A to-do task generation method and device

Abstract

The application discloses a to-do task generation method and device, and relates to the technical field of speech recognition. The method comprises the following steps: segmenting a recording audio signal and extracting features of the recording audio signal to obtain recording audio segment features; grouping the recording audio segment features based on similarities in a similarity matrix, and taking the recording audio segment features in each recording audio segment feature group as a cluster; calculating the similarity between the clusters based on the similarity matrix; clustering based on the similarity between the clusters to obtain a clustering tree; assigning a speaker label based on the clustering tree to obtain a target recording audio signal segment, and then performing speech-to-text processing on the target recording audio signal segment to obtain text content corresponding to the target recording audio signal segment; dynamically constructing a prompt word based on a keyword table, a historical task template and the text content; taking the text generation type as a task type corresponding to a preloaded model; and generating a to-do task by using a task generation model according to the prompt word. In this way, the efficiency of to-do task generation is improved.

Description

Technical Field

[0001] This application relates to the field of speech recognition technology, and in particular to a method and apparatus for generating to-do tasks. Background Technology

[0002] With the development of information technology, office automation systems are increasingly being used in enterprises, improving office efficiency.

[0003] In related technologies, traditional office automation systems include multiple functional modules, each used to perform different functions. For example, a meeting management module manages and publishes meetings and meeting minutes, while a task management module manages the tasks in the system. However, in traditional office automation systems, it is mainly necessary to manually extract tasks from the meeting minutes in the meeting management module and then publish them based on the task management module. However, relying solely on manual operation leads to delays in updating meeting minutes and publishing tasks, resulting in low office efficiency. Summary of the Invention

[0004] In view of this, this application provides a method and apparatus for generating to-do tasks, which is used to solve the problem of low office efficiency in the prior art.

[0005] The objective of this application can be achieved through the following technical solutions:

[0006] The first aspect of this application is to provide a method for generating to-do tasks, including:

[0007] Acquire the audio recording file, which includes the audio signal.

[0008] The recorded audio signal is segmented to obtain multiple recorded audio signal segments;

[0009] Feature extraction is performed on the recorded audio signal segments to obtain the features of the recorded audio segments;

[0010] Calculate the similarity between features of the recorded audio segments to obtain a similarity matrix;

[0011] Based on the similarity in the similarity matrix, the audio segment features are grouped to obtain multiple audio segment feature groups, and the audio segment features in each audio segment feature group are treated as a cluster.

[0012] Calculate the similarity between clusters based on the similarity matrix;

[0013] Clustering is performed based on the similarity between clusters to obtain a clustering tree;

[0014] Speaker labels are assigned based on clustering trees to obtain target audio signal segments, which include the corresponding speaker labels.

[0015] The target audio signal segment is processed for speech-to-text conversion to obtain the corresponding text content of the target audio signal segment.

[0016] Obtain the keyword list and historical task templates;

[0017] Dynamically generate prompts based on the keyword list, historical task templates, and text content;

[0018] Use the text generation type as the task type corresponding to the pre-loaded task generation model;

[0019] Based on the prompts, generate tasks to be done using a task generation model.

[0020] In an optional embodiment, before performing feature extraction on the recorded audio signal segment to obtain the recorded audio segment features, the method further includes:

[0021] A short-time Fourier transform is performed on the recorded audio signal segment to obtain the recorded audio spectrum, which includes the recorded audio amplitude spectrum and the recorded audio phase spectrum.

[0022] The noise amplitude spectrum is obtained by performing noise estimation on the amplitude spectrum of the recorded audio.

[0023] The enhanced amplitude spectrum of the recorded audio is obtained by performing a spectral subtraction operation on the noise amplitude spectrum.

[0024] The enhanced spectrum of the recorded audio is obtained based on the amplitude spectrum and phase spectrum of the recorded audio.

[0025] The inverse short-time Fourier transform of the audio enhancement spectrum is used to obtain the audio enhancement signal.

[0026] Feature extraction is performed on the recorded audio signal segments to obtain the features of the recorded audio segments, including:

[0027] Feature extraction is performed on the enhanced audio signal to obtain the features of the audio segments.

[0028] In one optional embodiment, a spectral subtraction operation is performed on the recorded audio amplitude spectrum based on the noise amplitude spectrum to obtain the enhanced amplitude spectrum of the recorded audio, including:

[0029] Based on the noise amplitude spectrum, the enhanced amplitude spectrum of the recorded audio is obtained by performing a spectral subtraction operation on the recorded audio amplitude spectrum using the following formula:

[0030]

[0031] in, This refers to the amplitude spectrum of the recorded audio signal at frequency f. This is the amplitude spectrum of the recorded audio signal at frequency f. This represents the noise amplitude spectrum at frequency f. For over-subtraction factor, This is the protection factor.

[0032] In an optional embodiment, before performing a spectral subtraction operation on the recorded audio amplitude spectrum based on the noise amplitude spectrum to obtain the enhanced amplitude spectrum of the recorded audio, the method further includes:

[0033] Calculate the signal-to-noise ratio of the recorded audio signal at frequency f;

[0034] Obtain the noise intensity of the recorded audio signal at frequency f;

[0035] The over-attenuation factor is adjusted based on the signal-to-noise ratio, noise intensity, and the initial value of the over-attenuation factor;

[0036] The protection factor is adjusted based on the initial values ​​of the signal-to-noise ratio, noise intensity, and protection factor.

