RagLeapFirst pre-alpha release of ragleap-voice: the model-management layer only.
There are no speech engines in this release. It cannot transcribe or synthesize yet.
What it does:
- A licence-checked model manifest (MIT, Apache-2.0, BSD, ISC and CC-BY with attribution are allowed).
- Lock files with a size and SHA-256 per file at a pinned revision.
- Hash-verified downloads from Hugging Face only (HTTPS, host allowlist, tokens never sent to other hosts), plus offline verification.
- A command line:
ragleap-voice models list | show | check | verify | pull | lock. - Benchmark tools: WER and CER metrics, CSV test manifests, a speech-to-text runner.
- Zero runtime dependencies. Python 3.10 or newer. MIT.
Status of models: all six manifest entries are still marked unverified, and none is a default. Indic Parler-TTS and its tokenizer are locked but have not had a human review.
Install:
pip install ragleap-voice
uvx ragleap-voice models list
Measurements (read speech, small samples, one GPU, with their limits) are in benchmarks/voice/README.md. Nothing in this release is a claim about any commercial service.
Planned for 0.1.0: transcribe() and synthesize() with engines as optional extras. The design draft is open for comments in issue #762.