一种目标说话人语音获取方法和系统
By designing a target speaker speech acquisition model in multi-speaker scenarios and using hybrid and reference corpora for feature separation and comparison, the problem of low voiceprint recognition accuracy in multi-speaker scenarios is solved, achieving more efficient target speaker recognition and application expansion.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- XIAMEN KUAISHANGTONG TECH CORP LTD
- Filing Date
- 2022-10-26
- Publication Date
- 2026-07-17
AI Technical Summary
In multi-speaker scenarios, existing voiceprint recognition technology struggles to accurately identify the target speaker, resulting in low recognition accuracy and limiting the application scenarios of voiceprint recognition systems.
By acquiring mixed corpora, reference corpora, and single-speaker corpora, and utilizing the target speaker's speech acquisition model, including a mixed speech interface module, a speech encoding module, a speaker extraction module, a reference speech interface module, a speaker encoding module, and a speaker comparison module, one-to-one feature scoring and speech decoding are performed to improve the model's training effect and recognition accuracy.
It effectively improves the accuracy of voiceprint recognition in multi-speaker scenarios, expands the application scenarios of voiceprint recognition, and enhances the robustness and recognition efficiency of the model.
Smart Images
Figure CN115881093B_ABST