[0037] In one optional embodiment, obtaining the noise intensity of the recorded audio signal at a frequency of f includes:

[0038] Obtain the midline of the recorded audio signal at frequency f. The midline is determined based on the maximum and minimum values ​​of the recorded audio signal at frequency f.

[0039] The noise intensity of the recorded audio signal at frequency f is calculated based on the signal value located on the center line, the average value of all signal values ​​located on the center line, and the number of signal values ​​located on the center line.

[0040] In an optional embodiment, before assigning speaker labels based on a clustering tree to obtain the target recorded audio signal segment, the method further includes:

[0041] Obtain voiceprint samples from attendees;

[0042] A voiceprint sample library is constructed based on voiceprint samples.

[0043] In one optional embodiment, speaker labels are assigned based on a clustering tree to obtain the target recorded audio signal segment, including:

[0044] Based on the voiceprint sample library, speaker labels are assigned to each cluster in the clustering tree.

[0045] In an optional embodiment, it further includes:

[0046] Edit the text content based on the received editing operation.

[0047] In an alternative embodiment, the method further includes: format conversion of the task to be performed.

[0048] A second aspect of this application is to provide a task generation device, characterized in that it includes:

[0049] The first acquisition module is used to acquire the recording file, which includes the recorded audio signal;

[0050] The segmentation module is used to segment the recorded audio signal into multiple recorded audio signal segments;

[0051] The extraction module is used to extract features from the recorded audio signal segments to obtain the features of the recorded audio segments;

[0052] The first calculation module is used to calculate the similarity between features of recorded audio segments and obtain a similarity matrix;

[0053] The grouping module is used to group the recorded audio segment features based on the similarity in the similarity matrix, resulting in multiple recorded audio segment feature groups, and the recorded audio segment features in each recorded audio segment feature group are treated as a cluster;

[0054] The second calculation module is used to calculate the similarity between clusters based on the similarity matrix;

[0055] The clustering module is used to perform clustering based on the similarity between clusters to obtain a clustering tree;

[0056] The allocation module is used to allocate speaker tags based on a clustering tree to obtain the target audio recording signal segment, which includes the corresponding speaker tag.

[0057] The speech-to-text module is used to process the target audio signal segment into speech and obtain the text content corresponding to the target audio signal segment.

[0058] The second acquisition module is used to acquire the keyword list and historical task templates;

[0059] The module is used to dynamically generate prompt words based on the keyword list, historical task templates, and text content;

[0060] The determination module is used to identify the text generation type as the task type corresponding to the pre-loaded task generation model.

[0061] The generation module is used to generate tasks to be done based on prompts and a task generation model.

[0062] A third aspect of this application is to provide an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to perform the method as described in the first aspect.

[0063] A fourth aspect of this application is to provide a computer-readable storage medium storing a computer program, which, when executed by a processor, performs the method as described in the first aspect.

[0064] Compared with existing technologies, the to-do task generation method provided in this application segments the recorded audio signal and extracts features to obtain audio segment features; it groups the audio segment features based on the similarity in the calculated similarity matrix to obtain multiple audio segment feature groups, and treats the audio segment features in each audio segment feature group as a cluster; it calculates the similarity between clusters based on the similarity matrix; it performs clustering based on the similarity between clusters to obtain a clustering tree; it assigns speaker labels based on the clustering tree to obtain the target audio signal segment, and then performs speech-to-text processing to obtain the text content corresponding to the target audio signal segment; it dynamically constructs prompt words based on a keyword list, historical task templates, and text content; it uses the text generation type as the task type corresponding to the pre-loaded model; and it generates to-do tasks using the task generation model based on the prompt words. This improves the efficiency of to-do task generation. Attached Figure Description

[0065] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0066] Figure 1 A flowchart illustrating a method for generating to-do tasks provided in an embodiment of this application;

[0067] Figure 2 Another flowchart illustrating the to-do task generation method provided in this application embodiment;

[0068] Figure 3 A structural block diagram of a task generation device provided in an embodiment of this application;

[0069] Figure 4 This is a structural block diagram of an electronic device for implementing a to-do task generation method, provided in an embodiment of this application. Detailed Implementation

[0070] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0071] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0072] It should be understood that in the embodiments of this application, "at least one" means one or more, and "more than one" means two or more. "And / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. The character " / " generally indicates that the related objects before and after it are in an "or" relationship. "Contains A, B and / or C" means containing any one, two, or three of A, B, and C.

[0073] It should be understood that in the embodiments of this application, "B corresponding to A", "B corresponding to A", "A corresponds to B" or "B corresponds to A" means that B is associated with A, and B can be determined based on A. Determining B based on A does not mean that B is determined solely based on A; B can also be determined based on A and / or other information.

[0074] To address the technical problems existing in related technologies, this application provides a method and apparatus for generating to-do tasks.

[0075] The task generation method provided in this application can be executed by an electronic device, which can be a terminal or a server. The terminal can be a smartphone, tablet, laptop, or other similar device. The server can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms. It is understood that this application does not limit the specific entity executing the task generation method.

[0076] The technical solution of this application will be described in detail below through specific embodiments. It should be noted that the following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments described below are used to explain the technical solution of this application and are not intended to limit actual use.

[0077] To address the technical problems existing in related technologies, embodiments of this application provide a method for generating to-do tasks, such as... Figure 1 As shown, Figure 1 This is a flowchart illustrating a method for generating to-do tasks according to an embodiment of this application. It should be noted that the steps shown may be executed in a logical order different from that shown in the flowchart. The method may include the following steps S101 to S113.

