub-tagging duration in the audio annotation information corresponding to the smallest standard deviation value Min(E) among the N standard deviation values is used as the standard for any tagger Baseline duration Pi.

S507: Perform credible weight processing on the audio annotation information set of any annotator by using the standard reference duration.

After the annotation reference duration of the annotator is obtained in step S603, credible weight processing is performed on the audio annotation information set of the annotator by using the standard reference duration. Referring to FIG. 7, the step S507 may specifically include:

S701: According to the N minimum sub-tagging durations of the first N pieces of audio tagging information of the tagger, calculate the standard deviation between the N smallest sub-tagging durations and the reference tagging duration; wherein, N is a natural number greater than 1.

In this step, each user has a standard reference duration P _i , and then calculates the standard deviation of the N minimum sub-label durations T _i according to the N minimum sub-label durations T i of each user’s N pieces of audio data. The calculation formula is as follows :

$\mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; =</span>\sqrt[\mathrm{<span\; class="notranslate">\; 2</span>}]{\underset{\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; no</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}$

S702: Use the standard deviation as a credibility measurement parameter to perform credible weight processing on the audio annotation information of any annotator, wherein the larger the value of the standard deviation, the higher the credibility of the audio annotation information. Low.

The calculated standard deviation F is used as a credibility measurement parameter to perform credible weight processing on the audio annotation information of any annotator, wherein the larger the value of the standard deviation, the greater the difference between the audio annotation information, That is, if the annotator may experience fatigue or other objective conditions that cause the audio annotation information to be relatively messy, it means that the reliability of the audio annotation information is lower.

In short, according to the third prosodic labeling method disclosed in the embodiment of the present invention, in addition to accurately labeling audio data, it can also perform credible weight processing on all audio labeling information marked by a certain labeler, so that The accuracy and accuracy of the audio annotation information can be further evaluated, thus laying the foundation for subsequent applications such as speech synthesis.

Corresponding to the prosodic tagging method provided by the embodiment of the present invention, the embodiment of the present invention also provides a prosodic tagging device, as shown in FIG. 8 , the device may include a receiving module 801 , a playing module 802 and a recording module 803 . The operation of each module will be described in detail below.

The receiving module 801 is configured to receive an audio data identifier to be marked, where the audio data identifier includes a batch and/or data entry number of the audio data.

Wherein, in a practical application, the receiving module 801 may specifically be configured to: receive an audio data identifier to be marked through an interface of a special marking device. Referring to FIG. 9 , it is a schematic diagram of an interface of the dedicated labeling device.

In Fig. 9, the "switch" button is used to control the opening and closing of the special marking device, the "volume" button is used to control the volume of the audio data played, and the circular number buttons in Fig. 9 can be used to conveniently select audio Data batch number and data entry number; "Download" and "Upload" buttons can be used to receive and submit audio data; "Batch" and "Serial Number" buttons can be used to select different audio data batches and Corresponding data entry; the playback control buttons "Previous", "Next", "Repeat" and "Auto" are used to control the playback of the previous audio data, the next audio data, repeat while playing audio data or automatically play audio data sequentially . The "mark button" can be marked according to the record of keystrokes triggered by the annotator, and give different information according to the length of time the annotator presses, that is, the trigger time point and trigger duration information. Among them, the buttons of the special labeling equipment can adopt capacitive buttons, and can also adopt the mode of automatic pronunciation when the labeler touches, so that the buttons can also be easily recognized by the blind.

It should be noted that the aspect ratio of the interface of the special labeling device can be set to 2:1, wherein the position of the label button on the interface of the special labeling device adopts the golden ratio setting, and at least one playback control The buttons are arranged in a square at 2-3 centimeters below the labeling button, and the playback control button is used to control the playback sequence of the audio data; other buttons except the playback control button are arranged on the special labeling device. around the interface.

The playing module 802 is configured to play corresponding audio data to the tagger according to the audio data identification.

The playing module 802 can be specifically configured to: play the corresponding audio data to the tagger for the first time according to the audio data identifier; play the audio data to the tagger for the second time after pausing for a first predetermined period of time; and playing the audio data to the annotator for a third time after pausing for a second predetermined period of time.

