A method and system for speech recognition and student formative assessment based on dynamic vocabulary injection.

Abstract

This invention discloses a method and system for speech recognition and student process evaluation based on dynamic vocabulary injection, relating to the fields of artificial intelligence and smart education technology. This invention aims to solve the problem of low recognition rate of homophones in existing speech recognition technologies, leading to inaccurate attribution of process evaluation results and high manual recording costs. The method includes: obtaining a student list for the target teaching scenario; pre-injecting the list as a dynamic vocabulary into an ASR (Automatic Speech Recognition) engine for decoding weight bias; acquiring real-time teaching speech streams, using the ASR engine with injected dynamic vocabulary for accurate recognition, and outputting text containing specific student names; using a Natural Language Processing (NLP) model to extract entities from the text, separating name entities and performance evaluation entities; finally, inputting the performance evaluation entities into an AI scoring model to automatically generate a quantitative score and include it in the student's process evaluation file. This invention completely solves the problem of recognizing homophones and proper nouns, realizing an accompanying, seamless, automated evaluation scoring closed loop, greatly reducing the recording burden on teachers.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence and smart education technology, specifically to a speech recognition and student process evaluation method and system based on dynamic vocabulary injection. Background Technology

[0002] Currently, formative assessment in educational settings is a key focus of teaching reform. However, relying on teachers to manually record or use tablets to score students significantly increases their workload and disrupts the teaching rhythm. Using speech recognition technology to achieve "accompanied assessment" (e.g., the teacher verbally saying, "Wang Zixuan, you answered the question actively, add one point") is becoming a trend. However, due to the large number of homophones and rare characters in Chinese names, general speech recognition models struggle to accurately identify student names in specific scenarios (e.g., misidentifying "Wang Zixuan" as "Wang Zixuan"). If the name entity is misidentified, subsequent evaluation records will become completely invalid or result in misattribution, causing a break in the closed loop of the entire AI scoring system. Summary of the Invention

[0003] The purpose of this invention is to overcome the shortcomings of the prior art and provide a complete process evaluation method and system that solves the problem of homophone name recognition at the front end of the speech recognition end and seamlessly connects to AI automatic scoring at the back end. Figure 1. Business process diagram of the student process assessment system; Figure 2. Architecture diagram of the student process assessment system. Figure 3 A schematic diagram illustrating the dynamic vocabulary insertion ASR decoding principle.

[0004] The core of this invention lies in: obtaining a real-time list of scenarios by connecting to the academic affairs system, and using contextual biasing technology to temporarily change the mapping weights from ASR acoustic features to text features to solve the problem of homophone recognition of proper nouns; furthermore, using a large AI model to understand non-standardized oral evaluations and transforming them into structured process evaluation data.

[0005] The specific implementation methods are described in detail with reference to Figures 1 to 3. Example 1: Classroom Accompanied Evaluation Scenario Before a Chinese language class in a primary school, the system automatically retrieves a list of 45 students (including student names: Zhang Wei, Li Zixuan, etc.) from Class 2, Grade 3 via a scheduling interface. List Injection Stage: The dynamic injection module loads these 45 names into the hot word table of the ASR engine, setting the bias weight coefficient to +15.0. Real-time Recognition Stage: The teacher, wearing a microphone, says in class, "Zhang Wei, you recited the ancient poem very fluently today. Li Zixuan, please don't whisper among your classmates." Speech Decoding Process: When encountering the pronunciation feature of "zhang wei," the general language model defaults to outputting the highly probable "Zhang Wei." However, because "Zhang Wei" is given additional weight in the dynamic word table, the network ultimately selects the path with the highest global score through the Viterbi decoding algorithm, accurately outputting "Zhang Wei"; similarly, it accurately outputs "Li Zixuan." Semantic Parsing Stage: The NLP model processes the above text. Entity 1: [Object: Zhang Wei, Event: Fluent recitation of classical poems]; Entity 2: [Object: Li Zixuan, Event: Whispering]. AI evaluation generation stage: The AI ​​scoring model matches "fluent recitation of classical poems" to the evaluation dimension "Academic Performance - Positive," generating a score of +2 points; it matches "whispering" to "Classroom Discipline - Disciplinary Violation," generating a score of -1 point. Finally, the data is seamlessly stored in the personal digital profiles of Zhang Wei and Li Zixuan.

Claims

1. A speech recognition and student formative assessment method based on dynamic vocabulary injection, characterized in that, Includes the following steps: S1: Obtain the list of target personnel in the target scenario; S2: Convert the target personnel list data into a dynamic vocabulary and inject it as a hot word bias parameter into the acoustic model or language model of the speech recognition (ASR) engine; S3: Collect real-time audio streams during the teaching process, decode the audio streams using the speech recognition engine with the injected dynamic vocabulary, and output structured text containing the accurate names of the target personnel. S4: Utilize a Natural Language Processing (NLP) model to extract entities and identify intents from the structured text, separating the corresponding personnel name entity and performance evaluation entity; S5: Input the performance evaluation entity into a pre-trained AI scoring model, combine it with preset educational evaluation dimension rules to automatically quantify and score it, and record the scoring results in the process evaluation database of the corresponding personnel.

2. The method according to claim 1, characterized in that, The injection method described in step S2 specifically includes: constructing a weighted finite state converter (WFST) dynamic decoding network, assigning higher state transition probability weights to the target personnel list data, so that when there are homophones or similar pronunciations in the decoding path, the names in the list are output first.

3. The method according to claim 1, characterized in that, The natural language processing (NLP) model described in step S4 includes a named entity recognition (NER) module for extracting names, actions, evaluation polarity (praise or criticism), and degree words.

4. The method according to claim 1, characterized in that, The AI ​​scoring model described in step S5 is based on the Large Language Model (LLM). Through the preset prompt word project, it maps the extracted performance evaluation entities to preset structured evaluation dimensions, including but not limited to: classroom focus, participation, logical thinking, and discipline, and generates corresponding increase or decrease scores.

5. A speech recognition and student process evaluation system based on dynamic vocabulary injection, characterized in that, include: List Acquisition Module: Used to acquire target personnel list data in the target scenario; Dynamic Injection Module: Used to convert the list data into a dynamic vocabulary and inject it into the speech recognition engine; Speech Recognition Module: Used to collect real-time audio streams and output text using the engine after injecting the dynamic vocabulary; Semantic parsing module: used to perform intent recognition and entity extraction on the output text, extracting personnel names and performance evaluation content; AI evaluation generation module: used to automatically score the extracted performance evaluation content, and generate and store process evaluation records.

Images