Upload 5 files

.gitignore

@@ -0,0 +1,7 @@
+__pycache__/
+*.pyc
+*.pyo
+.env
+generated_audio/
+*.mp3
+.DS_Store

Dockerfile

@@ -0,0 +1,33 @@
+# ─── Base image ───────────────────────────────────────────────
+FROM python:3.10-slim
+
+# ─── System deps needed by OpenCV & MediaPipe ─────────────────
+RUN apt-get update && apt-get install -y --no-install-recommends \
+        libgl1 \
+        libglib2.0-0 \
+        libsm6 \
+        libxext6 \
+        libxrender-dev \
+    && rm -rf /var/lib/apt/lists/*
+
+# ─── Working directory ────────────────────────────────────────
+WORKDIR /app
+
+# ─── Python deps ──────────────────────────────────────────────
+COPY requirements.txt .
+RUN pip install --no-cache-dir -r requirements.txt
+
+# ─── App code ─────────────────────────────────────────────────
+COPY app.py .
+
+# ─── Models directory (upload your .keras files here) ─────────
+#     Structure expected:
+#       models/asl_landmark_cnn_lstm_model.keras    ← English
+#       models/arsl_landmark_cnn_lstm_model.keras   ← Arabic
+COPY models/ ./models/
+
+# ─── HuggingFace Spaces uses port 7860 ────────────────────────
+EXPOSE 7860
+
+# ─── Start ────────────────────────────────────────────────────
+CMD ["uvicorn", "app:app", "--host", "0.0.0.0", "--port", "7860"]

README.md

@@ -1,10 +1,66 @@
 ---
-title: Sign Language
-emoji: 🏃
-colorFrom: yellow
-colorTo: red
+title: Sign Language Translator API
+emoji: 🤟
+colorFrom: blue
+colorTo: purple
 sdk: docker
 pinned: false
 ---
 
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+# 🤟 Sign Language Translator API
+
+Real-time **Arabic & English** Sign Language recognition using **CNN-BiLSTM + MediaPipe**.
+
+## Endpoints
+
+| Method | URL | Description |
+|--------|-----|-------------|
+| GET | `/` | API info |
+| GET | `/health` | Model status |
+| POST | `/predict/english` | Predict English ASL letter |
+| POST | `/predict/arabic` | Predict Arabic ARSL letter |
+
+## How to use
+
+Send a hand-gesture image (JPG/PNG) as `multipart/form-data`:
+
+```bash
+curl -X POST "https://<your-space>.hf.space/predict/english" \
+     -F "file=@hand.jpg"
+```
+
+### Response example
+
+```json
+{
+  "language": "english",
+  "predicted_label": "A",
+  "predicted_char": "A",
+  "confidence": 0.9823,
+  "above_threshold": true,
+  "message": "OK"
+}
+```
+
+```json
+{
+  "language": "arabic",
+  "predicted_label": "Alef",
+  "predicted_char": "ا",
+  "confidence": 0.9412,
+  "above_threshold": true,
+  "message": "OK"
+}
+```
+
+## Model details
+
+- Input: single hand-gesture image
+- MediaPipe extracts 21 landmarks → 63 features
+- Sequence replicated to 23 timesteps → shape `(1, 23, 63)`
+- CNN + BiLSTM → Softmax
+- Confidence threshold: **0.8**
+- English classes: 28 (A-Z + del + space)
+- Arabic classes: 33
+
+## Built by Ahmed Sobhy

