声音克隆方法、系统、电子设备及介质
By segmenting and grading reference speech to determine its stability, a timbre anchor segment library is constructed. High-risk boundary windows are identified and locally corrected, solving the problem of timbre drift accumulation in existing technologies and achieving efficient, low-latency long text audio cloning.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- CHONGQING MALYA MEDIA CO LTD
- Filing Date
- 2026-04-07
- Publication Date
- 2026-07-17
AI Technical Summary
Given the overlapping constraints of limited reference speech duration and uneven quality, multiple pause boundaries in the target text, and the need for near real-time output, existing voice cloning technologies struggle to identify high-risk timbre drift windows and cannot effectively perform local corrections, leading to the accumulation and spread of timbre deviations in long text synthesis.
By segmenting the reference speech into segments, calculating the stability score, and constructing a stable timbre anchor segment library, local corrections are made for high-risk boundary windows. Local backfilling and resynthesis are performed using boundary risk scores and continuous residuals to prevent bias propagation. The threshold is adjusted through an adaptive update mechanism.
It effectively identifies and corrects high-risk tone drift windows, avoids high latency in whole-sentence regeneration, ensures output tone stability, adapts to different recording quality conditions, and is suitable for mobile and online applications.
Smart Images
Figure CN121983069B_ABST