Quint 1.2 — real-time speech → semantic paragraphs

This release reorients Quint around real-time, CPU-only speech → semantic paragraphs.

New

• Streaming semantic chunker (quint.chunking.streaming): single-pass, no O(n²) similarity matrix, no lookahead — runs on a live stream with bounded memory. The split rule is self-calibrating (sim < mean − z·std), so it needs no per-corpus threshold tuning. Includes stream_chunks_live (a generator that yields each paragraph the instant it closes), stream_chunks_adaptive, and stream_chunks.
• Live mic demo on the Hugging Face Space: a fully-local CPU pipeline — Moonshine speech-to-text → punctuation restoration → streaming chunker → paragraphs. No GPU, no torch.

Docs

• README, PyPI metadata, and the Space now lead with real-time speech → semantic paragraphs.

pip install quintessentia