# pixi environments
.pixi/
__pycache__/
config_sv.smoke.yml

# fetched/trained artifacts and data (live on the 3090 box / downloaded)
voices/
output/
data/
recipe/
models/

# g2p/ holds the vendored neural-G2P CODE (tracked); its big runtime artifacts
# auto-download from HF (Joakim/swedish-kokoro under g2p/) — keep those out of git.
g2p/g2p_model.pt
g2p/g2p_model_snapshot.pt
g2p/lexicon.tsv
g2p/__pycache__/

# temporary audio
*.wav

# run logs
*.log

# large model weights — host on HuggingFace / Git LFS, NOT plain git (>100 MB).
# deploy/README.md, config.json, sv_female.pt (512 K) ARE tracked.
deploy/*.pth
deploy/*.onnx

# --- session data/artifacts (2026-07: keep code, exclude everything heavy) ---
data_*/
raw_*/
hf_bundle/
hf_pack/
.venv/
*.pt
!deploy/sv_female.pt
*.parquet
*.tar
ravdess*/
sweep_*/
listen_*/
joint*/
mv3_*/
surgery_*/
v3_e*/
mood_*/

# editor/agent config
.claude/
# large regenerable corpora (derived from public TTS-Swedish; regen via split_corpus.py)
data-gen/corpus_sv*.txt
data-gen/corpus_v3_*.txt
training/data/
training/outputs/
training/.venv/
data-gen/dataset*/
data-gen/output/

# generated per-run training configs (train_kokoro.py regenerates from template)
training/configs/train_*.yml

# python build artifacts
dist/
build/
*.egg-info/