app.py

@@ -0,0 +1,240 @@
+import os
+import cv2
+import uuid
+import base64
+import numpy as np
+import tensorflow as tf
+import mediapipe as mp
+
+from fastapi import FastAPI, File, UploadFile, HTTPException
+from fastapi.middleware.cors import CORSMiddleware
+from pydantic import BaseModel
+from typing import Optional
+from contextlib import asynccontextmanager
+
+
+#  PATHS
+BASE_DIR = os.path.dirname(os.path.abspath(__file__))
+
+ENGLISH_MODEL_PATH = os.path.join(BASE_DIR, "models", "asl_landmark_cnn_lstm_model.keras")
+ARABIC_MODEL_PATH  = os.path.join(BASE_DIR, "models", "arsl_landmark_cnn_lstm_model.keras")
+
+
+#  CLASSES
+ENGLISH_CLASSES = list("ABCDEFGHIJKLMNOPQRSTUVWXYZ") + ["del", "space"]
+
+ARABIC_CLASSES = [
+    "Ain", "Al", "Alef", "Beh", "Dad", "Dal", "Feh", "Ghain", "Hah", "Heh",
+    "Jeem", "Kaf", "Khah", "Laa", "Lam", "Meem", "Noon", "Qaf", "Reh", "Sad",
+    "Seen", "Sheen", "Tah", "Teh", "Teh_Marbuta", "Thal", "Theh", "Waw", "Yeh",
+    "Zah", "Zain", "del", "space"
+]
+
+ARABIC_MAP = {
+    "Alef": "ا", "Beh": "ب", "Teh": "ت", "Theh": "ث", "Jeem": "ج",
+    "Hah": "ح", "Khah": "خ", "Dal": "د", "Thal": "ذ", "Reh": "ر",
+    "Zain": "ز", "Seen": "س", "Sheen": "ش", "Sad": "ص", "Dad": "ض",
+    "Tah": "ط", "Zah": "ظ", "Ain": "ع", "Ghain": "غ", "Feh": "ف",
+    "Qaf": "ق", "Kaf": "ك", "Lam": "ل", "Meem": "م", "Noon": "ن",
+    "Heh": "ه", "Waw": "و", "Yeh": "ي", "Laa": "لا", "Al": "ال",
+    "Teh_Marbuta": "ة"
+}
+
+
+#  CONFIG
+TIMESTEPS            = 23
+CONFIDENCE_THRESHOLD = 0.8
+
+
+#  GLOBAL MODELS & MEDIAPIPE
+english_model = None
+arabic_model  = None
+hands_detector = None
+
+
+@asynccontextmanager
+async def lifespan(app: FastAPI):
+    global english_model, arabic_model, hands_detector
+
+    print("Loading models...")
+
+    if os.path.exists(ENGLISH_MODEL_PATH):
+        english_model = tf.keras.models.load_model(ENGLISH_MODEL_PATH)
+        print(" English model loaded")
+    else:
+        print(f"  English model not found at {ENGLISH_MODEL_PATH}")
+
+    if os.path.exists(ARABIC_MODEL_PATH):
+        arabic_model = tf.keras.models.load_model(ARABIC_MODEL_PATH)
+        print(" Arabic model loaded")
+    else:
+        print(f"  Arabic model not found at {ARABIC_MODEL_PATH}")
+
+    mp_hands = mp.solutions.hands
+    hands_detector = mp_hands.Hands(
+        static_image_mode=True,
+        max_num_hands=1,
+        min_detection_confidence=0.5
+    )
+    print(" MediaPipe ready")
+
+    yield
+
+    # cleanup
+    if hands_detector:
+        hands_detector.close()
+    print("Shutting down...")
+
+
+
+#  APP
+app = FastAPI(
+    title="Sign Language Translator API",
+    description="Real-time Arabic & English Sign Language recognition using CNN-BiLSTM + MediaPipe",
+    version="1.0.0",
+    lifespan=lifespan
+)
+
+app.add_middleware(
+    CORSMiddleware,
+    allow_origins=["*"],
+    allow_methods=["*"],
+    allow_headers=["*"],
+)
+
+
+
+#  SCHEMAS
+class PredictResponse(BaseModel):
+    language: str
+    predicted_label: str          # raw class name  e.g. "Alef"
+    predicted_char: str           # display char    e.g. "ا"
+    confidence: float
+    above_threshold: bool
+    message: str
+
+
+
+#  HELPERS
+def decode_image(data: bytes) -> np.ndarray:
+    """Decode uploaded image bytes → BGR numpy array."""
+    arr = np.frombuffer(data, dtype=np.uint8)
+    img = cv2.imdecode(arr, cv2.IMREAD_COLOR)
+    if img is None:
+        raise HTTPException(status_code=400, detail="Cannot decode image. Send a valid JPG/PNG.")
+    return img
+
+
+def extract_landmarks(img_bgr: np.ndarray) -> Optional[np.ndarray]:
+    """Run MediaPipe on a BGR image → 63-d landmark vector or None."""
+    img_rgb = cv2.cvtColor(img_bgr, cv2.COLOR_BGR2RGB)
+    results = hands_detector.process(img_rgb)
+
+    if not results.multi_hand_landmarks:
+        return None
+
+    hand = results.multi_hand_landmarks[0]
+    coords = []
+    for lm in hand.landmark:
+        coords.extend([lm.x, lm.y, lm.z])
+
