System and method for intelligently generating conference record based on WavLM model

Abstract

The invention discloses a system and method for intelligently generating conference records based on a WavLM model, and the system comprises a user interface module which is used for displaying a user interface; the voice processing and recognition module is used for preprocessing the collected voice signals and converting the voice signals into texts; the natural language processing module is used for processing the text and generating a conference record; and the database module is used for storing conference records and audio files. According to the method, the voice in the conference can be automatically identified, the text is automatically generated in a spokesman role dialogue form, and finally a high-quality conference record is formed.

Description

Technical Field

[0001] The present invention belongs to the technical field of speech recognition, and specifically is a system and method for intelligently generating conference records based on a WavLM model. Background Art

[0002] In modern businesses and organizations, meetings have become an important part of daily work. However, organizing meeting minutes is a time-consuming and error-prone task.

[0003] Common speech recognition functions can only convert speech into text, but cannot identify and distinguish the roles of speakers, requiring manual marking. A large number of meeting records need to be recorded and played back repeatedly, and then manually marked to indicate that a certain paragraph is the speech of speaker A, and a certain paragraph is the speech of speaker B... In the long run, the workload of meeting recorders is extremely high and their work efficiency is low. Summary of the invention

[0004] In view of the above problems, the present invention is proposed to provide a system and method for intelligently generating meeting records based on the WavLM model to overcome the above problems or at least partially solve the above problems.

[0005] To achieve the above object, the present invention adopts the following technical solutions:

[0006] A system for intelligently generating meeting records based on a WavLM model, the system comprising:

[0007] A user interface module, used for displaying a user interface;

[0008] The speech processing and recognition module is used to pre-process the collected speech signals and convert the speech signals into text;

[0009] A natural language processing module to process text and generate meeting records;

[0010] Database module, used to store meeting records and audio files.

[0011] Optionally, the system also includes a log management module for recording operation logs and user behavior logs.

[0012] Optionally, the user interface module includes:

[0013] Conference control unit, used to start, pause and end conference recording;

[0014] The meeting recording unit is used to display the transcription of the meeting in real time and supports manual editing and modification.

[0015] Optionally, the speech processing and recognition module includes:

[0016] An audio preprocessing unit, used to reduce noise and increase gain on the collected voice signal;

[0017] The WavLM model unit uses the WavLM model for speech recognition, converts speech signals into text, and divides the text into paragraphs based on different people's voices.

[0018] Optional natural language processing modules include:

[0019] Semantic analysis unit, used to extract keywords and identify intent from text;

[0020] The summary generation unit generates a conference summary based on the semantic analysis results.

[0021] The present invention also provides a method for intelligently generating conference records based on the WavLM model. The method is based on any of the aforementioned systems for intelligently generating conference records based on the WavLM model, and is characterized in that the method comprises:

[0022] Collecting voice signals;

[0023] Preprocessing the collected speech signals;

[0024] Perform speech recognition on the collected speech signals through the WavLM model;

[0025] Generate and store meeting minutes.

[0026] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0027] 1. The present invention utilizes the WavLM model to achieve real-time recording and organization of conference content through speech recognition, natural language processing and intelligent generation technology.

[0028] 2. The present invention can automatically recognize the speech in the meeting and automatically generate text in the form of speaker role dialogue, ultimately forming high-quality meeting records. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 A schematic diagram of a system flow for intelligently generating meeting records based on the WavLM model provided in an embodiment of the present application;

[0030] Figure 2 A schematic diagram of a system architecture for intelligently generating meeting minutes based on the WavLM model provided in an embodiment of the present application. DETAILED DESCRIPTION

[0031] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0032] Embodiment 1:

[0033] See also Figure 1 and Figure 2 This embodiment provides a system for intelligently generating meeting records based on the WavLM model, the system comprising:

[0034] The user interface module is used to display the user interface.

[0035] The user interface module includes:

[0036] The conference control unit is used to start, pause and end conference recording. Users can start, pause and end conference recording through this unit.

[0037] The meeting recording unit is used to display the transcription of the meeting in real time and supports manual editing and modification.

[0038] The speech processing and recognition module is used to pre-process the collected speech signals and convert the speech signals into text.

[0039] The speech processing and recognition module includes:

[0040] The audio preprocessing unit is used to perform preprocessing such as noise reduction and gain on the collected speech signal to improve the accuracy of speech recognition.

[0041] The WavLM model unit uses the WavLM model for speech recognition, converts speech signals into text, and divides the text into paragraphs based on the speaker roles according to different human voices.

