alpa/ gains the conversion from tiny-voca, the Rasa import, feature extraction, timing labels and the trainer. scripts/ has four helpers. TRAINING.md lists every command in the order it was run. Weights, browser files and samples are unchanged.
Card: link to the technical report at alpa.blog/report
New weights alpa-v0.2.pt and browser file onnx/alpa-v0.2.onnx: the female voice is now trained, and the male voice learned from twice the recordings (9.8 hours in all, 3 h 39 min of CPU training in total, no GPU). Samples in both voices; card updated. The v0.1 and v0.1.1 files are kept.
Same weights file as v0.1. The way the output is turned into sound is corrected while speaking (alpa/model.py, Polish); new browser file onnx/alpa-v0.1.1.onnx beside the v0.1 one; samples re-made; card updated. `--plain` gives the v0.1 sound.
Model card: banner, step diagram and three charts added