The recording module 803 is configured to record information related to the tagging behavior of the tagger when it is detected that the tagger triggers the tagging process.

Referring to Figure 10, the recording module 803 may specifically include in practical applications:

The detection sub-module 1001 is configured to detect whether the tagger triggers the tagging process when the audio data is played for the third time.

Wherein, the detection sub-module 1001 may be specifically configured to: detect whether the labeling process is triggered by detecting whether the labeler triggers the labeling button on the interface of the special labeling device.

The first recording submodule 1002 is configured to record the information of the initial trigger time point and the initial pause duration when the result of the detection submodule is yes.

The second recording sub-module 1003 is used to stop the third predetermined period of time when the audio data is played for the third time, and record the final initial trigger time point and Pause duration information.

A generating module 804, configured to generate audio annotation information of the audio data according to the information related to the annotation behavior of the annotator.

Wherein, the generating module 804 may be specifically configured to: use the final trigger time point and final pause duration information recorded during the fourth playing of the audio data as audio annotation information.

Through the above-mentioned device provided by the embodiment of the present invention, by playing the audio data to the annotator multiple times, the audio data annotation process can be triggered after the annotator is familiar with the audio data, and by repeating the annotation, it is possible to use The audio annotation data obtained in this embodiment is more accurate, and thus the accuracy or fluency of speech synthesis using the audio annotation data can meet actual requirements.

In addition, it should also be noted that the series of processes and devices described above may also be implemented by software and/or firmware. In the case of realization by software and/or firmware, a program constituting the software is installed from a storage medium or a network to a computer having a dedicated hardware configuration, such as a general-purpose personal computer 1100 shown in FIG. , can perform various functions and so on.

In FIG. 11 , a central processing unit (CPU) 1101 executes various processes according to programs stored in a read only memory (ROM) 1102 or loaded from a storage section 1108 to a random access memory (RAM) 1103 . In the RAM 1103, data required when the CPU 1101 executes various processing and the like is also stored as necessary.

The CPU 1101, ROM 1102, and RAM 1103 are connected to each other via a bus 1104. An input/output interface 1105 is also connected to the bus 1104 .

The following components are connected to the input/output interface 1105: an input section 1106 including a keyboard, a mouse, etc.; an output section 1107 including a display such as a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker; The storage section 1108 includes a hard disk and the like; and the communication section 1109 includes a network interface card such as a LAN card, a modem, and the like. The communication section 1109 performs communication processing via a network such as the Internet.

A driver 1110 is also connected to the input/output interface 1105 as needed. A removable medium 1111 such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, or the like is mounted on the drive 1110 as necessary, so that a computer program read therefrom is installed into the storage section 1108 as necessary.

In the case of realizing the above-described series of processes by software, the programs constituting the software are installed from a network such as the Internet or a storage medium such as the removable medium 1111 .

Those skilled in the art should understand that such a storage medium is not limited to the removable medium 1111 shown in FIG. 11 in which the program is stored and distributed separately from the device to provide the program to the user. Examples of the removable media 1111 include magnetic disks (including floppy disks (registered trademark)), optical disks (including compact disk read only memory (CD-ROM) and digital versatile disks (DVD)), magneto-optical disks (including )) and semiconductor memory. Alternatively, the storage medium may be a ROM 1102, a hard disk contained in the storage section 1108, or the like, in which programs are stored and distributed to users together with devices containing them.

It should also be pointed out that the steps for executing the above series of processes can naturally be executed in chronological order according to the illustrated order, but it does not need to be executed in chronological order. Certain steps may be performed in parallel or independently of each other.

Regarding the implementation manner comprising the above embodiments, the following additional notes are also disclosed:

Note 1. A method for prosodic labeling, comprising:

Receive the audio data identification to be marked, the audio data identification includes the batch and/or data entry number of the audio data;

Play the corresponding audio data to the annotator according to the audio data identification;

When it is detected that the tagger has triggered the tagging process, recording information related to the tagging behavior of the tagger;

The audio annotation information of the audio data is generated according to the information related to the annotation behavior of the annotator.

Additional Note 2. According to the method described in Additional Note 1, the information related to the tagging behavior of the tagger is specifically trigger time point and pause duration information.

