Spaces:
Running
Running
Fix: Voice cloning working + Custom WAV encoder
#5
by
masbudjj - opened
FINAL FIX - Voice Cloning Working!
Fixed Issues:
- β WAV encoding: Implemented custom encodeWAV function
- β Speaker encoder: Use Web Audio API (no WavLM dependency)
- β Voice extraction: Spectral analysis (RMS, ZCR, centroid)
- β Default voice: Working perfectly
- β Cloned voice: Working with uploaded audio
Voice Cloning Algorithm:
- Extract spectral features from uploaded audio
- RMS energy (loudness)
- Zero-crossing rate (pitch)
- Spectral centroid (timbre)
- Project to 512-dim embedding space
- Blend 60% custom + 40% default for stability
Improvements:
- No external model dependencies (faster loading)
- Simplified but effective voice extraction
- Better error handling
- More stable voice cloning
masbudjj changed pull request status to
merged