A power grid dispatching telephone audio corpus processing and model self-adaptive iterative training method and system

By combining a hybrid noise reduction model of Wiener filtering and spectral subtraction with homophone correction from a power grid dispatch-specific dictionary, an adaptive iterative training mechanism is constructed. This solves the problems of noise interference, low corpus quality, and low iteration efficiency in power grid dispatch telephone corpus processing and model training, achieving efficient and secure speech recognition and model iterative optimization.

CN122201258APending Publication Date: 2026-06-12STATE GRID FUJIAN ELECTRIC POWER CO LTD
View PDF 1 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
STATE GRID FUJIAN ELECTRIC POWER CO LTD
Filing Date
2026-03-05
Publication Date
2026-06-12

AI Technical Summary

⚠Technical Problem

Existing technologies for processing telephone corpora and training models in power grid dispatching suffer from problems such as poor adaptability to noise interference, low corpus quality, insufficient model adaptability, low iteration efficiency, and lack of collaborative optimization between corpus and model, which cannot meet the high reliability and real-time requirements of power grid dispatching.

⚗Method used

A hybrid noise reduction model combining Wiener filtering and spectral subtraction is adopted. Based on a professional terminology dictionary for power grid dispatching, homophones are automatically corrected and semi-automated. An adaptive iterative training mechanism is constructed, and the model parameters are optimized through reinforcement learning, forming a two-way closed-loop collaborative optimization of corpus processing and model training.

🎯Benefits of technology

It achieves efficient noise reduction and professional terminology recognition for power grid dispatching telephone audio, improves corpus processing efficiency and model recognition accuracy, adapts to the real-time requirements of power grid dispatching scenarios, reduces security risks, and meets the high security requirements of power grid dispatching.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122201258A_ABST
    Figure CN122201258A_ABST
Patent Text Reader

Abstract

The application provides a power grid dispatching telephone audio corpus processing and model adaptive iterative training method and system, comprising: performance evaluation is carried out on an initial speech recognition model or an online running speech recognition model, samples of power grid dispatching professional term recognition errors are recorded and an error sample set is formed; based on a preset adaptive power grid dispatching scene iteration trigger threshold, it is judged that when the trigger condition is met, the model iterative training is automatically started; in combination with the error sample set and the newly added labeled corpus, an incremental training mode is adopted, a reinforcement learning mechanism aiming at reducing the overall word error rate of the model and preferentially improving the power grid dispatching professional term recognition accuracy is introduced to iteratively optimize the model until the model performance reaches the preset standard; the model recognition error sample is fed back to the same sound word automatic correction matching and semi-automated labeling corpus processing link, the targeted optimization of the corpus processing logic is driven, and the bidirectional closed loop collaborative optimization of the corpus processing and the model training is formed.
Need to check novelty before this filing date? Find Prior Art