Shorten HF short_description

README.md

@@ -6,87 +6,5 @@ colorTo: blue
 sdk: docker
 pinned: false
 license: apache-2.0
-short_description: FastAPI service for Whisper ASR via HTTP POST /audio endpoint
----
-
-# Whisper ASR Space
-
-FastAPI service for **Hololens / Unity / clients**: `POST /audio` with the same format as local `ASR/src/web_realtime.py`.....
-
-Default **Hub model** in code: [`Thedeezat/ASR-Hearing-Impaired`](https://huggingface.co/Thedeezat/ASR-Hearing-Impaired). You can override with Space variable **`MODEL_ID`**.
-
-## Your checklist (what you do)
-
-### 1. Model on the Hub (already done if you use that repo)
-
-If you use another checkpoint: create a **model repo** and upload everything from `ASR/models/whisper_finetuned/` (`config.json`, tokenizer files, weights, etc.; use **Git LFS** for large files).
-
-### 2. Create this Space on Hugging Face
-
-1. Create a **new Space** → choose **Docker** (not Gradio).
-2. Name it (e.g. `YOUR_USERNAME/whisper-asr-space`).
-3. Clone the empty Space locally, then **copy into it** the files from **this folder** in your grad-project repo:
-   - `Dockerfile`
-   - `main.py`
-   - `requirements.txt`
-   - replace the auto-generated `README.md` **metadata** at the top with the YAML block in this file (title, `sdk: docker`, etc.)
-
-   Or: push this `ASR/ASR-Hearing-Impaired-API/` folder contents to the Space repo’s `main` branch.
-
-   ```bash
-   cd ASR/hf_space
-   git clone https://huggingface.co/spaces/YOUR_USERNAME/YOUR_SPACE
-   cp Dockerfile main.py requirements.txt YOUR_SPACE/
-   # Copy README.md from here (full file) into YOUR_SPACE/README.md
-   cd YOUR_SPACE
-   git add -A && git commit -m "Add ASR API" && git push
-   ```
-
-### 3. Private or gated model — **HF_TOKEN** (required)
-
-401 / “Repository Not Found” usually means the Space is not authenticated for **`Thedeezat/ASR-Hearing-Impaired`**.
-
-1. Open the Space → **Settings** → **Variables and secrets**.
-2. Under **Repository secrets** (not “Variables”), add:
-   - **Name:** `HF_TOKEN` (exactly; the app reads this and `HUGGING_FACE_HUB_TOKEN`).
-   - **Value:** a [fine-grained](https://huggingface.co/docs/hub/security-tokens#fine-grained-access-tokens) or **classic** token with **Read** access to model repo `Thedeezat/ASR-Hearing-Impaired`.
-3. **Redeploy** the Space (new commit or “Factory reboot”) after saving the secret.
-4. Do **not** paste the token in code, chat, or README. If it was exposed, **revoke** it on your [token settings](https://huggingface.co/settings/tokens) and create a new one.
-
-The API passes your token only to `from_pretrained` (it does **not** call `huggingface_hub.login()`), so startup avoids extra `whoami` calls that can trigger **HTTP 429** on Hugging Face.
-
-Optional: **`MODEL_ID`** variable only if you use another Hub repo than the default.
-
-### 3b. Speed / latency (optional — matches local `web_realtime`)
-
-These **Space variables** (or Docker `-e`) tune inference to match the fast local defaults:
-
-| Variable | Default | Meaning |
-|----------|---------|--------|
-| `ASR_WHISPER_NUM_BEAMS` | `1` | Greedy decode (fastest). Use `2` for slightly better quality, slower. |
-| `ASR_WHISPER_MAX_NEW_TOKENS` | `256` | Shorter decoder cap → faster responses. |
-| `ASR_MIN_RMS` | `0.009` | Drop very quiet chunks (noise) before the model. |
-
-Generation also uses Whisper **no_speech** / **compression_ratio** thresholds (same idea as `ASR/src/whisper_asr.py`) to reduce silence hallucinations, plus trailing-silence trim and light repetition cleanup on the server.
-
-### 4. Public URL for your HoloLens app
-
-After the Space builds (**Building** → **Running**), your API base is:
-
-`https://YOUR_USERNAME-YOUR_SPACE_NAME.hf.space`
-
-Examples:
-
-- `GET https://...hf.space/` — service info  
-- `GET https://...hf.space/health` — health  
-- `POST https://...hf.space/audio` — body = float32 PCM (same as `web_realtime`), header `X-Sample-Rate: 16000`  
-
-**Note:** Free Spaces **sleep after inactivity**; the first request after sleep can take a long time (cold start). Upgrade to paid hardware if you need always-hot GPU/CPU.
-
-## Local smoke test
-
-```bash
-docker build -t whisper-asr .
-docker run -e MODEL_ID=Thedeezat/ASR-Hearing-Impaired -p 7860:7860 whisper-asr
-curl http://127.0.0.1:7860/health
-```
+short_description: Whisper ASR HTTP API
+---