A speech recognition method and device for the power dispatching field, an electronic device, and a storage medium
By combining multi-stage iterative training with a speech recognition model generated from power dispatch terminology and scene noise, the problems of low accuracy and poor anti-interference ability in the field of power dispatching have been solved, achieving higher recognition accuracy and robustness.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- POWER DISPATCHING CONTROL CENT OF GUANGDONG POWER GRID CO LTD
- Filing Date
- 2026-04-30
- Publication Date
- 2026-07-17
AI Technical Summary
Existing speech recognition technologies face problems such as low accuracy in recognizing technical terms and poor anti-interference capabilities in the field of power dispatching. This is mainly due to the severe lack of coverage of technical terms and complex noise distribution in the training data, which makes the model prone to omissions or misrecognitions in real complex scenarios.
By acquiring historical real dispatch speech and scene-enhanced synthesized speech as training samples, and combining them with power dispatch terminology, multi-stage iterative training is carried out, including parameter optimization of encoder, predictor and decoder, to generate target speech recognition model and enhance its adaptability to professional terminology and complex noise.
It significantly improves the recognition accuracy and anti-interference robustness of the target speech recognition model in real business scenarios, solves the problems of omission and misrecognition of professional terms, and improves the recognition accuracy of scheduling instructions.
Smart Images
Figure CN122417015A_ABST