[0078] Step S101: Obtain the audio file.

[0079] In one alternative embodiment, the audio file may be obtained by recording the voices of the participants using a recording device during the meeting, and the audio file includes the recorded audio signal.

[0080] It should be noted that all attendees were informed in advance and their consent was obtained before the recording began; this meeting will be recorded.

[0081] In one specific embodiment, the meeting organizer logs into the office automation system, captures sound through a microphone, and pushes the recorded voice to the voiceprint recognition system in real time.

[0082] In one specific embodiment, a communication connection is established between the office automation system and the voiceprint recognition system based on a full-duplex communication protocol.

[0083] In this step, the voices of the participants are recorded during the meeting to generate audio files, which can then be used to generate to-do tasks.

[0084] In another optional embodiment, the to-do task generation method provided in this application further includes the following steps: obtaining voiceprint samples of attendees; and constructing a voiceprint sample library based on the voiceprint samples.

[0085] It should be noted that if no speaker is matched in the voiceprint sample database, the current speaker will be modified in real time, and the speaker's voiceprint will be saved to the voiceprint sample database.

[0086] In one specific embodiment, voiceprint samples are stored in a voiceprint sample library in the form of sets, with different sets corresponding to different meetings.

[0087] In this embodiment, a voiceprint sample database is constructed so that attendees can be identified based on the voiceprint sample database in the future.

[0088] Step S102: Segment the recorded audio signal to obtain multiple recorded audio signal segments.

[0089] In one optional embodiment, the recorded audio signal is segmented according to a preset time interval. For example, the preset time intervals are 200ms, 250ms, and 300ms.

[0090] Step S103: Extract features from the recorded audio signal segment to obtain the features of the recorded audio segment.

[0091] In one optional embodiment, VAD (Voice Activity Detector) is used to extract features from the recorded audio signal segments to obtain the features of the recorded audio segments.

[0092] In one optional embodiment, before extracting features from the recorded audio signal segment to obtain the features of the recorded audio segment, noise processing can be performed on the recorded audio signal segment first.

[0093] In one specific embodiment, the to-do task generation method provided in this application further includes the following steps:

[0094] A short-time Fourier transform is performed on the recorded audio signal segment to obtain the recorded audio spectrum, which includes the recorded audio amplitude spectrum and the recorded audio phase spectrum. Noise estimation is performed on the recorded audio amplitude spectrum to obtain the noise amplitude spectrum. Based on the noise amplitude spectrum, a spectral subtraction operation is performed on the recorded audio amplitude spectrum to obtain the recorded audio enhanced amplitude spectrum. Based on the recorded audio enhanced amplitude spectrum and the recorded audio phase spectrum, the recorded audio enhanced spectrum is obtained. An inverse short-time Fourier transform is performed on the recorded audio enhanced spectrum to obtain the recorded audio enhanced signal.

[0095] Correspondingly, feature extraction is performed on the recorded audio signal segment to obtain the recorded audio segment features. Specifically, this includes the following steps: feature extraction is performed on the recorded audio enhancement signal to obtain the recorded audio segment features.

[0096] In another specific embodiment, after obtaining the recorded audio spectrum, non-speech segments in the recorded audio signal segment are identified by VAD so that noise estimation can be performed subsequently based on the non-speech segments.

[0097] In another specific embodiment, the noise amplitude spectrum is obtained by using MCRA (Minimum Controlled Recursive Averaging) to estimate the noise amplitude spectrum of the recorded audio.

[0098] In a more specific embodiment, a spectral subtraction operation is performed on the recorded audio amplitude spectrum based on the noise amplitude spectrum to obtain the enhanced amplitude spectrum of the recorded audio, including:

[0099] Based on the noise amplitude spectrum, the enhanced amplitude spectrum of the recorded audio is obtained by performing a spectral subtraction operation on the recorded audio amplitude spectrum using the following formula:

[0100] (1);

[0101] in, This refers to the amplitude spectrum of the recorded audio signal at frequency f. This is the amplitude spectrum of the recorded audio signal at frequency f. This represents the noise amplitude spectrum at frequency f. For over-subtraction factor, This is the protection factor.

[0102] In a more specific embodiment, and It is set up based on actual needs; for example, when the meeting is held in a conference room, it will... Set to 2.0. Set to 0.01. When the meeting is held at a construction site, Set to 3.0, Set to 0.001. When the meeting is held in a KTV, [the value will be set to 0.001]. Set to 1.5. Set to 0.05.

[0103] The physical meaning is to eliminate the noise-dominant frequency band components. The physical meaning of this is to ensure that the result is not negative, avoid imaginary spectrum, and retain a small amount of noise. The physical meaning is to convert the energy domain into the amplitude domain while maintaining the waveform continuity when combined with the original phase spectrum.

[0104] In a more specific embodiment, the to-do task generation method provided in this application further includes the following steps:

[0105] Calculate the signal-to-noise ratio (SNR) of the recorded audio signal at frequency f; obtain the noise intensity of the recorded audio signal at frequency f; adjust the over-attenuation factor based on the initial values ​​of the SNR, noise intensity, and over-attenuation factor; adjust the protection factor based on the initial values ​​of the SNR, noise intensity, and protection factor.

[0106] In a more specific embodiment, obtaining the noise intensity of the recorded audio signal at a frequency of f specifically includes the following steps: obtaining the centerline of the recorded audio signal at a frequency of f; and calculating the noise intensity of the recorded audio signal at a frequency of f based on the signal values ​​located on the centerline, the average value of all signal values ​​located on the centerline, and the number of signal values ​​located on the centerline.