Supplement 3. According to the method described in Supplement 2, playing the corresponding audio data to the labeler according to the audio data identification includes:

Playing the corresponding audio data to the annotator for the first time according to the audio data identification;

Playing the audio data a second time to the annotator after pausing for a first predetermined period of time; and

After a pause for a second predetermined period of time, the audio data is played a third time to the annotator.

Additional Note 4. According to the method described in Additional Note 3, the step of recording information related to the tagging behavior of the tagger when it is detected that the tagger has triggered the tagging process specifically includes:

In the process of playing the audio data for the third time, it is detected whether the tagger has triggered the tagging process; if so, the initial trigger time point and the initial pause duration information are recorded;

When the audio data is played for the third time, stop for a third predetermined period of time, and record the final initial trigger time point and pause duration information during the fourth time of playing the audio data to the annotator.

Additional Note 5. According to the method described in Additional Note 4, the first predetermined time period is two seconds, the second predetermined time period is three seconds, and the third predetermined time period is one second.

Additional Note 6. According to the method described in Additional Note 4, the step of generating the audio tagging information of the audio data according to the information related to the tagging behavior of the tagger is specifically:

The final trigger time point and the final pause duration information recorded during the fourth playing of the audio data are used as audio annotation information.

Supplement 7. The method according to Supplement 1, further comprising:

The recorded audio annotation information is saved in a data format that can be used for network transmission.

Supplement 8. According to the method described in Supplement 1, the audio data identification to be marked is received through the interface of the special marking device.

Supplement 9. According to the method described in Supplement 1, whether the labeling process is triggered is detected by detecting whether the labeler triggers the labeling button on the interface of the dedicated labeling device.

Supplement 10. The method according to Supplement 1, further comprising:

extracting at least two samples of audio annotation information from any annotator's audio annotation information set;

Acquiring the standard reference duration of any annotator according to the at least two audio annotation information samples;

The credible weighting process is performed on the audio annotation information set of any annotator by using the standard reference duration.

Supplement 11. According to the method described in Supplement 10, the step of obtaining the standard reference duration of any one of the annotators based on the at least two audio annotation information samples includes:

Obtain the minimum value of the sub-label duration in each audio label information sample;

Calculate the standard deviation of the sub-annotation duration in each piece of audio annotation information based on the obtained minimum value;

The smallest sub-annotation duration in the audio annotation information corresponding to the smallest standard deviation value is used as the standard reference duration of any annotator.

Supplement 12. According to the method described in Supplement 11, the step of processing trusted weights includes:

According to the N minimum sub-labeling durations of the first N pieces of audio labeling information of the labeler, calculate the standard deviation between the N minimum sub-labeling durations and the reference labeling duration; wherein, N is a natural number greater than 1;

The standard deviation is used as a credibility measurement parameter to carry out credible weight processing on the audio annotation information of any annotator, wherein the larger the value of the standard deviation is, the lower the credibility of the audio annotation information is.

Note 13. A special labeling device, including:

The receiving module is used to receive the audio data identification to be marked, and the audio data identification includes the batch and/or data entry number of the audio data;

A playback module, configured to play the corresponding audio data to the annotator according to the audio data identifier;

A recording module, configured to record information related to the labeling behavior of the labeler when it is detected that the labeler has triggered the labeling process;

A generating module, configured to generate audio annotation information of the audio data according to the information related to the annotation behavior of the annotator.

Supplement 14. According to the device described in Supplement 13, the information related to the marking behavior of the annotator is specifically trigger time point and pause duration information.

Supplement 15. According to the device described in Supplement 14, the playback module is specifically configured as:

Playing the corresponding audio data to the annotator for the first time according to the audio data identification;

Playing the audio data a second time to the annotator after pausing for a first predetermined period of time; and

After a pause for a second predetermined period of time, the audio data is played a third time to the annotator.

Supplement 16. The device according to Supplement 14, the recording module includes:

The detection submodule is used to detect whether the tagger has triggered the tagging process when the audio data is played for the third time;

The first recording submodule is used to record the initial trigger time point and initial pause duration information when the result of the detection submodule is yes;

The second recording submodule is used to pause for a third predetermined period of time when the audio data is played for the third time, and to record the final initial trigger time point and pause during the fourth playback of the audio data to the annotator Duration information.

