Çekicin AI · M4 — voice-guidance pipeline (STT → LLM → TTS)
Configuration of the Çekicin voice assistant used by the Telegram bot and the /v1/voice/demo API.
- STT: Whisper large-v3-turbo (Workers AI) — word accuracy 97.4% clean, 94.9% in loud car noise on the Turkish driver-request set (https://huggingface.co/datasets/sosa123454321/cekicin-m4-voice-eval).
- LLM: @cf/openai/gpt-oss-120b, @cf/meta/llama-4-scout-17b-16e-instruct with the system prompt in
pipeline.json(safety first: 112, tow number, no fake dispatch). - TTS: not yet — answers are text. Candidates listed in
pipeline.json; the Space'sapp_gradio.pyalready wires MMS-TTS Turkish for the PRO/GPU version.
Live: https://t.me/Cekicin_bot (send a voice message) · Space: https://sosa123454321-cekicin-m4-voice.static.hf.space
Part of the Çekicin AI demo (6 roadmap modules). Not in the Çekicin mobile app yet. AI Lab: https://sosa123454321-cekicin-site.static.hf.space/ai-lab/status.html