[0107] It should be noted that the midline is determined based on the maximum and minimum values ​​of the recorded audio signal at frequency f. The maximum value line is the line connecting the maximum values ​​of the recorded audio signal at frequency f, and the minimum value line is the line connecting the minimum values ​​of the recorded audio signal at frequency f.

[0108] In a more specific embodiment, the noise intensity of the recorded audio signal at frequency f is calculated using the following formula, based on the signal value located on the center line, the mean of all signal values ​​located on the center line, and the number of signal values ​​located on the center line:

[0109] (2);

[0110] Where Q is the noise intensity of the recorded audio signal at frequency f, and G is the number of signal values ​​located on the midline. Let i be the i-th signal value located on the midline. It is the average of all signal values ​​located on the midline.

[0111] In a more specific embodiment, based on the signal-to-noise ratio, noise intensity, and the initial value of the over-attenuation factor, the over-attenuation factor is adjusted using the following formula to obtain the adjusted over-attenuation factor:

[0112] (3);

[0113] (4);

[0114] in, This is the adjusted over-reduction factor. is the initial value of the over-subtraction factor, and Q is the noise intensity.

[0115] It should be noted that when the signal-to-noise ratio is higher than the preset threshold, the over-reduction factor is adjusted using formula (4), and when the signal-to-noise ratio is lower than the preset threshold, the over-reduction factor is adjusted using formula (3).

[0116] In a more specific embodiment, based on the initial values ​​of the signal-to-noise ratio, noise intensity, and protection factor, the protection factor is adjusted using the following formula to obtain the adjusted protection factor:

[0117] (5);

[0118] (6);

[0119] in, This is the adjusted protection coefficient. Q is the initial value of the protection factor, and Q is the noise intensity.

[0120] It should be noted that when the signal-to-noise ratio is higher than the preset threshold, the over-reduction factor is adjusted using formula (6), and when the signal-to-noise ratio is lower than the preset threshold, the over-reduction factor is adjusted using formula (5).

[0121] In this embodiment, under low signal-to-noise ratio conditions, the noise is strong and the signal is weak, therefore a stronger noise estimation is needed to better separate the noise. Increase It can improve the ability to suppress noise, thereby reducing the impact of noise; Increase the degree of retention of control signals. It can retain more signal information while suppressing noise. Therefore, at low signal-to-noise ratios, increasing... This can prevent the boundary between noise and signal from being misidentified.

[0122] In high signal-to-noise ratio (SNR) scenarios, noise is low and the signal is strong, therefore excessive noise suppression is unnecessary. Reduce This helps reduce excessive suppression of the signal and avoids introducing more distortion. Reduce This helps avoid over-smoothing signal details and preserves the original signal shape, especially at high signal-to-noise ratios where noise has less impact. This ensures better signal retention.

[0123] Step S104: Calculate the similarity between the features of the recorded audio segments to obtain a similarity matrix.

[0124] It should be noted that the features of different audio segments recorded by the same speaker are close in distance in the vector space.

[0125] In one alternative embodiment, cosine similarity between features of recorded audio segments is calculated to obtain a similarity matrix.

[0126] Step S105: Group the recorded audio segment features based on the similarity in the similarity matrix to obtain multiple recorded audio segment feature groups, and treat the recorded audio segment features in each recorded audio segment feature group as a cluster.

[0127] In one optional embodiment, based on the similarity in the similarity matrix, a clustering algorithm is used to group the recorded audio segment features to obtain multiple recorded audio segment feature groups, and the recorded audio segment features in each recorded audio segment feature group are regarded as a cluster.

[0128] Step S106: Calculate the similarity between clusters based on the similarity matrix.

[0129] In one alternative embodiment, the similarity between clusters is calculated using the maximum similarity based on the similarity matrix. Alternatively, other methods may be used to calculate the similarity between clusters, and this application does not limit the specific methods used.

[0130] Step S107: Perform clustering based on the similarity between clusters to obtain a clustering tree.

[0131] In one optional embodiment, a hierarchical clustering algorithm is used to cluster based on the similarity between clusters, resulting in a cluster tree. The cluster tree reflects the acoustic similarity relationships between different speakers.

[0132] Step S108: Assign speaker labels based on the clustering tree to obtain the target recorded audio signal segment.

[0133] It should be noted that the target audio signal segment includes the corresponding speaker tag.

[0134] In one optional embodiment, speaker labels are assigned based on a clustering tree to obtain the target recorded audio signal segment, specifically including the following steps:

[0135] Based on the voiceprint sample library, speaker labels are assigned to each cluster in the clustering tree.

[0136] That is, clusters are matched with real names, and the real names are used as speaker tags.

[0137] Step S109: Perform speech-to-text processing on the target audio signal segment to obtain the text content corresponding to the target audio signal segment.

[0138] In one optional embodiment, ASR (Automatic Speech Recognition) is used to process the target recorded audio signal segment into text to obtain the text content corresponding to the target recorded audio signal segment.

[0139] Step S110: Obtain the keyword list and historical task templates.

[0140] In one alternative embodiment, the keyword table is maintained through a maintainable interface, and the keyword table includes a serial number, a keyword, and the number of times the keyword is used.

[0141] In one optional embodiment, keywords in the keyword list can be added, modified, or deleted later, and keywords can also be imported in batches through templates.