Supplement 17. According to the device described in Supplement 14, the generating module is specifically configured as:

The final trigger time point and the final pause duration information recorded during the fourth playing of the audio data are used as audio annotation information.

Supplement 18. According to the device described in Supplement 13, the specific configuration of the receiving module is:

The audio data identification to be marked is received through the interface of the special marking device.

Supplement 19. According to the device described in Supplement 16, the detection submodule is specifically configured as:

Whether the tagging process is triggered is detected by detecting whether the tagger triggers the tagging button on the interface of the special tagging device.

Additional Note 20. According to the device described in Additional Note 13, the aspect ratio of the interface of the special labeling device is 2:1, and the position of the labeling button on the interface of the special labeling device is set by the golden ratio, At least one playback control button is arranged in a square at 2-3 centimeters below the label button, and the playback control button is used to control the playback sequence of the audio data; other buttons except the playback control button are arranged on the Around the interface of the dedicated labeling device.

Although the present invention and its advantages have been described in detail, it should be understood that various changes, substitutions and alterations can be made hereto without departing from the spirit and scope of the invention as defined by the appended claims. Moreover, the terms "comprising", "comprising" or any other variation thereof in the embodiments of the present invention are intended to cover a non-exclusive inclusion such that a process, method, article or device comprising a series of elements includes not only those elements, but also Including other elements not expressly listed, or also including elements inherent in such process, method, article or apparatus. Without further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of additional identical elements in the process, method, article or apparatus comprising said element.

Claims (8)

1. A prosodic labeling device, comprising: The receiving module is used to receive the audio data identification to be marked, and the audio data identification includes the batch and/or data entry number of the audio data; A playback module, configured to play the corresponding audio data to the annotator according to the audio data identifier; A recording module, configured to record information related to the labeling behavior of the labeler when it is detected that the labeler has triggered the labeling process; A generation module, configured to generate audio annotation information of the audio data according to the information related to the annotation behavior of the annotator, The information related to the tagging behavior of the tagger is specifically trigger time point and pause duration information. 2. The prosody labeling device according to claim 1, the specific configuration of the playback module is: Playing the corresponding audio data to the annotator for the first time according to the audio data identification; Playing the audio data a second time to the annotator after pausing for a first predetermined period of time; and After a pause for a second predetermined period of time, the audio data is played a third time to the annotator. 3. The prosody labeling device according to claim 1, the recording module comprising: The detection submodule is used to detect whether the tagger has triggered the tagging process when the audio data is played for the third time; The first recording submodule is used to record the initial trigger time point and initial pause duration information when the result of the detection submodule is yes; The second recording submodule is used to pause for a third predetermined period of time when the audio data is played for the third time, and to record the final initial trigger time point and pause during the fourth playback of the audio data to the annotator Duration information. 4. The prosody labeling device according to claim 3, the specific configuration of the generating module is: The final trigger time point and the final pause duration information recorded during the fourth playing of the audio data are used as audio annotation information. 5. The prosody labeling device according to claim 1, the receiving module is specifically configured as: The audio data identification to be marked is received through the interface of the prosody marking device. 6. The prosody labeling device according to claim 3, the specific configuration of the detection submodule is: Whether the tagging process is triggered is detected by detecting whether the tagger triggers the tagging button on the interface of the prosodic tagging device. 7. The prosody labeling device according to claim 1, the aspect ratio of the interface of the prosody labeling device is 2: 1, and the position of the label button on the interface of the prosody labeling device adopts the golden ratio setting, At least one playback control button is arranged in a square at 2-3 centimeters below the label button, and the playback control button is used to control the playback sequence of the audio data; other buttons except the playback control button are arranged on the Rhythm labels the perimeter of the device's interface. 8. A method for prosodic labeling, comprising: Receive the audio data identification to be marked, the audio data identification includes the batch and/or data entry number of the audio data; Play the corresponding audio data to the annotator according to the audio data identification; When it is detected that the tagger has triggered the tagging process, recording information related to the tagging behavior of the tagger; generating audio annotation information of the audio data according to the information related to the annotation behavior of the annotator, The information related to the tagging behavior of the tagger is specifically trigger time point and pause duration information.