🎙️ Edge STT Capability Probe

Phase 0 spike — run all three tests, then share the report back on WhatsApp.

Environment detecting…

Tier 1 · Web Speech API not tested

Transcript appears here…
Pick the language you will actually speak first — Web Speech does not auto-detect; if you speak Telugu while this is set to English you get garbage. Then Start, allow the mic, and speak a test sentence (e.g. “I want a loan of 10 lakh rupees.” / “నాకు 10 లక్షల రూపాయల లోన్ కావాలి.”).

Tier 2 · Keyboard dictation not tested

This tests the phone keyboard's built-in voice typing (Gboard / iOS dictation). We can't trigger it from code — tap the mic key on the keyboard itself, speak, then press one of the buttons above.

Tier 3 · Audio capture (server STT path) not tested

Speak any language — do NOT pick a language, just talk. Record a sentence like “Naaku 10 lakh loan kaavaali” or “मुझे 15 लाख का लोन चाहिए”, then tap Transcribe. The clip goes to a server Whisper model that auto-detects the language, and an LLM extracts the intent/amount. (For this test the clip is sent to Workers AI; a production deployment self-hosts the model in-country.)

Verdict & report pending

Run the tests above, then the verdict appears here.

Debug log