[0142] In one optional embodiment, the historical task template data source is historically generated to-do tasks, and the target number is controlled according to usage rate and capacity. The template content includes time nodes, responsible persons, and task content. For example, responsible person A completes task B before the deadline.

[0143] Step S111: Dynamically construct prompt words based on the keyword list, historical task templates, and text content.

[0144] Prompt words refer to the text input when interacting with the task generation model, used to guide the model to generate specific outputs. The task generation model refers to a "large parameter" model trained using massive amounts of data and powerful computing capabilities. It possesses high versatility and generalization ability, and can be applied to fields such as natural language processing, image recognition, and speech recognition. It can be categorized into large language models, large visual models, multimodal large models, and basic large models, etc. For example, the task generation model in this embodiment can be a large language model.

[0145] In one alternative embodiment, the task and output format of the prompt words can be determined based on the meeting minutes format and target text content of the target meeting, thereby constructing the prompt words.

[0146] For example, the meeting minutes format of the target meeting is selected as the output format for the prompt words, and the prompt words are constructed based on the target text content and the meeting minutes format.

[0147] Step S112: Use the text generation type as the task type corresponding to the pre-loaded task generation model.

[0148] In one alternative embodiment, a task generation model is preloaded with the training objective of "generating structured tasks from unstructured text". The task type of the model can be conditional text generation or information extraction.

[0149] Step S113: Generate to-do tasks using the task generation model based on the prompt words.

[0150] In one alternative embodiment, prompts are input into a task generation model, inference is performed, and a structured list of tasks to be done is output.

[0151] In an alternative embodiment, the to-do task generation method provided in this application further includes the following steps:

[0152] Preprocessing operations are performed on the text content. The preprocessing operations include at least one of the following: cleaning operations, formatting operations, and sentence segmentation operations.

[0153] In one specific embodiment, the cleaning operation may refer to removing irrelevant content such as greetings and small talk from the text. The formatting standardization operation refers to standardizing the text content to include date / time formats. The sentence segmentation operation refers to breaking long paragraphs into short sentences with independent semantic meaning, using a hybrid sentence segmentation process, i.e., using punctuation marks + semantic meaning.

[0154] In another optional embodiment, the to-do task generation method provided in this application further includes the following steps:

[0155] Edit the text content based on the received editing operation.

[0156] In this embodiment, the accuracy of subsequent task generation is improved by modifying the text content.

[0157] In another optional embodiment, the to-do task generation method provided in this application embodiment further includes the following steps: converting the format of the to-do task so that the generated task can be "understood by the machine" and "used by the system".

[0158] like Figure 2 As shown in the figure, this application embodiment also provides a method for generating to-do tasks, which specifically includes the following steps:

[0159] Step 1: Obtain the keyword list set by calling the keyword API (Application Programming Interface).

[0160] Step 2: Obtain the collection of historical task templates based on the task template API.

[0161] Step 3: Enter the target text content.

[0162] Step 4: Dynamically construct prompt words based on the keyword list, historical task templates, and text content.

[0163] Through the above design, the system can respond in real time to changes in the keyword list and historical task templates, keep the generated results consistent with the latest enterprise standards, and achieve modular management of prompt word components;

[0164] Step 5: Load the task generation model and select the text generation type;

[0165] Step 6: Based on the dynamically constructed prompts, use the task generation model to extract the task to-do list. This process generates 1 to N different to-do tasks.

[0166] Step 7: Generate the corresponding format for the to-do list using the rules engine.

[0167] It should be noted that different rule engines can achieve different format conversions.

[0168] Corresponding to the method for generating to-do tasks provided in the embodiments of this application, the embodiments of this application also provide a device for generating to-do tasks, such as... Figure 3 As shown, the task generation device includes:

[0169] The first acquisition module 301 is used to acquire an audio recording file, which includes audio signals.

[0170] The segmentation module 302 is used to segment the recorded audio signal to obtain multiple recorded audio signal segments;

[0171] Extraction module 303 is used to extract features from the recorded audio signal segment to obtain the features of the recorded audio segment;

[0172] The first calculation module 304 is used to calculate the similarity between features of recorded audio segments to obtain a similarity matrix;

[0173] Grouping module 305 is used to group the recorded audio segment features based on the similarity in the similarity matrix to obtain multiple recorded audio segment feature groups, and to treat the recorded audio segment features in each recorded audio segment feature group as a cluster;

[0174] The second calculation module 306 is used to calculate the similarity between clusters based on the similarity matrix;

[0175] Clustering module 307 is used to perform clustering based on the similarity between clusters to obtain a clustering tree;

[0176] The allocation module 308 is used to allocate speaker tags based on a clustering tree to obtain a target audio recording signal segment, wherein the target audio recording signal segment includes the corresponding speaker tags.

[0177] The speech-to-text module 309 is used to process the target recorded audio signal segment into speech and obtain the text content corresponding to the target recorded audio signal segment.

[0178] The second acquisition module 310 is used to acquire the keyword list and historical task templates;

[0179] Module 311 is used to dynamically construct prompt words based on the keyword list, historical task templates, and text content;

[0180] Module 312 is used to determine the text generation type as the task type corresponding to the pre-loaded task generation model.

[0181] The generation module 313 is used to generate to-do tasks based on prompt words using a task generation model.

[0182] Corresponding to the method for generating to-do tasks provided in the embodiments of this application, the embodiments of this application also provide an electronic device for executing the method for generating to-do tasks, such as... Figure 4 As shown, the electronic device includes: a processor 401; and a memory 402 for storing a program for generating a task to be done. After the device is powered on and the program for generating the task to be done is run by the processor, the following steps are performed:

[0183] Acquire the audio recording file, which includes the audio signal.