+    lm_arr = np.array(coords, dtype=np.float32)
+
+    # Mirror right hand so model always sees left-hand orientation
+    if results.multi_handedness:
+        label = results.multi_handedness[0].classification[0].label
+        if label == "Right":
+            lm_arr[0::3] = 1.0 - lm_arr[0::3]
+
+    return lm_arr
+
+
+def build_sequence(lm: np.ndarray) -> np.ndarray:
+    """Repeat single frame 23× → (1, 23, 63) normalised tensor."""
+    x_seq = np.repeat(lm[np.newaxis, :], TIMESTEPS, axis=0)   # (23, 63)
+    x_seq = x_seq[np.newaxis, :, :]                            # (1, 23, 63)
+    max_val = np.max(np.abs(x_seq))
+    if max_val != 0:
+        x_seq = x_seq / max_val
+    return x_seq
+
+
+def run_inference(model, x_seq: np.ndarray, classes: list) -> tuple[str, float]:
+    """Return (predicted_class_name, confidence)."""
+    probs     = model.predict(x_seq, verbose=0)
+    idx       = int(np.argmax(probs))
+    conf      = float(probs[0][idx])
+    label     = classes[idx] if idx < len(classes) else "unknown"
+    return label, conf
+
+
+
+#  ROUTES
+@app.get("/")
+def root():
+    return {
+        "name": "Sign Language Translator API",
+        "endpoints": {
+            "predict_english": "POST /predict/english",
+            "predict_arabic":  "POST /predict/arabic",
+            "health":          "GET  /health"
+        }
+    }
+
+
+@app.get("/health")
+def health():
+    return {
+        "status": "ok",
+        "english_model_loaded": english_model is not None,
+        "arabic_model_loaded":  arabic_model  is not None,
+    }
+
+
+@app.post("/predict/english", response_model=PredictResponse)
+async def predict_english(file: UploadFile = File(...)):
+    if english_model is None:
+        raise HTTPException(status_code=503, detail="English model not loaded.")
+
+    img   = decode_image(await file.read())
+    lm    = extract_landmarks(img)
+
+    if lm is None:
+        raise HTTPException(status_code=422, detail="No hand detected in the image.")
+
+    x_seq        = build_sequence(lm)
+    label, conf  = run_inference(english_model, x_seq, ENGLISH_CLASSES)
+
+    return PredictResponse(
+        language        = "english",
+        predicted_label = label,
+        predicted_char  = label,           # same for English
+        confidence      = round(conf, 4),
+        above_threshold = conf >= CONFIDENCE_THRESHOLD,
+        message         = "OK" if conf >= CONFIDENCE_THRESHOLD else f"Low confidence ({conf:.2f})"
+    )
+
+
+@app.post("/predict/arabic", response_model=PredictResponse)
+async def predict_arabic(file: UploadFile = File(...)):
+    if arabic_model is None:
+        raise HTTPException(status_code=503, detail="Arabic model not loaded.")
+
+    img   = decode_image(await file.read())
+    lm    = extract_landmarks(img)
+
+    if lm is None:
+        raise HTTPException(status_code=422, detail="No hand detected in the image.")
+
+    x_seq        = build_sequence(lm)
+    label, conf  = run_inference(arabic_model, x_seq, ARABIC_CLASSES)
+    char         = ARABIC_MAP.get(label, label)
+
+    return PredictResponse(
+        language        = "arabic",
+        predicted_label = label,
+        predicted_char  = char,
+        confidence      = round(conf, 4),
+        above_threshold = conf >= CONFIDENCE_THRESHOLD,
+        message         = "OK" if conf >= CONFIDENCE_THRESHOLD else f"Low confidence ({conf:.2f})"
+    )

requirements.txt

@@ -0,0 +1,21 @@
+# ── Web framework ──────────────────────────────
+fastapi==0.115.12
+uvicorn[standard]==0.34.0
+python-multipart==0.0.20
+pydantic==2.11.3
+
+# ── ML / CV ────────────────────────────────────
+# tensorflow-cpu أخف بكتير على السيرفر (مفيش GPU على HuggingFace Free)
+tensorflow-cpu==2.17.0
+keras==3.13.2
+mediapipe==0.10.14
+opencv-python-headless==4.11.0.86
+numpy==1.26.4
+
+# ── Arabic support ─────────────────────────────
+arabic-reshaper==3.0.0
+python-bidi==0.6.7
+pillow==12.1.1
+
+# ── Utils ──────────────────────────────────────
+requests==2.32.5