Upload README.md with huggingface_hub

README.md

@@ -8,19 +8,20 @@ app_file: app.py
 pinned: true
 license: cc-by-4.0
 models:
-- Tonykip/parakeet-tdt-0.6b-kalenjin
+- Tonykip/whisper-kalenjin-v3-turbo
 short_description: Transcribe spoken Kalenjin (Kipsigis & Nandi)
 ---
 
 # Kalenjin Speech Recognition 🎙️
 
-Try the first **open** automatic speech recognition model for **Kalenjin** (Kipsigis & Nandi):
-record a phrase or upload a clip and get a transcription.
+Try the recommended open ASR model for **Kalenjin** (Kipsigis & Nandi): record a phrase or upload a
+clip and get a transcription.
 
-Powered by [`Tonykip/parakeet-tdt-0.6b-kalenjin`](https://huggingface.co/Tonykip/parakeet-tdt-0.6b-kalenjin)
-— an honest fine-tune of NVIDIA's `parakeet-tdt-0.6b-v3` (FastConformer + TDT), **not trained from
-scratch**. It preserves the velar-nasal `ng'` letter and is dialect-balanced (Kipsigis ≈ Nandi).
+Powered by [`Tonykip/whisper-kalenjin-v3-turbo`](https://huggingface.co/Tonykip/whisper-kalenjin-v3-turbo)
+— a fine-tune of OpenAI's Whisper-large-v3-turbo. Long clips are transcribed with **chunked decoding +
+anti-repetition controls**, so the model stays stable on long-form speech (no looping).
 
-**Note on speed:** on free CPU hardware transcription takes a few seconds per clip. For
-near-instant results, switch this Space to a GPU or ZeroGPU runtime in **Settings → Hardware**
-(the app already uses `@spaces.GPU` when a GPU is present).
+**Speed:** on free CPU hardware transcription takes a few seconds per clip. For near-instant results,
+switch this Space to a GPU or ZeroGPU runtime in **Settings → Hardware** (the app auto-detects a GPU).
+
+_A separate experimental Parakeet fine-tune exists but is not recommended for use._