[0184] The recorded audio signal is segmented to obtain multiple recorded audio signal segments;

[0185] Feature extraction is performed on the recorded audio signal segments to obtain the features of the recorded audio segments;

[0186] Calculate the similarity between features of the recorded audio segments to obtain a similarity matrix;

[0187] Based on the similarity in the similarity matrix, the audio segment features are grouped to obtain multiple audio segment feature groups, and the audio segment features in each audio segment feature group are treated as a cluster.

[0188] Calculate the similarity between clusters based on the similarity matrix;

[0189] Clustering is performed based on the similarity between clusters to obtain a clustering tree;

[0190] Speaker labels are assigned based on clustering trees to obtain target audio signal segments, which include the corresponding speaker labels.

[0191] The target audio signal segment is processed for speech-to-text conversion to obtain the corresponding text content of the target audio signal segment.

[0192] Obtain the keyword list and historical task templates;

[0193] Dynamically generate prompts based on the keyword list, historical task templates, and text content;

[0194] Use the text generation type as the task type corresponding to the pre-loaded task generation model;

[0195] Based on the prompts, generate tasks to be done using a task generation model.

[0196] Corresponding to the to-do task generation method provided in the embodiments of this application, the embodiments of this application also provide a computer-readable storage medium storing a program for the to-do task generation method, which is executed by a processor to perform the following steps:

[0197] Acquire the audio recording file, which includes the audio signal.

[0198] The recorded audio signal is segmented to obtain multiple recorded audio signal segments;

[0199] Feature extraction is performed on the recorded audio signal segments to obtain the features of the recorded audio segments;

[0200] Calculate the similarity between features of the recorded audio segments to obtain a similarity matrix;

[0201] Based on the similarity in the similarity matrix, the audio segment features are grouped to obtain multiple audio segment feature groups, and the audio segment features in each audio segment feature group are treated as a cluster.

[0202] Calculate the similarity between clusters based on the similarity matrix;

[0203] Clustering is performed based on the similarity between clusters to obtain a clustering tree;

[0204] Speaker labels are assigned based on clustering trees to obtain target audio signal segments, which include the corresponding speaker labels.

[0205] The target audio signal segment is processed for speech-to-text conversion to obtain the corresponding text content of the target audio signal segment.

[0206] Obtain the keyword list and historical task templates;

[0207] Dynamically generate prompts based on the keyword list, historical task templates, and text content;

[0208] Use the text generation type as the task type corresponding to the pre-loaded task generation model;

[0209] Based on the prompts, generate tasks to be done using a task generation model.

[0210] Corresponding to the to-do task generation method provided in the embodiments of this application, the embodiments of this application also provide a computer program containing instructions, which, when executed by a computer, cause the computer to perform the following steps:

[0211] Acquire the audio recording file, which includes the audio signal.

[0212] The recorded audio signal is segmented to obtain multiple recorded audio signal segments;

[0213] Feature extraction is performed on the recorded audio signal segments to obtain the features of the recorded audio segments;

[0214] Calculate the similarity between features of the recorded audio segments to obtain a similarity matrix;

[0215] Based on the similarity in the similarity matrix, the audio segment features are grouped to obtain multiple audio segment feature groups, and the audio segment features in each audio segment feature group are treated as a cluster.

[0216] Calculate the similarity between clusters based on the similarity matrix;

[0217] Clustering is performed based on the similarity between clusters to obtain a clustering tree;

[0218] Speaker labels are assigned based on clustering trees to obtain target audio signal segments, which include the corresponding speaker labels.

[0219] The target audio signal segment is processed for speech-to-text conversion to obtain the corresponding text content of the target audio signal segment.

[0220] Obtain the keyword list and historical task templates;

[0221] Dynamically generate prompts based on the keyword list, historical task templates, and text content;

[0222] Use the text generation type as the task type corresponding to the pre-loaded task generation model;

[0223] Based on the prompts, generate tasks to be done using a task generation model.

[0224] It should be noted that for a detailed description of the to-do task generation device, electronic device and computer-readable storage medium provided in the embodiments of this application, please refer to the relevant description of the to-do task generation method embodiments provided in the embodiments of this application, which will not be repeated here.

[0225] Although this application discloses preferred embodiments as described above, it is not intended to limit this application. Any person skilled in the art can make possible changes and modifications without departing from the spirit and scope of this application. Therefore, the scope of protection of this application should be determined by the scope defined in the claims of this application.

[0226] In a typical configuration, an electronic device includes one or more processors (Central Processing Units), input / output interfaces, network interfaces, and memory.

[0227] Memory may include non-persistent storage in computer-readable media, such as random access memory and / or non-volatile memory, like read-only memory or flash memory. Memory is an example of computer-readable media.

[0228] Computer-readable media, including both permanent and non-permanent, removable and non-removable media, can store information using any method or technology. Information can be computer-readable operations, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PCM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EPROM), flash memory or other memory technologies, compact disc read-only memory (CD-ROM), digital video disc (DMCD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include non-transitory computer-readable media, such as modulated data signals and carrier waves.

[0229] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, compact disc read-only memory, optical storage, etc.) containing computer-usable program code.

[0230] Although this application discloses preferred embodiments as described above, it is not intended to limit this application. Any person skilled in the art can make possible changes and modifications without departing from the spirit and scope of this application. Therefore, the scope of protection of this application should be determined by the scope defined in this application.

