Rgains/ibibio-tts

# ibibio-tts-lab Reproducible data preparation and research tooling for an Ibibio (`ibb`) Common Voice multi-speaker TTS baseline. ## Quick start ```bash UV_CACHE_DIR=.uv-cache uv sync --python 3.11 UV_CACHE_DIR=.uv-cache uv run python scripts/validate_environment.py UV_CACHE_DIR=.uv-cache uv run pytest UV_CACHE_DIR=.uv-cache uv run ruff check . UV_CACHE_DIR=.uv-cache uv run mypy src scripts ``` The validator checks the supported Python version, FFmpeg tools, free disk space, and (when present) the Common Voice archive checksum. Run it before downloading or importing data. A non-zero exit means the environment is not ready for the next pipeline stage. ## Configuration Copy `.env.example` to `.env` and fill in real values; `.env` is git-ignored and must never be committed. `.env.example` must only ever hold placeholders. - `YOUR_API_KEY` — Mozilla Data Collective API key used to fetch the Common Voice dataset archive. - `DATASET_ID` — Common Voice Scripted Speech 26.0 (Ibibio) dataset ID, `cmqiohial00zgnq07ybhhu7ni`. Leave this at the pinned value; see `PROTOCOL.md` before changing it. With `.env` configured, run the pipeline in order (`import_common_voice.py` downloads the archive itself via the Mozilla Data Collective API, verifies its checksum against `reports/common_voice_ibb_archive.sha256`, and extracts it into `data/raw/common_voice_ibb/` — no manual placement needed): ```bash UV_CACHE_DIR=.uv-cache uv run python scripts/import_common_voice.py UV_CACHE_DIR=.uv-cache uv run python scripts/audit_audio.py UV_CACHE_DIR=.uv-cache uv run python scripts/analyse_speakers.py UV_CACHE_DIR=.uv-cache uv run python scripts/build_manifests.py ``` This produces, in order: an import manifest with every rejection tracked (`data/interim/.../imported.parquet`), decoded 24 kHz mono PCM16 WAV clips plus an audio audit (`reports/audio_audit.*`), an anonymous per-speaker audit (`reports/speaker_audit.*`), and leakage-safe 90/5/5 train/validation/ test manifests with a leak …