🌾 Sahel-Voice — Minimal Baseline

Zero-shot Whisper → CohereLabs/aya-expanse-32b → MMS-TTS, with a curated Bambara/Pular phrasebook short-circuit in front of the LLM. No adapters, no memory, no polish. This is the field-test baseline — see docs/baseline_rebuild.md.
Input language

Language you're speaking/typing. Drives Whisper hint (fr/en only) and bam_normalize (bam only).
Output language

Language the LLM should reply in. Also picks the TTS voice.

What's intentionally missing: LoRA adapters, memory/vocabulary persistence, speaker ID, Waxal/F5 TTS, IoT sensor integration, phrase-matcher shortcuts. All of those live in app.py — this is the stripped-down baseline used to measure what Whisper zero-shot does on real Bambara/Fula recordings and to collect a real-user eval set.

The Text tab skips Whisper — it's for fast iteration on the LLM + TTS path, not for field-test measurement.

How the two boxes differ: the top pair is a phrasebook lookup (no LLM, instant, gold-curated translation). If your input isn't in the curated list you'll see (no curated translation) — click Generate reply to get a dialect-anchored LLM response in the bottom pair.