Claims

1. A to-do task generation method, characterized by, include: Acquire an audio recording file, wherein the audio recording file includes audio signals; The recorded audio signal is segmented to obtain multiple recorded audio signal segments; A short-time Fourier transform is performed on the recorded audio signal segment to obtain the recorded audio spectrum, which includes the recorded audio amplitude spectrum and the recorded audio phase spectrum. The noise amplitude spectrum is obtained by using MCRA to estimate the noise amplitude spectrum of the recorded audio. Based on the noise amplitude spectrum, the recorded audio amplitude spectrum is subtracted using the following formula to obtain the enhanced amplitude spectrum of the recorded audio: wherein is the recorded audio amplitude spectrum of the recorded audio signal at frequency f, is the recorded audio amplitude spectrum of the recorded audio signal at frequency f, is the noise amplitude spectrum of the noise at frequency f, is an over-reduction factor, is a protection factor; Based on the enhanced amplitude spectrum and the enhanced phase spectrum of the recorded audio, the enhanced spectrum of the recorded audio is obtained; Perform an inverse short-time Fourier transform on the enhanced audio spectrum of the recorded audio to obtain the enhanced audio signal of the recorded audio. The VAD is used to extract features from the audio enhancement signal to obtain the audio segment features. Calculate the cosine similarity between the features of the recorded audio segments to obtain a similarity matrix; The recorded audio segment features are grouped based on the similarity in the similarity matrix to obtain multiple recorded audio segment feature groups, and the recorded audio segment features in each recorded audio segment feature group are treated as a cluster; Based on the similarity matrix, the similarity between the clusters is calculated using the maximum similarity method; Based on the similarity between the clusters, a hierarchical clustering algorithm is used to perform clustering to obtain a clustering tree; Speaker tags are assigned based on the clustering tree to obtain target audio recording segments, wherein the target audio recording segments include corresponding speaker tags; The target audio recording signal segment is processed for speech-to-text conversion to obtain the text content corresponding to the target audio recording signal segment. Obtain the keyword list and historical task templates; Prompt words are dynamically constructed based on the keyword list, the historical task template, and the text content; Use the text generation type as the task type corresponding to the pre-loaded task generation model; Based on the prompt words, generate to-do tasks using the task generation model; Before performing a spectral subtraction operation on the recorded audio amplitude spectrum based on the noise amplitude spectrum to obtain the enhanced amplitude spectrum of the recorded audio, the method further includes: Calculate the signal-to-noise ratio (SNR) of the recorded audio signal at frequency f; obtain the midline of the recorded audio signal at frequency f, the midline being determined based on the maximum and minimum values ​​of the recorded audio signal at frequency f, the maximum value line being the line connecting the maximum values ​​of the recorded audio signal at frequency f, and the minimum value line being the line connecting the minimum values ​​of the recorded audio signal at frequency f; calculate the noise intensity of the recorded audio signal at frequency f based on the signal values ​​located on the midline, the average of all signal values ​​located on the midline, and the number of signal values ​​located on the midline; adjust the over-attenuation factor based on the SNR, the noise intensity, and the initial value of the over-attenuation factor; adjust the protection factor based on the SNR, the noise intensity, and the initial value of the protection factor. The calculation of the noise intensity of the recorded audio signal at frequency f based on the signal value located on the midline, the average of all signal values ​​located on the midline, and the number of signal values ​​located on the midline includes: Based on the signal value located on the center line, the average of all signal values ​​located on the center line, and the number of signal values ​​located on the center line, the noise intensity of the recorded audio signal at frequency f is calculated using the following formula: ； where Q is the noise intensity of the recorded audio signal at a frequency of f, G is the number of signal values located on the midline, G is the ith signal value located on the midline, G is the average of all signal values located on the midline; The adjustment of the over-attenuation factor based on the signal-to-noise ratio, the noise intensity, and the initial value of the over-attenuation factor includes: If the signal-to-noise ratio is higher than a preset threshold, the over-attenuation factor is adjusted using the following formula based on the signal-to-noise ratio, the noise intensity, and the initial value of the over-attenuation factor: ； wherein, is the adjusted over-reduction factor, is the initial value of the over-reduction factor, Q is the noise intensity; If the signal-to-noise ratio is lower than a preset threshold, the over-attenuation factor is adjusted using the following formula based on the signal-to-noise ratio, the noise intensity, and the initial value of the over-attenuation factor: ； wherein, is the adjusted over-reduction factor, is the initial value of the over-reduction factor, Q is the noise intensity; The adjustment of the protection coefficient based on the signal-to-noise ratio, the noise intensity, and the initial value of the protection coefficient includes: If the signal-to-noise ratio is higher than a preset threshold, the protection coefficient is adjusted using the following formula based on the signal-to-noise ratio, the noise intensity, and the initial value of the protection coefficient: ； wherein is the adjusted protection coefficient, is the initial value of the protection coefficient, Q is the noise intensity; If the signal-to-noise ratio is lower than a preset threshold, the protection coefficient is adjusted using the following formula based on the signal-to-noise ratio, the noise intensity, and the initial value of the protection coefficient: ； wherein is the adjusted protection coefficient, is the initial value of the protection coefficient, Q is the noise intensity; Also includes: Edit the text content based on the received editing operation; Also includes: Convert the format of the tasks to be done.