[0042] Natural language processing module, used to process text and generate meeting records.

[0043] The natural language processing module includes:

[0044] The semantic analysis unit is used to extract keywords and identify intent from text to ensure the accuracy and logic of the recorded content.

[0045] The summary generation unit generates concise and clear conference summaries based on the semantic analysis results.

[0046] The database module is used to store meeting records and audio files and generate meeting summaries.

[0047] The log management module is used to record operation logs and user behavior logs to facilitate system maintenance and problem tracking.

[0048] WavLM (Waveform Language Model) is an advanced speech model proposed by Microsoft Research, which is designed to handle the complexity and diversity of speech signals. WavLM is based on the Transformer architecture and can effectively perform speech recognition, speech enhancement and other speech processing tasks through pre-training on large-scale speech data.

[0049] The core technologies of the WavLM model include the following aspects:

[0050] 1. Transformer architecture: WavLM adopts a Transformer-based encoder structure, which takes advantage of the self-attention mechanism in modeling long-distance dependencies and can better capture the timing information in speech signals.

[0051] 2. Pre-training and fine-tuning: WavLM learns rich speech representations by pre-training on large-scale unlabeled speech data, and then fine-tunes on specific tasks to improve the performance of specific tasks.

[0052] 3. Noise stability: WavLM takes the impact of noise into consideration during its design. By training on speech data in different noise environments, the model’s stability to noise is enhanced, and it can maintain a high recognition accuracy in complex real-world environments.

[0053] 4. Multi-task learning: WavLM adopts a multi-task learning method during the pre-training process to learn shared representations of different speech tasks (such as speech recognition, speech enhancement, speaker recognition, etc.), which improves the generalization ability and performance of the model.

[0054] This embodiment uses the WavLM model to achieve real-time recording and organization of meeting content through speech recognition, natural language processing and intelligent generation technology. This embodiment can automatically recognize the speech in the meeting and automatically generate text in the form of speaker role dialogue, ultimately forming high-quality meeting records.

[0055] Embodiment 2:

[0056] Embodiment 2 discloses a method for intelligently generating conference records based on the WavLM model. The method is based on the above-mentioned system for intelligently generating conference records based on the WavLM model. The method comprises the following steps:

[0057] S1. Collect voice signals.

[0058] S2. Preprocess the collected speech signals.

[0059] S3. Perform speech recognition on the collected speech signals through the WavLM model.

[0060] S4. Generate meeting minutes and store them.

[0061] The above contents are further detailed descriptions of the present invention in combination with specific preferred embodiments, and it cannot be determined that the specific implementation of the present invention is limited to these descriptions. For ordinary technicians in the technical field to which the present invention belongs, several simple deductions or substitutions can be made without departing from the concept of the present invention, which should be regarded as falling within the protection scope of the present invention.

Claims

1. A system for intelligently generating meeting records based on the WavLM model, characterized in that: The system comprises: A user interface module, used for displaying a user interface; The speech processing and recognition module is used to pre-process the collected speech signals and convert the speech signals into text; A natural language processing module to process text and generate meeting records; Database module, used to store meeting records and audio files.

2. A system for intelligently generating meeting records based on the WavLM model as claimed in claim 1, characterized in that: The system also includes a log management module for recording operation logs and user behavior logs.

3. A system for intelligently generating meeting records based on the WavLM model as claimed in claim 1, characterized in that: The user interface module includes: Conference control unit, used to start, pause and end conference recording; The meeting recording unit is used to display the transcription of the meeting in real time and supports manual editing and modification.

4. A system for intelligently generating meeting records based on the WavLM model as claimed in claim 1, characterized in that: The speech processing and recognition module includes: An audio preprocessing unit, used to reduce noise and increase gain on the collected voice signal; The WavLM model unit uses the WavLM model for speech recognition, converts speech signals into text, and divides the text into paragraphs based on different people's voices.

5. A system for intelligently generating meeting records based on the WavLM model as claimed in claim 1, characterized in that: The natural language processing module includes: Semantic analysis unit, used to extract keywords and identify intent from text; The summary generation unit generates a conference summary based on the semantic analysis results.

6. A method for intelligently generating meeting records based on the WavLM model, the method is based on the system for intelligently generating meeting records based on the WavLM model as described in any one of claims 1 to 5, characterized in that: The method comprises: Collecting voice signals; Preprocessing the collected speech signals; Perform speech recognition on the collected speech signals through the WavLM model; Generate and store meeting minutes.

Images