Commit History
-
v2: fine-tuned model fixing sentence-initial word collapse (.?!)
- Replaced onnx/json with fine-tuned checkpoint (epoch 18, val_mel 0.3567) - Dataset 27h -> 28h (added 1,372 targeted sentence-initial phrases) - Regression scan: collapsed words 4/8 -> 1/41 - Inference defaults unchanged (noise 0.6 / noise_w 0.4) - Updated sample audiob182e7d
-
Replace sample with serena_high_sample_1 audio
committa0d5b66d -
it_IT-serena-high piper voice (config A inference) + sample
committa32368e9