2. The to-do list generation method according to claim 1, characterized by, Before assigning speaker labels based on the clustering tree to obtain the target recorded audio signal segment, the method further includes: Obtain voiceprint samples from attendees; Based on the aforementioned voiceprint samples, a voiceprint sample library is constructed.

3. The to-do list generation method according to claim 2, characterized by, The step of assigning speaker labels based on the clustering tree to obtain the target recorded audio signal segment includes: Based on the voiceprint sample library, speaker labels are assigned to each cluster in the clustering tree.

4. A to-do task generating apparatus characterized by comprising: include: The first acquisition module is used to acquire an audio recording file, wherein the audio recording file includes an audio signal. The segmentation module is used to segment the recorded audio signal to obtain multiple recorded audio signal segments; The extraction module is used to extract features from the recorded audio signal segment using VAD to obtain the recorded audio segment features; The first calculation module is used to calculate the cosine similarity between the features of the recorded audio segments to obtain a similarity matrix; The grouping module is used to group the recorded audio segment features based on the similarity in the similarity matrix to obtain multiple recorded audio segment feature groups, and to treat the recorded audio segment features in each recorded audio segment feature group as a cluster; The second calculation module is used to calculate the similarity between the clusters based on the similarity matrix and using the maximum similarity. The clustering module is used to perform clustering based on the similarity between the clusters using a hierarchical clustering algorithm to obtain a clustering tree; The allocation module is used to allocate speaker tags based on the clustering tree to obtain a target recorded audio signal segment, wherein the target recorded audio signal segment includes the corresponding speaker tag; The speech-to-text module is used to process the target recorded audio signal segment into speech-to-text to obtain the text content corresponding to the target recorded audio signal segment. The second acquisition module is used to acquire the keyword list and historical task templates; The construction module is used to dynamically construct prompt words based on the keyword list, the historical task template, and the text content; The determination module is used to identify the text generation type as the task type corresponding to the pre-loaded task generation model. The generation module is used to generate to-do tasks based on the prompt words and the task generation model. The device also includes a module that performs the following operations: The recorded audio signal segment is subjected to a short-time Fourier transform to obtain the recorded audio spectrum, which includes the recorded audio amplitude spectrum and the recorded audio phase spectrum; noise amplitude spectrum is obtained by using MCRA to estimate the noise of the recorded audio amplitude spectrum. Based on the noise amplitude spectrum, the recorded audio amplitude spectrum is subtracted using the following formula to obtain the enhanced amplitude spectrum of the recorded audio: wherein is the recorded audio amplitude spectrum of the recorded audio signal at frequency f, is the recorded audio amplitude spectrum of the recorded audio signal at frequency f, is the noise amplitude spectrum of the noise at frequency f, is the over-reduction factor, is the protection factor; Based on the recorded audio enhancement amplitude spectrum and the recorded audio phase spectrum, the recorded audio enhancement spectrum is obtained; the recorded audio enhancement signal is obtained by performing an inverse short-time Fourier transform on the recorded audio enhancement spectrum. The device also includes a module that performs the following operations: Calculate the signal-to-noise ratio (SNR) of the recorded audio signal at frequency f; obtain the midline of the recorded audio signal at frequency f, the midline being determined based on the maximum and minimum values ​​of the recorded audio signal at frequency f, the maximum value line being the line connecting the maximum values ​​of the recorded audio signal at frequency f, and the minimum value line being the line connecting the minimum values ​​of the recorded audio signal at frequency f; calculate the noise intensity of the recorded audio signal at frequency f based on the signal values ​​located on the midline, the average of all signal values ​​located on the midline, and the number of signal values ​​located on the midline; adjust the over-attenuation factor based on the SNR, the noise intensity, and the initial value of the over-attenuation factor; adjust the protection factor based on the SNR, the noise intensity, and the initial value of the protection factor. Based on the signal value located on the center line, the average of all signal values ​​located on the center line, and the number of signal values ​​located on the center line, the noise intensity of the recorded audio signal at frequency f is calculated using the following formula: ； Where Q is the noise intensity of the recorded audio signal at frequency f, and G is the number of signal values ​​located on the midline. Let i be the i-th signal value located on the midline. This is the average of all signal values ​​located on the midline; If the signal-to-noise ratio is higher than a preset threshold, the over-attenuation factor is adjusted using the following formula based on the signal-to-noise ratio, the noise intensity, and the initial value of the over-attenuation factor: ； in, This is the adjusted over-reduction factor. The initial value of the over-subtraction factor is given, and Q is the noise intensity. If the signal-to-noise ratio is lower than a preset threshold, the over-attenuation factor is adjusted using the following formula based on the signal-to-noise ratio, the noise intensity, and the initial value of the over-attenuation factor: ； in, This is the adjusted over-reduction factor. The initial value of the over-subtraction factor is given, and Q is the noise intensity. If the signal-to-noise ratio is higher than a preset threshold, the protection coefficient is adjusted using the following formula based on the signal-to-noise ratio, the noise intensity, and the initial value of the protection coefficient: ； in, This is the adjusted protection coefficient. Here, Q is the initial value of the protection factor, and Q is the noise intensity. If the signal-to-noise ratio is lower than a preset threshold, the protection coefficient is adjusted using the following formula based on the signal-to-noise ratio, the noise intensity, and the initial value of the protection coefficient: ； in, This is the adjusted protection coefficient. Here, Q is the initial value of the protection factor, and Q is the noise intensity. The device also includes modules that perform the following operations: Edit the text content based on the received editing operation; The device also includes modules that perform the following operations: Convert the format of the tasks to be done.

Images